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.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _LUSTRE_DLM_H__
38 #define _LUSTRE_DLM_H__
40 /** \defgroup ldlm ldlm
45 #if defined(__linux__)
46 #include <linux/lustre_dlm.h>
47 #elif defined(__APPLE__)
48 #include <darwin/lustre_dlm.h>
49 #elif defined(__WINNT__)
50 #include <winnt/lustre_dlm.h>
52 #error Unsupported operating system.
55 #include <lustre_lib.h>
56 #include <lustre_net.h>
57 #include <lustre_import.h>
58 #include <lustre_handles.h>
59 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
65 #define OBD_LDLM_DEVICENAME "ldlm"
67 #ifdef HAVE_BGL_SUPPORT
68 /* 1.5 times the maximum 128 tasks available in VN mode */
69 #define LDLM_DEFAULT_LRU_SIZE 196
71 #define LDLM_DEFAULT_LRU_SIZE (100 * cfs_num_online_cpus())
73 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
74 #define LDLM_CTIME_AGE_LIMIT (10)
79 ELDLM_LOCK_CHANGED = 300,
80 ELDLM_LOCK_ABORTED = 301,
81 ELDLM_LOCK_REPLACED = 302,
82 ELDLM_NO_LOCK_DATA = 303,
84 ELDLM_NAMESPACE_EXISTS = 400,
85 ELDLM_BAD_NAMESPACE = 401
89 LDLM_NAMESPACE_SERVER = 1 << 0,
90 LDLM_NAMESPACE_CLIENT = 1 << 1
93 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
95 /* If the server returns one of these flags, then the lock was put on that list.
96 * If the client sends one of these flags (during recovery ONLY!), it wants the
97 * lock added to the specified list, no questions asked. -p */
98 #define LDLM_FL_BLOCK_GRANTED 0x000002
99 #define LDLM_FL_BLOCK_CONV 0x000004
100 #define LDLM_FL_BLOCK_WAIT 0x000008
102 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
103 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
104 * queued for sending. */
105 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
106 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
108 /* Lock is being replayed. This could probably be implied by the fact that one
109 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
110 #define LDLM_FL_REPLAY 0x000100
112 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
113 #define LDLM_FL_LOCAL_ONLY 0x000400
115 /* don't run the cancel callback under ldlm_cli_cancel_unused */
116 #define LDLM_FL_FAILED 0x000800
118 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
119 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
120 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
121 /* was LDLM_FL_WARN until 2.0.0 0x008000 */
122 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
124 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
127 /* file & record locking */
128 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
129 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
131 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
132 * the LVB filling happens _after_ the lock has been granted, so another thread
133 * can match`t before the LVB has been updated. As a dirty hack, we set
134 * LDLM_FL_LVB_READY only after we've done the LVB poop.
135 * this is only needed on lov/osc now, where lvb is actually used and callers
136 * must set it in input flags.
138 * The proper fix is to do the granting inside of the completion AST, which can
139 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
140 * pretty high-risk, though, and would need a lot more testing. */
142 #define LDLM_FL_LVB_READY 0x100000
144 /* A lock contributes to the kms calculation until it has finished the part
145 * of it's cancelation that performs write back on its dirty pages. It
146 * can remain on the granted list during this whole time. Threads racing
147 * to update the kms after performing their writeback need to know to
148 * exclude each others locks from the calculation as they walk the granted
150 #define LDLM_FL_KMS_IGNORE 0x200000
152 /* Immediatelly cancel such locks when they block some other locks. Send
153 * cancel notification to original lock holder, but expect no reply. This is
154 * for clients (like liblustre) that cannot be expected to reliably response
155 * to blocking ast. */
156 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
158 /* Flags flags inherited from parent lock when doing intents. */
159 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
161 /* completion ast to be executed */
162 #define LDLM_FL_CP_REQD 0x1000000
164 /* cleanup_resource has already handled the lock */
165 #define LDLM_FL_CLEANED 0x2000000
167 /* optimization hint: LDLM can run blocking callback from current context
168 * w/o involving separate thread. in order to decrease cs rate */
169 #define LDLM_FL_ATOMIC_CB 0x4000000
171 /* was LDLM_FL_ASYNC until 2.0.0 0x8000000 */
173 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
174 * such that server send blocking ast for conflict locks to this client for
175 * the 1st operation, whereas the 2nd operation has canceled this lock and
176 * is waiting for rpc_lock which is taken by the 1st operation.
