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 #if defined(__linux__)
41 #include <linux/lustre_dlm.h>
42 #elif defined(__APPLE__)
43 #include <darwin/lustre_dlm.h>
44 #elif defined(__WINNT__)
45 #include <winnt/lustre_dlm.h>
47 #error Unsupported operating system.
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <lustre_import.h>
53 #include <lustre_handles.h>
54 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
60 #define OBD_LDLM_DEVICENAME "ldlm"
62 #ifdef HAVE_BGL_SUPPORT
63 /* 1.5 times the maximum 128 tasks available in VN mode */
64 #define LDLM_DEFAULT_LRU_SIZE 196
66 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
68 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
69 #define LDLM_CTIME_AGE_LIMIT (10)
74 ELDLM_LOCK_CHANGED = 300,
75 ELDLM_LOCK_ABORTED = 301,
76 ELDLM_LOCK_REPLACED = 302,
77 ELDLM_NO_LOCK_DATA = 303,
79 ELDLM_NAMESPACE_EXISTS = 400,
80 ELDLM_BAD_NAMESPACE = 401
84 LDLM_NAMESPACE_SERVER = 1 << 0,
85 LDLM_NAMESPACE_CLIENT = 1 << 1
88 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
90 /* If the server returns one of these flags, then the lock was put on that list.
91 * If the client sends one of these flags (during recovery ONLY!), it wants the
92 * lock added to the specified list, no questions asked. -p */
93 #define LDLM_FL_BLOCK_GRANTED 0x000002
94 #define LDLM_FL_BLOCK_CONV 0x000004
95 #define LDLM_FL_BLOCK_WAIT 0x000008
97 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
98 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
99 * queued for sending. */
100 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
101 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
103 /* Lock is being replayed. This could probably be implied by the fact that one
104 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
105 #define LDLM_FL_REPLAY 0x000100
107 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
108 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
110 /* don't run the cancel callback under ldlm_cli_cancel_unused */
111 #define LDLM_FL_FAILED 0x000800
113 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
114 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
115 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
116 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
117 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
119 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
122 /* file & record locking */
123 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
124 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
126 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
127 * the LVB filling happens _after_ the lock has been granted, so another thread
128 * can match`t before the LVB has been updated. As a dirty hack, we set
129 * LDLM_FL_LVB_READY only after we've done the LVB poop.
130 * this is only needed on lov/osc now, where lvb is actually used and callers
131 * must set it in input flags.
133 * The proper fix is to do the granting inside of the completion AST, which can
134 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
135 * pretty high-risk, though, and would need a lot more testing. */
137 #define LDLM_FL_LVB_READY 0x100000
139 /* A lock contributes to the kms calculation until it has finished the part
140 * of it's cancelation that performs write back on its dirty pages. It
141 * can remain on the granted list during this whole time. Threads racing
142 * to update the kms after performing their writeback need to know to
143 * exclude each others locks from the calculation as they walk the granted
145 #define LDLM_FL_KMS_IGNORE 0x200000
147 /* Immediatelly cancel such locks when they block some other locks. Send
148 * cancel notification to original lock holder, but expect no reply. This is
149 * for clients (like liblustre) that cannot be expected to reliably response
150 * to blocking ast. */
151 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
153 /* Flags flags inherited from parent lock when doing intents. */
154 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
156 /* completion ast to be executed */
157 #define LDLM_FL_CP_REQD 0x1000000
159 /* cleanup_resource has already handled the lock */
160 #define LDLM_FL_CLEANED 0x2000000
162 /* optimization hint: LDLM can run blocking callback from current context
163 * w/o involving separate thread. in order to decrease cs rate */
164 #define LDLM_FL_ATOMIC_CB 0x4000000
166 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
167 #define LDLM_FL_ASYNC 0x8000000
169 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
170 * such that server send blocking ast for conflict locks to this client for
171 * the 1st operation, whereas the 2nd operation has canceled this lock and
172 * is waiting for rpc_lock which is taken by the 1st operation.
173 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
174 * ELC code to cancel it.
175 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
176 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
177 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
178 * to send a separate CANCEL rpc. */
179 #define LDLM_FL_BL_AST 0x10000000
180 #define LDLM_FL_BL_DONE 0x20000000
182 /* measure lock contention and return -EUSERS if locking contention is high */
183 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
185 /* These are flags that are mapped into the flags and ASTs of blocking locks */
186 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
188 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
189 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
192 * --------------------------------------------------------------------------
193 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
194 * 0x80000000 will not be sent over the wire.
