1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _LUSTRE_DLM_H__
38 #define _LUSTRE_DLM_H__
40 /** \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 /* see ldlm_cli_cancel_unused */
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 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
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 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
172 #define LDLM_FL_ASYNC 0x8000000
174 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
175 * such that server send blocking ast for conflict locks to this client for
176 * the 1st operation, whereas the 2nd operation has canceled this lock and
177 * is waiting for rpc_lock which is taken by the 1st operation.
178 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
179 * ELC code to cancel it.
180 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
181 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
182 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
183 * to send a separate CANCEL rpc. */
184 #define LDLM_FL_BL_AST 0x10000000
185 #define LDLM_FL_BL_DONE 0x20000000
187 /* measure lock contention and return -EUSERS if locking contention is high */
188 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
190 /* These are flags that are mapped into the flags and ASTs of blocking locks */
191 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
193 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
194 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
197 * --------------------------------------------------------------------------
198 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
199 * 0x80000000 will not be sent over the wire.
200 * --------------------------------------------------------------------------
203 /* Used for marking lock as an target for -EINTR while cp_ast sleep
204 * emulation + race with upcoming bl_ast. */
205 #define LDLM_FL_FAIL_LOC 0x100000000ULL
207 /* The blocking callback is overloaded to perform two functions. These flags
208 * indicate which operation should be performed. */
209 #define LDLM_CB_BLOCKING 1
210 #define LDLM_CB_CANCELING 2
212 /* compatibility matrix */
213 #define LCK_COMPAT_EX LCK_NL
214 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
215 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
216 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
217 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
218 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
219 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
220 #define LCK_COMPAT_COS (LCK_COS)
222 extern ldlm_mode_t lck_compat_array[];
224 static inline void lockmode_verify(ldlm_mode_t mode)
226 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
229 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
231 return (lck_compat_array[exist_mode] & new_mode);
236 * cluster name spaces
240 #define DLM_OST_NAMESPACE 1
241 #define DLM_MDS_NAMESPACE 2
244 - do we just separate this by security domains and use a prefix for
245 multiple namespaces in the same domain?
255 * waiting_locks_spinlock
270 struct ldlm_resource;
271 struct ldlm_namespace;
273 struct ldlm_pool_ops {
274 int (*po_recalc)(struct ldlm_pool *pl);
275 int (*po_shrink)(struct ldlm_pool *pl, int nr,
276 unsigned int gfp_mask);
277 int (*po_setup)(struct ldlm_pool *pl, int limit);
281 * One second for pools thread check interval. Each pool has own period.
283 #define LDLM_POOLS_THREAD_PERIOD (1)
286 * ~6% margin for modest pools. See ldlm_pool.c for details.
288 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
291 * Default recalc period for server side pools in sec.
293 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
296 * Default recalc period for client side pools in sec.
298 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
302 * Pool proc directory.
304 cfs_proc_dir_entry_t *pl_proc_dir;
306 * Pool name, should be long enough to contain compound proc entry name.
310 * Lock for protecting slv/clv updates.
312 cfs_spinlock_t pl_lock;
314 * Number of allowed locks in in pool, both, client and server side.
316 cfs_atomic_t pl_limit;
318 * Number of granted locks in
320 cfs_atomic_t pl_granted;
324 cfs_atomic_t pl_grant_rate;
328 cfs_atomic_t pl_cancel_rate;
330 * Grant speed (GR-CR) per T.
332 cfs_atomic_t pl_grant_speed;
334 * Server lock volume. Protected by pl_lock.
336 __u64 pl_server_lock_volume;
338 * Current biggest client lock volume. Protected by pl_lock.
340 __u64 pl_client_lock_volume;
342 * Lock volume factor. SLV on client is calculated as following:
343 * server_slv * lock_volume_factor.
345 cfs_atomic_t pl_lock_volume_factor;
347 * Time when last slv from server was obtained.
349 time_t pl_recalc_time;
351 * Recalc period for pool.