177 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
178 * ELC code to cancel it.
179 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
180 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
181 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
182 * to send a separate CANCEL rpc. */
183 #define LDLM_FL_BL_AST 0x10000000
184 #define LDLM_FL_BL_DONE 0x20000000
186 /* measure lock contention and return -EUSERS if locking contention is high */
187 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
189 /* These are flags that are mapped into the flags and ASTs of blocking locks */
190 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
192 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
193 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
196 * --------------------------------------------------------------------------
197 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
198 * 0x80000000 will not be sent over the wire.
199 * --------------------------------------------------------------------------
202 /* Used for marking lock as an target for -EINTR while cp_ast sleep
203 * emulation + race with upcoming bl_ast. */
204 #define LDLM_FL_FAIL_LOC 0x100000000ULL
206 /* Used while processing the unused list to know that we have already
207 * handled this lock and decided to skip it */
208 #define LDLM_FL_SKIPPED 0x200000000ULL
210 /* The blocking callback is overloaded to perform two functions. These flags
211 * indicate which operation should be performed. */
212 #define LDLM_CB_BLOCKING 1
213 #define LDLM_CB_CANCELING 2
215 /* compatibility matrix */
216 #define LCK_COMPAT_EX LCK_NL
217 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
218 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
219 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
220 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
221 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
222 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
223 #define LCK_COMPAT_COS (LCK_COS)
225 extern ldlm_mode_t lck_compat_array[];
227 static inline void lockmode_verify(ldlm_mode_t mode)
229 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
232 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
234 return (lck_compat_array[exist_mode] & new_mode);
239 * cluster name spaces
243 #define DLM_OST_NAMESPACE 1
244 #define DLM_MDS_NAMESPACE 2
247 - do we just separate this by security domains and use a prefix for
248 multiple namespaces in the same domain?
258 * waiting_locks_spinlock
273 struct ldlm_resource;
274 struct ldlm_namespace;
276 struct ldlm_pool_ops {
277 int (*po_recalc)(struct ldlm_pool *pl);
278 int (*po_shrink)(struct ldlm_pool *pl, int nr,
279 unsigned int gfp_mask);
280 int (*po_setup)(struct ldlm_pool *pl, int limit);
284 * One second for pools thread check interval. Each pool has own period.
286 #define LDLM_POOLS_THREAD_PERIOD (1)
289 * ~6% margin for modest pools. See ldlm_pool.c for details.
291 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
294 * Default recalc period for server side pools in sec.
296 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
299 * Default recalc period for client side pools in sec.
301 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
305 * Pool proc directory.
307 cfs_proc_dir_entry_t *pl_proc_dir;
309 * Pool name, should be long enough to contain compound proc entry name.
313 * Lock for protecting slv/clv updates.
315 cfs_spinlock_t pl_lock;
317 * Number of allowed locks in in pool, both, client and server side.
319 cfs_atomic_t pl_limit;
321 * Number of granted locks in
323 cfs_atomic_t pl_granted;
327 cfs_atomic_t pl_grant_rate;
331 cfs_atomic_t pl_cancel_rate;
333 * Server lock volume. Protected by pl_lock.
335 __u64 pl_server_lock_volume;
337 * Current biggest client lock volume. Protected by pl_lock.
339 __u64 pl_client_lock_volume;
341 * Lock volume factor. SLV on client is calculated as following:
342 * server_slv * lock_volume_factor.
344 cfs_atomic_t pl_lock_volume_factor;
346 * Time when last slv from server was obtained.
348 time_t pl_recalc_time;
350 * Recalc period for pool.
352 time_t pl_recalc_period;
354 * Recalc and shrink ops.
356 struct ldlm_pool_ops *pl_ops;
358 * Number of planned locks for next period.