195 * --------------------------------------------------------------------------
198 /* Used for marking lock as an target for -EINTR while cp_ast sleep
199 * emulation + race with upcoming bl_ast. */
200 #define LDLM_FL_FAIL_LOC 0x100000000ULL
202 /* The blocking callback is overloaded to perform two functions. These flags
203 * indicate which operation should be performed. */
204 #define LDLM_CB_BLOCKING 1
205 #define LDLM_CB_CANCELING 2
207 /* compatibility matrix */
208 #define LCK_COMPAT_EX LCK_NL
209 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
210 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
211 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
212 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
213 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
214 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
215 #define LCK_COMPAT_COS (LCK_COS)
217 extern ldlm_mode_t lck_compat_array[];
219 static inline void lockmode_verify(ldlm_mode_t mode)
221 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
224 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
226 return (lck_compat_array[exist_mode] & new_mode);
231 * cluster name spaces
235 #define DLM_OST_NAMESPACE 1
236 #define DLM_MDS_NAMESPACE 2
239 - do we just separate this by security domains and use a prefix for
240 multiple namespaces in the same domain?
250 * waiting_locks_spinlock
265 struct ldlm_resource;
266 struct ldlm_namespace;
268 struct ldlm_pool_ops {
269 int (*po_recalc)(struct ldlm_pool *pl);
270 int (*po_shrink)(struct ldlm_pool *pl, int nr,
271 unsigned int gfp_mask);
272 int (*po_setup)(struct ldlm_pool *pl, int limit);
276 * One second for pools thread check interval. Each pool has own period.
278 #define LDLM_POOLS_THREAD_PERIOD (1)
281 * 5% margin for modest pools. See ldlm_pool.c for details.
283 #define LDLM_POOLS_MODEST_MARGIN (5)
286 * Default recalc period for server side pools in sec.
288 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
291 * Default recalc period for client side pools in sec.
293 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
297 * Pool proc directory.
299 cfs_proc_dir_entry_t *pl_proc_dir;
301 * Pool name, should be long enough to contain compound proc entry name.
305 * Lock for protecting slv/clv updates.
309 * Number of allowed locks in in pool, both, client and server side.
313 * Number of granted locks in
319 atomic_t pl_grant_rate;
323 atomic_t pl_cancel_rate;
325 * Grant speed (GR-CR) per T.
327 atomic_t pl_grant_speed;
329 * Server lock volume. Protected by pl_lock.
331 __u64 pl_server_lock_volume;
333 * Current biggest client lock volume. Protected by pl_lock.
335 __u64 pl_client_lock_volume;
337 * Lock volume factor. SLV on client is calculated as following:
338 * server_slv * lock_volume_factor.
340 atomic_t pl_lock_volume_factor;
342 * Time when last slv from server was obtained.
344 time_t pl_recalc_time;
346 * Recalc period for pool.
348 time_t pl_recalc_period;
350 * Recalc and shrink ops.
352 struct ldlm_pool_ops *pl_ops;
354 * Number of planned locks for next period.
360 struct lprocfs_stats *pl_stats;
363 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
364 void *req_cookie, ldlm_mode_t mode, int flags,
367 struct ldlm_valblock_ops {
368 int (*lvbo_init)(struct ldlm_resource *res);
369 int (*lvbo_update)(struct ldlm_resource *res,
370 struct ptlrpc_request *r,
375 LDLM_NAMESPACE_GREEDY = 1 << 0,
376 LDLM_NAMESPACE_MODEST = 1 << 1
380 * Default values for the "max_nolock_size", "contention_time" and
381 * "contended_locks" namespace tunables.
383 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
384 #define NS_DEFAULT_CONTENTION_SECONDS 2
385 #define NS_DEFAULT_CONTENDED_LOCKS 32
387 struct ldlm_namespace {
389 * Namespace name. Used for logging, etc.
394 * Is this a client-side lock tree?
396 ldlm_side_t ns_client;
399 * Namespce connect flags supported by server (may be changed via proc,
400 * lru resize may be disabled/enabled).
402 __u64 ns_connect_flags;
405 * Client side orig connect flags supported by server.
407 __u64 ns_orig_connect_flags;
410 * Hash table for namespace.
412 struct list_head *ns_hash;
413 spinlock_t ns_hash_lock;
416 * Count of resources in the hash.
421 * All root resources in namespace.
423 struct list_head ns_root_list;
426 * Position in global namespace list.
428 struct list_head ns_list_chain;
431 * All root resources in namespace.