353 time_t pl_recalc_period;
355 * Recalc and shrink ops.
357 struct ldlm_pool_ops *pl_ops;
359 * Number of planned locks for next period.
365 struct lprocfs_stats *pl_stats;
368 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
369 void *req_cookie, ldlm_mode_t mode, int flags,
372 struct ldlm_valblock_ops {
373 int (*lvbo_init)(struct ldlm_resource *res);
374 int (*lvbo_update)(struct ldlm_resource *res,
375 struct ptlrpc_request *r,
380 LDLM_NAMESPACE_GREEDY = 1 << 0,
381 LDLM_NAMESPACE_MODEST = 1 << 1
385 * Default values for the "max_nolock_size", "contention_time" and
386 * "contended_locks" namespace tunables.
388 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
389 #define NS_DEFAULT_CONTENTION_SECONDS 2
390 #define NS_DEFAULT_CONTENDED_LOCKS 32
392 struct ldlm_namespace {
394 * Namespace name. Used for logging, etc.
399 * Is this a client-side lock tree?
401 ldlm_side_t ns_client;
404 * Namespce connect flags supported by server (may be changed via proc,
405 * lru resize may be disabled/enabled).
407 __u64 ns_connect_flags;
410 * Client side orig connect flags supported by server.
412 __u64 ns_orig_connect_flags;
415 * Hash table for namespace.
418 cfs_spinlock_t ns_hash_lock;
421 * Count of resources in the hash.
426 * All root resources in namespace.
428 cfs_list_t ns_root_list;
431 * Position in global namespace list.
433 cfs_list_t ns_list_chain;
436 * All root resources in namespace.
438 cfs_list_t ns_unused_list;
440 cfs_spinlock_t ns_unused_lock;
442 unsigned int ns_max_unused;
443 unsigned int ns_max_age;
444 unsigned int ns_timeouts;
448 unsigned int ns_ctime_age_limit;
451 * Next debug dump, jiffies.
453 cfs_time_t ns_next_dump;
455 cfs_atomic_t ns_locks;
457 ldlm_res_policy ns_policy;
458 struct ldlm_valblock_ops *ns_lvbo;
460 cfs_waitq_t ns_waitq;
461 struct ldlm_pool ns_pool;
462 ldlm_appetite_t ns_appetite;
465 * If more than \a ns_contended_locks found, the resource is considered
468 unsigned ns_contended_locks;
471 * The resource remembers contended state during \a ns_contention_time,
474 unsigned ns_contention_time;
477 * Limit size of nolock requests, in bytes.
479 unsigned ns_max_nolock_size;
482 * Backward link to obd, required for ldlm pool to store new SLV.
484 struct obd_device *ns_obd;
486 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
489 static inline int ns_is_client(struct ldlm_namespace *ns)
492 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
493 LDLM_NAMESPACE_SERVER)));
494 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
495 ns->ns_client == LDLM_NAMESPACE_SERVER);
496 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
499 static inline int ns_is_server(struct ldlm_namespace *ns)
502 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
503 LDLM_NAMESPACE_SERVER)));
504 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
505 ns->ns_client == LDLM_NAMESPACE_SERVER);
506 return ns->ns_client == LDLM_NAMESPACE_SERVER;
509 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
512 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
517 * Resource hash table
521 #define RES_HASH_BITS 12
522 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
523 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
527 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
528 struct ldlm_lock_desc *new, void *data,
530 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
532 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
533 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
535 /* Interval node data for each LDLM_EXTENT lock */
536 struct ldlm_interval {
537 struct interval_node li_node; /* node for tree mgmt */
538 cfs_list_t li_group; /* the locks which have the same
539 * policy - group of the policy */
541 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
543 /* the interval tree must be accessed inside the resource lock. */
544 struct ldlm_interval_tree {
545 /* tree size, this variable is used to count
546 * granted PW locks in ldlm_extent_policy()*/
548 ldlm_mode_t lit_mode; /* lock mode */
549 struct interval_node *lit_root; /* actually ldlm_interval */
552 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
556 * Must be first in the structure.