364 struct lprocfs_stats *pl_stats;
367 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
368 void *req_cookie, ldlm_mode_t mode, int flags,
371 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
373 struct ldlm_valblock_ops {
374 int (*lvbo_init)(struct ldlm_resource *res);
375 int (*lvbo_update)(struct ldlm_resource *res,
376 struct ptlrpc_request *r,
378 int (*lvbo_free)(struct ldlm_resource *res);
382 LDLM_NAMESPACE_GREEDY = 1 << 0,
383 LDLM_NAMESPACE_MODEST = 1 << 1
387 * Default values for the "max_nolock_size", "contention_time" and
388 * "contended_locks" namespace tunables.
390 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
391 #define NS_DEFAULT_CONTENTION_SECONDS 2
392 #define NS_DEFAULT_CONTENDED_LOCKS 32
394 struct ldlm_ns_bucket {
396 struct ldlm_namespace *nsb_namespace;
397 /** estimated lock callback time */
398 struct adaptive_timeout nsb_at_estimate;
402 /** ldlm namespace lock stats */
409 LDLM_NS_TYPE_UNKNOWN = 0,
424 struct ldlm_namespace {
426 * Backward link to obd, required for ldlm pool to store new SLV.
428 struct obd_device *ns_obd;
431 * Is this a client-side lock tree?
433 ldlm_side_t ns_client;
438 cfs_hash_t *ns_rs_hash;
443 cfs_spinlock_t ns_lock;
446 * big refcount (by bucket)
448 cfs_atomic_t ns_bref;
451 * Namespce connect flags supported by server (may be changed via proc,
452 * lru resize may be disabled/enabled).
454 __u64 ns_connect_flags;
457 * Client side orig connect flags supported by server.
459 __u64 ns_orig_connect_flags;
462 * Position in global namespace list.
464 cfs_list_t ns_list_chain;
467 * All root resources in namespace.
469 cfs_list_t ns_unused_list;
472 unsigned int ns_max_unused;
473 unsigned int ns_max_age;
474 unsigned int ns_timeouts;
478 unsigned int ns_ctime_age_limit;
481 * Next debug dump, jiffies.
483 cfs_time_t ns_next_dump;
485 ldlm_res_policy ns_policy;
486 struct ldlm_valblock_ops *ns_lvbo;
488 cfs_waitq_t ns_waitq;
489 struct ldlm_pool ns_pool;
490 ldlm_appetite_t ns_appetite;
493 * If more than \a ns_contended_locks found, the resource is considered
496 unsigned ns_contended_locks;
499 * The resource remembers contended state during \a ns_contention_time,
502 unsigned ns_contention_time;
505 * Limit size of nolock requests, in bytes.
507 unsigned ns_max_nolock_size;
508 /* callback to cancel locks before replaying it during recovery */
509 ldlm_cancel_for_recovery ns_cancel_for_recovery;
513 struct lprocfs_stats *ns_stats;
516 static inline int ns_is_client(struct ldlm_namespace *ns)
519 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
520 LDLM_NAMESPACE_SERVER)));
521 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
522 ns->ns_client == LDLM_NAMESPACE_SERVER);
523 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
526 static inline int ns_is_server(struct ldlm_namespace *ns)
529 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
530 LDLM_NAMESPACE_SERVER)));
531 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
532 ns->ns_client == LDLM_NAMESPACE_SERVER);
533 return ns->ns_client == LDLM_NAMESPACE_SERVER;
536 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
539 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
542 static inline void ns_register_cancel(struct ldlm_namespace *ns,
543 ldlm_cancel_for_recovery arg)
546 ns->ns_cancel_for_recovery = arg;
551 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
552 struct ldlm_lock_desc *new, void *data,
554 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
556 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
557 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
559 /* Interval node data for each LDLM_EXTENT lock */
560 struct ldlm_interval {
561 struct interval_node li_node; /* node for tree mgmt */
562 cfs_list_t li_group; /* the locks which have the same
563 * policy - group of the policy */
565 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
567 /* the interval tree must be accessed inside the resource lock. */
568 struct ldlm_interval_tree {
569 /* tree size, this variable is used to count
570 * granted PW locks in ldlm_extent_policy()*/
572 ldlm_mode_t lit_mode; /* lock mode */
573 struct interval_node *lit_root; /* actually ldlm_interval */
576 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
580 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
581 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
582 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
584 } ldlm_cancel_flags_t;
588 * Must be first in the structure.