433 struct list_head ns_unused_list;
435 spinlock_t ns_unused_lock;
437 unsigned int ns_max_unused;
438 unsigned int ns_max_age;
439 unsigned int ns_timeouts;
443 unsigned int ns_ctime_age_limit;
446 * Next debug dump, jiffies.
448 cfs_time_t ns_next_dump;
452 ldlm_res_policy ns_policy;
453 struct ldlm_valblock_ops *ns_lvbo;
455 cfs_waitq_t ns_waitq;
456 struct ldlm_pool ns_pool;
457 ldlm_appetite_t ns_appetite;
460 * If more than \a ns_contended_locks found, the resource is considered
463 unsigned ns_contended_locks;
466 * The resource remembers contended state during \a ns_contention_time,
469 unsigned ns_contention_time;
472 * Limit size of nolock requests, in bytes.
474 unsigned ns_max_nolock_size;
477 * Backward link to obd, required for ldlm pool to store new SLV.
479 struct obd_device *ns_obd;
481 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
484 static inline int ns_is_client(struct ldlm_namespace *ns)
487 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
488 LDLM_NAMESPACE_SERVER)));
489 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
490 ns->ns_client == LDLM_NAMESPACE_SERVER);
491 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
494 static inline int ns_is_server(struct ldlm_namespace *ns)
497 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
498 LDLM_NAMESPACE_SERVER)));
499 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
500 ns->ns_client == LDLM_NAMESPACE_SERVER);
501 return ns->ns_client == LDLM_NAMESPACE_SERVER;
504 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
507 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
512 * Resource hash table
516 #define RES_HASH_BITS 12
517 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
518 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
522 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
523 struct ldlm_lock_desc *new, void *data,
525 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
527 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
528 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
530 /* Interval node data for each LDLM_EXTENT lock */
531 struct ldlm_interval {
532 struct interval_node li_node; /* node for tree mgmt */
533 struct list_head li_group; /* the locks which have the same
534 * policy - group of the policy */
536 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
538 /* the interval tree must be accessed inside the resource lock. */
539 struct ldlm_interval_tree {
540 /* tree size, this variable is used to count
541 * granted PW locks in ldlm_extent_policy()*/
543 ldlm_mode_t lit_mode; /* lock mode */
544 struct interval_node *lit_root; /* actually ldlm_interval */
547 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
551 * Must be first in the structure.
553 struct portals_handle l_handle;
555 * Lock reference count.
559 * Internal spinlock protects l_resource. we should hold this lock
560 * first before grabbing res_lock.
564 * ldlm_lock_change_resource() can change this.
566 struct ldlm_resource *l_resource;
568 * Protected by ns_hash_lock. List item for client side lru list.
570 struct list_head l_lru;
572 * Protected by lr_lock, linkage to resource's lock queues.
574 struct list_head l_res_link;
576 * Tree node for ldlm_extent.
578 struct ldlm_interval *l_tree_node;
580 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
583 struct hlist_node l_exp_hash;
585 * Protected by lr_lock. Requested mode.
587 ldlm_mode_t l_req_mode;
589 * Granted mode, also protected by lr_lock.
591 ldlm_mode_t l_granted_mode;
593 * Lock enqueue completion handler.
595 ldlm_completion_callback l_completion_ast;
597 * Lock blocking ast handler.
599 ldlm_blocking_callback l_blocking_ast;
601 * Lock glimpse handler.
603 ldlm_glimpse_callback l_glimpse_ast;
604 ldlm_weigh_callback l_weigh_ast;
609 struct obd_export *l_export;
611 * Lock connection export.
613 struct obd_export *l_conn_export;
616 * Remote lock handle.
618 struct lustre_handle l_remote_handle;
620 ldlm_policy_data_t l_policy_data;
623 * Protected by lr_lock. Various counters: readers, writers, etc.
629 * Set for locks that were removed from class hash table and will be
630 * destroyed when last reference to them is released. Set by
631 * ldlm_lock_destroy_internal().
633 * Protected by lock and resource locks.
638 * If the lock is granted, a process sleeps on this waitq to learn when
639 * it's no longer in use. If the lock is not granted, a process sleeps
640 * on this waitq to learn when it becomes granted.
645 * Seconds. it will be updated if there is any activity related to
646 * the lock, e.g. enqueue the lock or send block AST.
648 cfs_time_t l_last_activity;
651 * Jiffies. Should be converted to time if needed.