558 struct portals_handle l_handle;
560 * Lock reference count.
564 * Internal spinlock protects l_resource. we should hold this lock
565 * first before grabbing res_lock.
567 cfs_spinlock_t l_lock;
569 * ldlm_lock_change_resource() can change this.
571 struct ldlm_resource *l_resource;
573 * Protected by ns_hash_lock. List item for client side lru list.
577 * Protected by lr_lock, linkage to resource's lock queues.
579 cfs_list_t l_res_link;
581 * Tree node for ldlm_extent.
583 struct ldlm_interval *l_tree_node;
585 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
588 cfs_hlist_node_t l_exp_hash;
590 * Protected by lr_lock. Requested mode.
592 ldlm_mode_t l_req_mode;
594 * Granted mode, also protected by lr_lock.
596 ldlm_mode_t l_granted_mode;
598 * Lock enqueue completion handler.
600 ldlm_completion_callback l_completion_ast;
602 * Lock blocking ast handler.
604 ldlm_blocking_callback l_blocking_ast;
606 * Lock glimpse handler.
608 ldlm_glimpse_callback l_glimpse_ast;
609 ldlm_weigh_callback l_weigh_ast;
614 struct obd_export *l_export;
616 * Lock connection export.
618 struct obd_export *l_conn_export;
621 * Remote lock handle.
623 struct lustre_handle l_remote_handle;
625 ldlm_policy_data_t l_policy_data;
628 * Protected by lr_lock. Various counters: readers, writers, etc.
634 * Set for locks that were removed from class hash table and will be
635 * destroyed when last reference to them is released. Set by
636 * ldlm_lock_destroy_internal().
638 * Protected by lock and resource locks.
643 * If the lock is granted, a process sleeps on this waitq to learn when
644 * it's no longer in use. If the lock is not granted, a process sleeps
645 * on this waitq to learn when it becomes granted.
650 * Seconds. it will be updated if there is any activity related to
651 * the lock, e.g. enqueue the lock or send block AST.
653 cfs_time_t l_last_activity;
656 * Jiffies. Should be converted to time if needed.
658 cfs_time_t l_last_used;
660 struct ldlm_extent l_req_extent;
663 * Client-side-only members.
667 * Temporary storage for an LVB received during an enqueue operation.
673 cfs_spinlock_t l_extents_list_lock;
674 cfs_list_t l_extents_list;
676 cfs_list_t l_cache_locks_list;
679 * Server-side-only members.
682 /** connection cookie for the client originated the operation. */
683 __u64 l_client_cookie;
686 * Protected by elt_lock. Callbacks pending.
688 cfs_list_t l_pending_chain;
690 cfs_time_t l_callback_timeout;
693 * Pid which created this lock.
698 * For ldlm_add_ast_work_item().
702 * For ldlm_add_ast_work_item().
706 * For ldlm_add_ast_work_item().
710 struct ldlm_lock *l_blocking_lock;
714 * Protected by lr_lock, linkages to "skip lists".
716 cfs_list_t l_sl_mode;
717 cfs_list_t l_sl_policy;
718 struct lu_ref l_reference;
719 #if LUSTRE_TRACKS_LOCK_EXP_REFS
720 /* Debugging stuff for bug 20498, for tracking export
722 /** number of export references taken */
724 /** link all locks referencing one export */
725 cfs_list_t l_exp_refs_link;
726 /** referenced export object */
727 struct obd_export *l_exp_refs_target;
731 struct ldlm_resource {
732 struct ldlm_namespace *lr_namespace;
734 /* protected by ns_hash_lock */
736 struct ldlm_resource *lr_parent; /* 0 for a root resource */
737 cfs_list_t lr_children; /* list head for child resources */
738 cfs_list_t lr_childof; /* part of ns_root_list if root res,
739 * part of lr_children if child */
740 cfs_spinlock_t lr_lock;
742 /* protected by lr_lock */
743 cfs_list_t lr_granted;
744 cfs_list_t lr_converting;
745 cfs_list_t lr_waiting;
746 ldlm_mode_t lr_most_restr;
747 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
748 struct ldlm_res_id lr_name;
749 cfs_atomic_t lr_refcount;
751 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
753 /* Server-side-only lock value block elements */
754 cfs_semaphore_t lr_lvb_sem;
758 /* when the resource was considered as contended */
759 cfs_time_t lr_contention_time;
761 * List of references to this resource. For debugging.