590 struct portals_handle l_handle;
592 * Lock reference count.
596 * Internal spinlock protects l_resource. we should hold this lock
597 * first before grabbing res_lock.
599 cfs_spinlock_t l_lock;
601 * ldlm_lock_change_resource() can change this.
603 struct ldlm_resource *l_resource;
605 * Protected by ns_hash_lock. List item for client side lru list.
609 * Protected by lr_lock, linkage to resource's lock queues.
611 cfs_list_t l_res_link;
613 * Tree node for ldlm_extent.
615 struct ldlm_interval *l_tree_node;
617 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
620 cfs_hlist_node_t l_exp_hash;
622 * Protected by lr_lock. Requested mode.
624 ldlm_mode_t l_req_mode;
626 * Granted mode, also protected by lr_lock.
628 ldlm_mode_t l_granted_mode;
630 * Lock enqueue completion handler.
632 ldlm_completion_callback l_completion_ast;
634 * Lock blocking ast handler.
636 ldlm_blocking_callback l_blocking_ast;
638 * Lock glimpse handler.
640 ldlm_glimpse_callback l_glimpse_ast;
641 ldlm_weigh_callback l_weigh_ast;
646 struct obd_export *l_export;
648 * Lock connection export.
650 struct obd_export *l_conn_export;
653 * Remote lock handle.
655 struct lustre_handle l_remote_handle;
657 ldlm_policy_data_t l_policy_data;
660 * Protected by lr_lock. Various counters: readers, writers, etc.
666 * Set for locks that were removed from class hash table and will be
667 * destroyed when last reference to them is released. Set by
668 * ldlm_lock_destroy_internal().
670 * Protected by lock and resource locks.
674 * flag whether this is a server namespace lock
678 * If the lock is granted, a process sleeps on this waitq to learn when
679 * it's no longer in use. If the lock is not granted, a process sleeps
680 * on this waitq to learn when it becomes granted.
685 * Seconds. it will be updated if there is any activity related to
686 * the lock, e.g. enqueue the lock or send block AST.
688 cfs_time_t l_last_activity;
691 * Jiffies. Should be converted to time if needed.
693 cfs_time_t l_last_used;
695 struct ldlm_extent l_req_extent;
698 * Client-side-only members.
702 * Temporary storage for an LVB received during an enqueue operation.
710 * Server-side-only members.
713 /** connection cookie for the client originated the operation. */
714 __u64 l_client_cookie;
717 * Protected by elt_lock. Callbacks pending.
719 cfs_list_t l_pending_chain;
721 cfs_time_t l_callback_timeout;
724 * Pid which created this lock.
729 * For ldlm_add_ast_work_item().
733 * For ldlm_add_ast_work_item().
737 * For ldlm_add_ast_work_item().
741 struct ldlm_lock *l_blocking_lock;
745 * Protected by lr_lock, linkages to "skip lists".
747 cfs_list_t l_sl_mode;
748 cfs_list_t l_sl_policy;
749 struct lu_ref l_reference;
750 #if LUSTRE_TRACKS_LOCK_EXP_REFS
751 /* Debugging stuff for bug 20498, for tracking export
753 /** number of export references taken */
755 /** link all locks referencing one export */
756 cfs_list_t l_exp_refs_link;
757 /** referenced export object */
758 struct obd_export *l_exp_refs_target;
762 struct ldlm_resource {
763 struct ldlm_ns_bucket *lr_ns_bucket;
765 /* protected by ns_hash_lock */
766 cfs_hlist_node_t lr_hash;
767 cfs_spinlock_t lr_lock;
769 /* protected by lr_lock */
770 cfs_list_t lr_granted;
771 cfs_list_t lr_converting;
772 cfs_list_t lr_waiting;
773 ldlm_mode_t lr_most_restr;
774 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
775 struct ldlm_res_id lr_name;
776 cfs_atomic_t lr_refcount;
778 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
780 /* Server-side-only lock value block elements */
781 cfs_semaphore_t lr_lvb_sem;
785 /* when the resource was considered as contended */
786 cfs_time_t lr_contention_time;
788 * List of references to this resource. For debugging.