653 cfs_time_t l_last_used;
655 struct ldlm_extent l_req_extent;
658 * Client-side-only members.
662 * Temporary storage for an LVB received during an enqueue operation.
668 spinlock_t l_extents_list_lock;
669 struct list_head l_extents_list;
671 struct list_head l_cache_locks_list;
674 * Server-side-only members.
677 /** connection cookie for the client originated the operation. */
678 __u64 l_client_cookie;
681 * Protected by elt_lock. Callbacks pending.
683 struct list_head l_pending_chain;
685 cfs_time_t l_callback_timeout;
688 * Pid which created this lock.
693 * For ldlm_add_ast_work_item().
695 struct list_head l_bl_ast;
697 * For ldlm_add_ast_work_item().
699 struct list_head l_cp_ast;
701 * For ldlm_add_ast_work_item().
703 struct list_head l_rk_ast;
705 struct ldlm_lock *l_blocking_lock;
709 * Protected by lr_lock, linkages to "skip lists".
711 struct list_head l_sl_mode;
712 struct list_head l_sl_policy;
713 struct lu_ref l_reference;
714 #if LUSTRE_TRACKS_LOCK_EXP_REFS
715 /* Debugging stuff for bug 20498, for tracking export
717 /** number of export references taken */
719 /** link all locks referencing one export */
720 struct list_head l_exp_refs_link;
721 /** referenced export object */
722 struct obd_export *l_exp_refs_target;
726 struct ldlm_resource {
727 struct ldlm_namespace *lr_namespace;
729 /* protected by ns_hash_lock */
730 struct list_head lr_hash;
731 struct ldlm_resource *lr_parent; /* 0 for a root resource */
732 struct list_head lr_children; /* list head for child resources */
733 struct list_head lr_childof; /* part of ns_root_list if root res,
734 * part of lr_children if child */
737 /* protected by lr_lock */
738 struct list_head lr_granted;
739 struct list_head lr_converting;
740 struct list_head lr_waiting;
741 ldlm_mode_t lr_most_restr;
742 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
743 struct ldlm_res_id lr_name;
744 atomic_t lr_refcount;
746 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
748 /* Server-side-only lock value block elements */
749 struct semaphore lr_lvb_sem;
753 /* when the resource was considered as contended */
754 cfs_time_t lr_contention_time;
756 * List of references to this resource. For debugging.
758 struct lu_ref lr_reference;
761 struct ldlm_ast_work {
762 struct ldlm_lock *w_lock;
764 struct ldlm_lock_desc w_desc;
765 struct list_head w_list;
771 /* ldlm_enqueue parameters common */
772 struct ldlm_enqueue_info {
773 __u32 ei_type; /* Type of the lock being enqueued. */
774 __u32 ei_mode; /* Mode of the lock being enqueued. */
775 void *ei_cb_bl; /* blocking lock callback */
776 void *ei_cb_cp; /* lock completion callback */
777 void *ei_cb_gl; /* lock glimpse callback */
778 void *ei_cb_wg; /* lock weigh callback */
779 void *ei_cbdata; /* Data to be passed into callbacks. */
780 short ei_async:1; /* async request */
783 extern struct obd_ops ldlm_obd_ops;
785 extern char *ldlm_lockname[];
786 extern char *ldlm_typename[];
787 extern char *ldlm_it2str(int it);
789 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
792 if (((level) & D_CANTMASK) != 0 || \
793 ((libcfs_debug & (level)) != 0 && \
794 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
795 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
796 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
798 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
803 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
804 struct libcfs_debug_msg_data *data, const char *fmt,
806 __attribute__ ((format (printf, 4, 5)));
808 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
809 static cfs_debug_limit_state_t _ldlm_cdls; \
810 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
811 __FILE__, __FUNCTION__, __LINE__, \
815 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
816 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
818 #define LDLM_DEBUG(lock, fmt, a...) do { \
819 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
820 __FILE__, __FUNCTION__, __LINE__, \
823 #else /* !LIBCFS_DEBUG */
824 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
825 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
826 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
830 #define LDLM_DEBUG_NOLOCK(format, a...) \
831 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
833 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
834 int first_enq, ldlm_error_t *err,
835 struct list_head *work_list);
841 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
842 #define LDLM_ITER_STOP 2 /* stop iterating */
844 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
845 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
847 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
849 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
851 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
852 ldlm_res_iterator_t iter, void *closure);
854 int ldlm_replay_locks(struct obd_import *imp);
855 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
856 ldlm_iterator_t iter, void *data);
859 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
862 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
866 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
867 void *data, int flag);
868 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
869 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
870 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