763 struct lu_ref lr_reference;
766 struct ldlm_ast_work {
767 struct ldlm_lock *w_lock;
769 struct ldlm_lock_desc w_desc;
776 /* ldlm_enqueue parameters common */
777 struct ldlm_enqueue_info {
778 __u32 ei_type; /* Type of the lock being enqueued. */
779 __u32 ei_mode; /* Mode of the lock being enqueued. */
780 void *ei_cb_bl; /* blocking lock callback */
781 void *ei_cb_cp; /* lock completion callback */
782 void *ei_cb_gl; /* lock glimpse callback */
783 void *ei_cb_wg; /* lock weigh callback */
784 void *ei_cbdata; /* Data to be passed into callbacks. */
785 short ei_async:1; /* async request */
788 extern struct obd_ops ldlm_obd_ops;
790 extern char *ldlm_lockname[];
791 extern char *ldlm_typename[];
792 extern char *ldlm_it2str(int it);
794 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
797 if (((level) & D_CANTMASK) != 0 || \
798 ((libcfs_debug & (level)) != 0 && \
799 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
800 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
801 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
803 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
808 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
809 struct libcfs_debug_msg_data *data, const char *fmt,
811 __attribute__ ((format (printf, 4, 5)));
813 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
814 static cfs_debug_limit_state_t _ldlm_cdls; \
815 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
816 __FILE__, __FUNCTION__, __LINE__, \
820 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
821 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
823 #define LDLM_DEBUG(lock, fmt, a...) do { \
824 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
825 __FILE__, __FUNCTION__, __LINE__, \
828 #else /* !LIBCFS_DEBUG */
829 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
830 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
831 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
835 #define LDLM_DEBUG_NOLOCK(format, a...) \
836 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
838 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
839 int first_enq, ldlm_error_t *err,
840 cfs_list_t *work_list);
846 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
847 #define LDLM_ITER_STOP 2 /* stop iterating */
849 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
850 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
852 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
854 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
856 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
857 ldlm_res_iterator_t iter, void *closure);
859 int ldlm_replay_locks(struct obd_import *imp);
860 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
861 ldlm_iterator_t iter, void *data);
864 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
867 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
871 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
872 void *data, int flag);
873 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
874 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
875 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
876 ldlm_blocking_callback, ldlm_glimpse_callback);
877 int ldlm_handle_convert(struct ptlrpc_request *req);
878 int ldlm_handle_cancel(struct ptlrpc_request *req);
879 int ldlm_request_cancel(struct ptlrpc_request *req,
880 const struct ldlm_request *dlm_req, int first);
881 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
882 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
883 void ldlm_revoke_export_locks(struct obd_export *exp);
884 int ldlm_get_ref(void);
885 void ldlm_put_ref(void);
886 int ldlm_init_export(struct obd_export *exp);
887 void ldlm_destroy_export(struct obd_export *exp);
890 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
891 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
892 void ldlm_lock2handle(const struct ldlm_lock *lock,
893 struct lustre_handle *lockh);
894 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
895 void ldlm_cancel_callback(struct ldlm_lock *);
896 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
898 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
900 return __ldlm_handle2lock(h, 0);
903 #define LDLM_LOCK_REF_DEL(lock) \
904 lu_ref_del(&lock->l_reference, "handle", cfs_current())
906 static inline struct ldlm_lock *
907 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
909 struct ldlm_lock *lock;
911 lock = __ldlm_handle2lock(h, flags);
913 LDLM_LOCK_REF_DEL(lock);
917 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
918 struct ptlrpc_request *r, int increase)
920 if (res->lr_namespace->ns_lvbo &&
921 res->lr_namespace->ns_lvbo->lvbo_update) {
922 return res->lr_namespace->ns_lvbo->lvbo_update(res, r,
928 int ldlm_error2errno(ldlm_error_t error);
929 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
930 * confuses user-space. */
931 #if LUSTRE_TRACKS_LOCK_EXP_REFS
932 void ldlm_dump_export_locks(struct obd_export *exp);
936 * Release a temporary lock reference obtained by ldlm_handle2lock() or
937 * __ldlm_handle2lock().