790 struct lu_ref lr_reference;
792 struct inode *lr_lvb_inode;
796 ldlm_ns_name(struct ldlm_namespace *ns)
798 return ns->ns_rs_hash->hs_name;
801 static inline struct ldlm_namespace *
802 ldlm_res_to_ns(struct ldlm_resource *res)
804 return res->lr_ns_bucket->nsb_namespace;
807 static inline struct ldlm_namespace *
808 ldlm_lock_to_ns(struct ldlm_lock *lock)
810 return ldlm_res_to_ns(lock->l_resource);
814 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
816 return ldlm_ns_name(ldlm_lock_to_ns(lock));
819 static inline struct adaptive_timeout *
820 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
822 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
825 struct ldlm_ast_work {
826 struct ldlm_lock *w_lock;
828 struct ldlm_lock_desc w_desc;
835 /* ldlm_enqueue parameters common */
836 struct ldlm_enqueue_info {
837 __u32 ei_type; /* Type of the lock being enqueued. */
838 __u32 ei_mode; /* Mode of the lock being enqueued. */
839 void *ei_cb_bl; /* blocking lock callback */
840 void *ei_cb_cp; /* lock completion callback */
841 void *ei_cb_gl; /* lock glimpse callback */
842 void *ei_cb_wg; /* lock weigh callback */
843 void *ei_cbdata; /* Data to be passed into callbacks. */
844 short ei_async:1; /* async request */
847 extern struct obd_ops ldlm_obd_ops;
849 extern char *ldlm_lockname[];
850 extern char *ldlm_typename[];
851 extern char *ldlm_it2str(int it);
853 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
856 if (((level) & D_CANTMASK) != 0 || \
857 ((libcfs_debug & (level)) != 0 && \
858 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
859 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
860 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
862 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
867 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
868 struct libcfs_debug_msg_data *data, const char *fmt,
870 __attribute__ ((format (printf, 4, 5)));
872 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
873 static cfs_debug_limit_state_t _ldlm_cdls; \
874 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
875 __FILE__, __FUNCTION__, __LINE__, \
879 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
880 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
882 #define LDLM_DEBUG(lock, fmt, a...) do { \
883 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
884 __FILE__, __FUNCTION__, __LINE__, \
887 #else /* !LIBCFS_DEBUG */
888 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
889 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
890 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
894 #define LDLM_DEBUG_NOLOCK(format, a...) \
895 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
897 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
898 int first_enq, ldlm_error_t *err,
899 cfs_list_t *work_list);
905 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
906 #define LDLM_ITER_STOP 2 /* stop iterating */
908 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
909 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
911 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
913 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
916 int ldlm_replay_locks(struct obd_import *imp);
917 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
918 ldlm_iterator_t iter, void *data);
921 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
924 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
928 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
929 void *data, int flag);
930 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
931 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
932 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
933 ldlm_blocking_callback, ldlm_glimpse_callback);
934 int ldlm_handle_convert(struct ptlrpc_request *req);
935 int ldlm_handle_cancel(struct ptlrpc_request *req);
936 int ldlm_request_cancel(struct ptlrpc_request *req,
937 const struct ldlm_request *dlm_req, int first);
938 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
939 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
940 void ldlm_revoke_export_locks(struct obd_export *exp);
941 int ldlm_get_ref(void);
942 void ldlm_put_ref(void);
943 int ldlm_init_export(struct obd_export *exp);
944 void ldlm_destroy_export(struct obd_export *exp);
947 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
948 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
949 void ldlm_lock2handle(const struct ldlm_lock *lock,
950 struct lustre_handle *lockh);
951 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
952 void ldlm_cancel_callback(struct ldlm_lock *);
953 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
955 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
957 return __ldlm_handle2lock(h, 0);
960 #define LDLM_LOCK_REF_DEL(lock) \
961 lu_ref_del(&lock->l_reference, "handle", cfs_current())
963 static inline struct ldlm_lock *
964 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
966 struct ldlm_lock *lock;
968 lock = __ldlm_handle2lock(h, flags);
970 LDLM_LOCK_REF_DEL(lock);
974 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
975 struct ptlrpc_request *r, int increase)
977 if (ldlm_res_to_ns(res)->ns_lvbo &&
978 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
979 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
985 int ldlm_error2errno(ldlm_error_t error);
986 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
987 * confuses user-space. */
988 #if LUSTRE_TRACKS_LOCK_EXP_REFS
989 void ldlm_dump_export_locks(struct obd_export *exp);
993 * Release a temporary lock reference obtained by ldlm_handle2lock() or
994 * __ldlm_handle2lock().