871 ldlm_blocking_callback, ldlm_glimpse_callback);
872 int ldlm_handle_convert(struct ptlrpc_request *req);
873 int ldlm_handle_cancel(struct ptlrpc_request *req);
874 int ldlm_request_cancel(struct ptlrpc_request *req,
875 const struct ldlm_request *dlm_req, int first);
876 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
877 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
878 void ldlm_revoke_export_locks(struct obd_export *exp);
879 int ldlm_get_ref(void);
880 void ldlm_put_ref(void);
881 int ldlm_init_export(struct obd_export *exp);
882 void ldlm_destroy_export(struct obd_export *exp);
885 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
886 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
887 void ldlm_lock2handle(const struct ldlm_lock *lock,
888 struct lustre_handle *lockh);
889 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
890 void ldlm_cancel_callback(struct ldlm_lock *);
891 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
893 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
895 return __ldlm_handle2lock(h, 0);
898 #define LDLM_LOCK_REF_DEL(lock) \
899 lu_ref_del(&lock->l_reference, "handle", cfs_current())
901 static inline struct ldlm_lock *
902 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
904 struct ldlm_lock *lock;
906 lock = __ldlm_handle2lock(h, flags);
908 LDLM_LOCK_REF_DEL(lock);
912 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
913 struct ptlrpc_request *r, int increase)
915 if (res->lr_namespace->ns_lvbo &&
916 res->lr_namespace->ns_lvbo->lvbo_update) {
917 return res->lr_namespace->ns_lvbo->lvbo_update(res, r,
923 int ldlm_error2errno(ldlm_error_t error);
924 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
925 * confuses user-space. */
926 #if LUSTRE_TRACKS_LOCK_EXP_REFS
927 void ldlm_dump_export_locks(struct obd_export *exp);
931 * Release a temporary lock reference obtained by ldlm_handle2lock() or
932 * __ldlm_handle2lock().
934 #define LDLM_LOCK_PUT(lock) \
936 LDLM_LOCK_REF_DEL(lock); \
937 /*LDLM_DEBUG((lock), "put");*/ \
938 ldlm_lock_put(lock); \
942 * Release a lock reference obtained by some other means (see
945 #define LDLM_LOCK_RELEASE(lock) \
947 /*LDLM_DEBUG((lock), "put");*/ \
948 ldlm_lock_put(lock); \
951 #define LDLM_LOCK_GET(lock) \
953 ldlm_lock_get(lock); \
954 /*LDLM_DEBUG((lock), "get");*/ \
958 #define ldlm_lock_list_put(head, member, count) \
960 struct ldlm_lock *_lock, *_next; \
962 list_for_each_entry_safe(_lock, _next, head, member) { \
965 list_del_init(&_lock->member); \
966 LDLM_LOCK_RELEASE(_lock); \
971 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
972 void ldlm_lock_put(struct ldlm_lock *lock);
973 void ldlm_lock_destroy(struct ldlm_lock *lock);
974 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
975 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
976 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
977 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
978 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
979 void ldlm_lock_allow_match(struct ldlm_lock *lock);
980 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
981 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
982 const struct ldlm_res_id *, ldlm_type_t type,
983 ldlm_policy_data_t *, ldlm_mode_t mode,
984 struct lustre_handle *, int unref);
985 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
987 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
988 void ldlm_lock_cancel(struct ldlm_lock *lock);
989 void ldlm_reprocess_all(struct ldlm_resource *res);
990 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
991 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
992 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
993 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
996 struct ldlm_namespace *
997 ldlm_namespace_new(struct obd_device *obd, char *name,
998 ldlm_side_t client, ldlm_appetite_t apt);
999 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1000 void ldlm_namespace_free(struct ldlm_namespace *ns,
1001 struct obd_import *imp, int force);
1002 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1003 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1004 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1005 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1006 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
1007 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
1008 void ldlm_namespace_get(struct ldlm_namespace *ns);
1009 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
1010 int ldlm_proc_setup(void);
1012 void ldlm_proc_cleanup(void);
1014 static inline void ldlm_proc_cleanup(void) {}
1017 /* resource.c - internal */
1018 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1019 struct ldlm_resource *parent,
1020 const struct ldlm_res_id *,
1021 ldlm_type_t type, int create);
1022 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1023 int ldlm_resource_putref(struct ldlm_resource *res);
1024 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
1025 struct ldlm_lock *lock);
1026 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1027 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1028 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1029 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1030 void ldlm_resource_dump(int level, struct ldlm_resource *);