939 #define LDLM_LOCK_PUT(lock) \
941 LDLM_LOCK_REF_DEL(lock); \
942 /*LDLM_DEBUG((lock), "put");*/ \
943 ldlm_lock_put(lock); \
947 * Release a lock reference obtained by some other means (see
950 #define LDLM_LOCK_RELEASE(lock) \
952 /*LDLM_DEBUG((lock), "put");*/ \
953 ldlm_lock_put(lock); \
956 #define LDLM_LOCK_GET(lock) \
958 ldlm_lock_get(lock); \
959 /*LDLM_DEBUG((lock), "get");*/ \
963 #define ldlm_lock_list_put(head, member, count) \
965 struct ldlm_lock *_lock, *_next; \
967 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
970 cfs_list_del_init(&_lock->member); \
971 LDLM_LOCK_RELEASE(_lock); \
976 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
977 void ldlm_lock_put(struct ldlm_lock *lock);
978 void ldlm_lock_destroy(struct ldlm_lock *lock);
979 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
980 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
981 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
982 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
983 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
984 void ldlm_lock_allow_match(struct ldlm_lock *lock);
985 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
986 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
987 const struct ldlm_res_id *, ldlm_type_t type,
988 ldlm_policy_data_t *, ldlm_mode_t mode,
989 struct lustre_handle *, int unref);
990 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
992 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
993 void ldlm_lock_cancel(struct ldlm_lock *lock);
994 void ldlm_reprocess_all(struct ldlm_resource *res);
995 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
996 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
997 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
998 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1001 struct ldlm_namespace *
1002 ldlm_namespace_new(struct obd_device *obd, char *name,
1003 ldlm_side_t client, ldlm_appetite_t apt);
1004 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1005 void ldlm_namespace_free(struct ldlm_namespace *ns,
1006 struct obd_import *imp, int force);
1007 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1008 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1009 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1010 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1011 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
1012 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
1013 void ldlm_namespace_get(struct ldlm_namespace *ns);
1014 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
1015 int ldlm_proc_setup(void);
1017 void ldlm_proc_cleanup(void);
1019 static inline void ldlm_proc_cleanup(void) {}
1022 /* resource.c - internal */
1023 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1024 struct ldlm_resource *parent,
1025 const struct ldlm_res_id *,
1026 ldlm_type_t type, int create);
1027 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1028 int ldlm_resource_putref(struct ldlm_resource *res);
1029 void ldlm_resource_add_lock(struct ldlm_resource *res,
1031 struct ldlm_lock *lock);
1032 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1033 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1034 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1035 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1036 void ldlm_resource_dump(int level, struct ldlm_resource *);
1037 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1038 const struct ldlm_res_id *);
1040 #define LDLM_RESOURCE_ADDREF(res) do { \
1041 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1044 #define LDLM_RESOURCE_DELREF(res) do { \
1045 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1048 struct ldlm_callback_suite {
1049 ldlm_completion_callback lcs_completion;
1050 ldlm_blocking_callback lcs_blocking;
1051 ldlm_glimpse_callback lcs_glimpse;
1052 ldlm_weigh_callback lcs_weigh;
1055 /* ldlm_request.c */
1056 int ldlm_expired_completion_wait(void *data);
1057 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1058 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1059 void *data, int flag);
1060 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1061 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1062 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1063 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1064 struct ldlm_enqueue_info *einfo,