996 #define LDLM_LOCK_PUT(lock) \
998 LDLM_LOCK_REF_DEL(lock); \
999 /*LDLM_DEBUG((lock), "put");*/ \
1000 ldlm_lock_put(lock); \
1004 * Release a lock reference obtained by some other means (see
1007 #define LDLM_LOCK_RELEASE(lock) \
1009 /*LDLM_DEBUG((lock), "put");*/ \
1010 ldlm_lock_put(lock); \
1013 #define LDLM_LOCK_GET(lock) \
1015 ldlm_lock_get(lock); \
1016 /*LDLM_DEBUG((lock), "get");*/ \
1020 #define ldlm_lock_list_put(head, member, count) \
1022 struct ldlm_lock *_lock, *_next; \
1024 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1027 cfs_list_del_init(&_lock->member); \
1028 LDLM_LOCK_RELEASE(_lock); \
1033 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1034 void ldlm_lock_put(struct ldlm_lock *lock);
1035 void ldlm_lock_destroy(struct ldlm_lock *lock);
1036 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1037 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1038 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1039 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1040 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1041 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1042 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1043 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1044 const struct ldlm_res_id *, ldlm_type_t type,
1045 ldlm_policy_data_t *, ldlm_mode_t mode,
1046 struct lustre_handle *, int unref);
1047 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1049 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1050 void ldlm_lock_cancel(struct ldlm_lock *lock);
1051 void ldlm_reprocess_all(struct ldlm_resource *res);
1052 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1053 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
1054 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1055 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1058 struct ldlm_namespace *
1059 ldlm_namespace_new(struct obd_device *obd, char *name,
1060 ldlm_side_t client, ldlm_appetite_t apt,
1061 ldlm_ns_type_t ns_type);
1062 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1063 void ldlm_namespace_free(struct ldlm_namespace *ns,
1064 struct obd_import *imp, int force);
1065 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1066 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1067 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1068 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1069 void ldlm_namespace_get(struct ldlm_namespace *ns);
1070 void ldlm_namespace_put(struct ldlm_namespace *ns);
1071 int ldlm_proc_setup(void);
1073 void ldlm_proc_cleanup(void);
1075 static inline void ldlm_proc_cleanup(void) {}
1078 /* resource.c - internal */
1079 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1080 struct ldlm_resource *parent,
1081 const struct ldlm_res_id *,
1082 ldlm_type_t type, int create);
1083 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1084 int ldlm_resource_putref(struct ldlm_resource *res);
1085 void ldlm_resource_add_lock(struct ldlm_resource *res,
1087 struct ldlm_lock *lock);
1088 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1089 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1090 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1091 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1092 void ldlm_resource_dump(int level, struct ldlm_resource *);
1093 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1094 const struct ldlm_res_id *);
1096 #define LDLM_RESOURCE_ADDREF(res) do { \
1097 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1100 #define LDLM_RESOURCE_DELREF(res) do { \
1101 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1104 struct ldlm_callback_suite {
1105 ldlm_completion_callback lcs_completion;
1106 ldlm_blocking_callback lcs_blocking;
1107 ldlm_glimpse_callback lcs_glimpse;
1108 ldlm_weigh_callback lcs_weigh;
1111 /* ldlm_request.c */
1112 int ldlm_expired_completion_wait(void *data);
1113 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1114 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1115 void *data, int flag);
1116 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1117 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1118 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1119 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1120 struct ldlm_enqueue_info *einfo,
1121 const struct ldlm_res_id *res_id,
1122 ldlm_policy_data_t const *policy, int *flags,