1031 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1032 const struct ldlm_res_id *);
1034 #define LDLM_RESOURCE_ADDREF(res) do { \
1035 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1038 #define LDLM_RESOURCE_DELREF(res) do { \
1039 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1042 struct ldlm_callback_suite {
1043 ldlm_completion_callback lcs_completion;
1044 ldlm_blocking_callback lcs_blocking;
1045 ldlm_glimpse_callback lcs_glimpse;
1046 ldlm_weigh_callback lcs_weigh;
1049 /* ldlm_request.c */
1050 int ldlm_expired_completion_wait(void *data);
1051 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1052 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1053 void *data, int flag);
1054 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1055 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1056 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1057 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1058 struct ldlm_enqueue_info *einfo,
1059 const struct ldlm_res_id *res_id,
1060 ldlm_policy_data_t *policy, int *flags,
1061 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1063 int ldlm_prep_enqueue_req(struct obd_export *exp,
1064 struct ptlrpc_request *req,
1065 struct list_head *cancels,
1067 int ldlm_prep_elc_req(struct obd_export *exp,
1068 struct ptlrpc_request *req,
1069 int version, int opc, int canceloff,
1070 struct list_head *cancels, int count);
1071 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1072 const struct ldlm_request *dlm_req,
1073 const struct ldlm_callback_suite *cbs);
1074 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1075 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1076 int *flags, void *lvb, __u32 lvb_len,
1077 struct lustre_handle *lockh, int rc);
1078 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1079 const struct ldlm_res_id *res_id,
1080 ldlm_type_t type, ldlm_policy_data_t *policy,
1081 ldlm_mode_t mode, int *flags,
1082 ldlm_blocking_callback blocking,
1083 ldlm_completion_callback completion,
1084 ldlm_glimpse_callback glimpse,
1085 void *data, __u32 lvb_len,
1086 const __u64 *client_cookie,
1087 struct lustre_handle *lockh);
1088 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1089 void *data, __u32 data_len);
1090 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1091 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1092 int ldlm_handle_convert0(struct ptlrpc_request *req,
1093 const struct ldlm_request *dlm_req);
1094 int ldlm_cli_cancel(struct lustre_handle *lockh);
1095 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1096 int flags, void *opaque);
1097 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1098 const struct ldlm_res_id *res_id,
1099 ldlm_policy_data_t *policy,
1100 ldlm_mode_t mode, int flags, void *opaque);
1101 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1102 int count, int flags);
1103 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1104 struct list_head *cancels,
1105 ldlm_policy_data_t *policy,
1106 ldlm_mode_t mode, int lock_flags,
1107 int cancel_flags, void *opaque);
1108 int ldlm_cli_cancel_list(struct list_head *head, int count,
1109 struct ptlrpc_request *req, int flags);
1112 /* This has to be here because recursive inclusion sucks. */
1113 int intent_disposition(struct ldlm_reply *rep, int flag);
1114 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1117 /* ioctls for trying requests */
1118 #define IOC_LDLM_TYPE 'f'
1119 #define IOC_LDLM_MIN_NR 40
1121 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1122 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1123 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1124 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1125 #define IOC_LDLM_MAX_NR 43
1128 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1129 * than one lock_res is dead-lock safe.
1131 enum lock_res_type {
1136 static inline void lock_res(struct ldlm_resource *res)
1138 spin_lock(&res->lr_lock);
1141 static inline void lock_res_nested(struct ldlm_resource *res,
1142 enum lock_res_type mode)
1144 spin_lock_nested(&res->lr_lock, mode);
1148 static inline void unlock_res(struct ldlm_resource *res)
1150 spin_unlock(&res->lr_lock);
1153 static inline void check_res_locked(struct ldlm_resource *res)
1155 LASSERT_SPIN_LOCKED(&res->lr_lock);
1158 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1159 void unlock_res_and_lock(struct ldlm_lock *lock);
1162 void ldlm_pools_recalc(ldlm_side_t client);
1163 int ldlm_pools_init(void);
1164 void ldlm_pools_fini(void);
1166 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1167 int idx, ldlm_side_t client);
1168 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1169 unsigned int gfp_mask);
1170 void ldlm_pool_fini(struct ldlm_pool *pl);
1171 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1172 int ldlm_pool_recalc(struct ldlm_pool *pl);
1173 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1174 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1175 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1176 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1177 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1178 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1179 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1180 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1181 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);