1065 const struct ldlm_res_id *res_id,
1066 ldlm_policy_data_t const *policy, int *flags,
1067 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1069 int ldlm_prep_enqueue_req(struct obd_export *exp,
1070 struct ptlrpc_request *req,
1071 cfs_list_t *cancels,
1073 int ldlm_prep_elc_req(struct obd_export *exp,
1074 struct ptlrpc_request *req,
1075 int version, int opc, int canceloff,
1076 cfs_list_t *cancels, int count);
1077 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1078 const struct ldlm_request *dlm_req,
1079 const struct ldlm_callback_suite *cbs);
1080 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1081 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1082 int *flags, void *lvb, __u32 lvb_len,
1083 struct lustre_handle *lockh, int rc);
1084 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1085 const struct ldlm_res_id *res_id,
1086 ldlm_type_t type, ldlm_policy_data_t *policy,
1087 ldlm_mode_t mode, int *flags,
1088 ldlm_blocking_callback blocking,
1089 ldlm_completion_callback completion,
1090 ldlm_glimpse_callback glimpse,
1091 void *data, __u32 lvb_len,
1092 const __u64 *client_cookie,
1093 struct lustre_handle *lockh);
1094 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1095 void *data, __u32 data_len);
1096 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1097 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1098 int ldlm_handle_convert0(struct ptlrpc_request *req,
1099 const struct ldlm_request *dlm_req);
1100 int ldlm_cli_cancel(struct lustre_handle *lockh);
1101 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1102 int flags, void *opaque);
1103 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1104 const struct ldlm_res_id *res_id,
1105 ldlm_policy_data_t *policy,
1106 ldlm_mode_t mode, int flags, void *opaque);
1107 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1108 int count, int flags);
1109 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1110 cfs_list_t *cancels,
1111 ldlm_policy_data_t *policy,
1112 ldlm_mode_t mode, int lock_flags,
1113 int cancel_flags, void *opaque);
1114 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1115 struct ptlrpc_request *req, int flags);
1118 /* This has to be here because recursive inclusion sucks. */
1119 int intent_disposition(struct ldlm_reply *rep, int flag);
1120 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1123 /* ioctls for trying requests */
1124 #define IOC_LDLM_TYPE 'f'
1125 #define IOC_LDLM_MIN_NR 40
1127 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1128 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1129 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1130 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1131 #define IOC_LDLM_MAX_NR 43
1134 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1135 * than one lock_res is dead-lock safe.
1137 enum lock_res_type {
1142 static inline void lock_res(struct ldlm_resource *res)
1144 cfs_spin_lock(&res->lr_lock);
1147 static inline void lock_res_nested(struct ldlm_resource *res,
1148 enum lock_res_type mode)
1150 cfs_spin_lock_nested(&res->lr_lock, mode);
1154 static inline void unlock_res(struct ldlm_resource *res)
1156 cfs_spin_unlock(&res->lr_lock);
1159 static inline void check_res_locked(struct ldlm_resource *res)
1161 LASSERT_SPIN_LOCKED(&res->lr_lock);
1164 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1165 void unlock_res_and_lock(struct ldlm_lock *lock);
1168 void ldlm_pools_recalc(ldlm_side_t client);
1169 int ldlm_pools_init(void);
1170 void ldlm_pools_fini(void);
1172 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1173 int idx, ldlm_side_t client);
1174 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1175 unsigned int gfp_mask);
1176 void ldlm_pool_fini(struct ldlm_pool *pl);
1177 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1178 int ldlm_pool_recalc(struct ldlm_pool *pl);
1179 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1180 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1181 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1182 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1183 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1184 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1185 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1186 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1187 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);