1123 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1125 int ldlm_prep_enqueue_req(struct obd_export *exp,
1126 struct ptlrpc_request *req,
1127 cfs_list_t *cancels,
1129 int ldlm_prep_elc_req(struct obd_export *exp,
1130 struct ptlrpc_request *req,
1131 int version, int opc, int canceloff,
1132 cfs_list_t *cancels, int count);
1133 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1134 const struct ldlm_request *dlm_req,
1135 const struct ldlm_callback_suite *cbs);
1136 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1137 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1138 int *flags, void *lvb, __u32 lvb_len,
1139 struct lustre_handle *lockh, int rc);
1140 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1141 const struct ldlm_res_id *res_id,
1142 ldlm_type_t type, ldlm_policy_data_t *policy,
1143 ldlm_mode_t mode, int *flags,
1144 ldlm_blocking_callback blocking,
1145 ldlm_completion_callback completion,
1146 ldlm_glimpse_callback glimpse,
1147 void *data, __u32 lvb_len,
1148 const __u64 *client_cookie,
1149 struct lustre_handle *lockh);
1150 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1151 void *data, __u32 data_len);
1152 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1153 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1154 int ldlm_handle_convert0(struct ptlrpc_request *req,
1155 const struct ldlm_request *dlm_req);
1156 int ldlm_cli_cancel(struct lustre_handle *lockh);
1157 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1158 ldlm_cancel_flags_t flags, void *opaque);
1159 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1160 const struct ldlm_res_id *res_id,
1161 ldlm_policy_data_t *policy,
1163 ldlm_cancel_flags_t flags,
1165 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1166 int count, ldlm_cancel_flags_t flags);
1167 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1168 cfs_list_t *cancels,
1169 ldlm_policy_data_t *policy,
1170 ldlm_mode_t mode, int lock_flags,
1171 ldlm_cancel_flags_t cancel_flags, void *opaque);
1172 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1173 ldlm_cancel_flags_t flags);
1174 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1175 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1178 /* This has to be here because recursive inclusion sucks. */
1179 int intent_disposition(struct ldlm_reply *rep, int flag);
1180 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1183 /* ioctls for trying requests */
1184 #define IOC_LDLM_TYPE 'f'
1185 #define IOC_LDLM_MIN_NR 40
1187 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1188 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1189 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1190 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1191 #define IOC_LDLM_MAX_NR 43
1194 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1195 * than one lock_res is dead-lock safe.
1197 enum lock_res_type {
1202 static inline void lock_res(struct ldlm_resource *res)
1204 cfs_spin_lock(&res->lr_lock);
1207 static inline void lock_res_nested(struct ldlm_resource *res,
1208 enum lock_res_type mode)
1210 cfs_spin_lock_nested(&res->lr_lock, mode);
1214 static inline void unlock_res(struct ldlm_resource *res)
1216 cfs_spin_unlock(&res->lr_lock);
1219 static inline void check_res_locked(struct ldlm_resource *res)
1221 LASSERT_SPIN_LOCKED(&res->lr_lock);
1224 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1225 void unlock_res_and_lock(struct ldlm_lock *lock);
1228 void ldlm_pools_recalc(ldlm_side_t client);
1229 int ldlm_pools_init(void);
1230 void ldlm_pools_fini(void);
1232 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1233 int idx, ldlm_side_t client);
1234 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1235 unsigned int gfp_mask);
1236 void ldlm_pool_fini(struct ldlm_pool *pl);
1237 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1238 int ldlm_pool_recalc(struct ldlm_pool *pl);
1239 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1240 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1241 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1242 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1243 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1244 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1245 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1246 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1247 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);