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 <lustre_export.h> /* for obd_export, for LDLM_DEBUG */
55 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
61 #define OBD_LDLM_DEVICENAME "ldlm"
63 #ifdef HAVE_BGL_SUPPORT
64 /* 1.5 times the maximum 128 tasks available in VN mode */
65 #define LDLM_DEFAULT_LRU_SIZE 196
67 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
69 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
70 #define LDLM_CTIME_AGE_LIMIT (10)
75 ELDLM_LOCK_CHANGED = 300,
76 ELDLM_LOCK_ABORTED = 301,
77 ELDLM_LOCK_REPLACED = 302,
78 ELDLM_NO_LOCK_DATA = 303,
80 ELDLM_NAMESPACE_EXISTS = 400,
81 ELDLM_BAD_NAMESPACE = 401
85 LDLM_NAMESPACE_SERVER = 1 << 0,
86 LDLM_NAMESPACE_CLIENT = 1 << 1
89 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
91 /* If the server returns one of these flags, then the lock was put on that list.
92 * If the client sends one of these flags (during recovery ONLY!), it wants the
93 * lock added to the specified list, no questions asked. -p */
94 #define LDLM_FL_BLOCK_GRANTED 0x000002
95 #define LDLM_FL_BLOCK_CONV 0x000004
96 #define LDLM_FL_BLOCK_WAIT 0x000008
98 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
99 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
100 * queued for sending. */
101 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
102 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
104 /* Lock is being replayed. This could probably be implied by the fact that one
105 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
106 #define LDLM_FL_REPLAY 0x000100
108 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
109 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
111 /* don't run the cancel callback under ldlm_cli_cancel_unused */
112 #define LDLM_FL_FAILED 0x000800
114 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
115 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
116 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
117 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
118 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
120 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
123 /* file & record locking */
124 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
125 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
127 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
128 * the LVB filling happens _after_ the lock has been granted, so another thread
129 * can match`t before the LVB has been updated. As a dirty hack, we set
130 * LDLM_FL_LVB_READY only after we've done the LVB poop.
131 * this is only needed on lov/osc now, where lvb is actually used and callers
132 * must set it in input flags.
134 * The proper fix is to do the granting inside of the completion AST, which can
135 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
136 * pretty high-risk, though, and would need a lot more testing. */
138 #define LDLM_FL_LVB_READY 0x100000
140 /* A lock contributes to the kms calculation until it has finished the part
141 * of it's cancelation that performs write back on its dirty pages. It
142 * can remain on the granted list during this whole time. Threads racing
143 * to update the kms after performing their writeback need to know to
144 * exclude each others locks from the calculation as they walk the granted
146 #define LDLM_FL_KMS_IGNORE 0x200000
148 /* Immediatelly cancel such locks when they block some other locks. Send
149 * cancel notification to original lock holder, but expect no reply. This is
150 * for clients (like liblustre) that cannot be expected to reliably response
151 * to blocking ast. */
152 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
154 /* Flags flags inherited from parent lock when doing intents. */
155 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
157 /* completion ast to be executed */
158 #define LDLM_FL_CP_REQD 0x1000000
160 /* cleanup_resource has already handled the lock */
161 #define LDLM_FL_CLEANED 0x2000000
163 /* optimization hint: LDLM can run blocking callback from current context
164 * w/o involving separate thread. in order to decrease cs rate */
165 #define LDLM_FL_ATOMIC_CB 0x4000000
167 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
168 #define LDLM_FL_ASYNC 0x8000000
170 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
171 * such that server send blocking ast for conflict locks to this client for
172 * the 1st operation, whereas the 2nd operation has canceled this lock and
173 * is waiting for rpc_lock which is taken by the 1st operation.
174 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
175 * ELC code to cancel it.
176 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
177 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
178 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
179 * to send a separate CANCEL rpc. */
180 #define LDLM_FL_BL_AST 0x10000000
181 #define LDLM_FL_BL_DONE 0x20000000
183 /* measure lock contention and return -EUSERS if locking contention is high */
184 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
186 /* These are flags that are mapped into the flags and ASTs of blocking locks */
187 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
189 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
190 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
193 * --------------------------------------------------------------------------
194 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
195 * 0x80000000 will not be sent over the wire.
196 * --------------------------------------------------------------------------
199 /* Used for marking lock as an target for -EINTR while cp_ast sleep
200 * emulation + race with upcoming bl_ast. */
201 #define LDLM_FL_FAIL_LOC 0x100000000ULL
203 /* The blocking callback is overloaded to perform two functions. These flags
204 * indicate which operation should be performed. */
205 #define LDLM_CB_BLOCKING 1
206 #define LDLM_CB_CANCELING 2
208 /* compatibility matrix */
209 #define LCK_COMPAT_EX LCK_NL
210 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
211 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
212 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
213 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
214 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
215 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
216 #define LCK_COMPAT_COS (LCK_COS)
218 extern ldlm_mode_t lck_compat_array[];
220 static inline void lockmode_verify(ldlm_mode_t mode)
222 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
225 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
227 return (lck_compat_array[exist_mode] & new_mode);
232 * cluster name spaces
236 #define DLM_OST_NAMESPACE 1
237 #define DLM_MDS_NAMESPACE 2
240 - do we just separate this by security domains and use a prefix for
241 multiple namespaces in the same domain?
251 * waiting_locks_spinlock
266 struct ldlm_resource;
267 struct ldlm_namespace;
269 struct ldlm_pool_ops {
270 int (*po_recalc)(struct ldlm_pool *pl);
271 int (*po_shrink)(struct ldlm_pool *pl, int nr,
272 unsigned int gfp_mask);
273 int (*po_setup)(struct ldlm_pool *pl, int limit);
277 * One second for pools thread check interval. Each pool has own period.
279 #define LDLM_POOLS_THREAD_PERIOD (1)
282 * 5% margin for modest pools. See ldlm_pool.c for details.
284 #define LDLM_POOLS_MODEST_MARGIN (5)
287 * Default recalc period for server side pools in sec.
289 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
292 * Default recalc period for client side pools in sec.
294 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
298 * Pool proc directory.
300 cfs_proc_dir_entry_t *pl_proc_dir;
302 * Pool name, should be long enough to contain compound proc entry name.
306 * Lock for protecting slv/clv updates.
310 * Number of allowed locks in in pool, both, client and server side.
314 * Number of granted locks in
320 atomic_t pl_grant_rate;
324 atomic_t pl_cancel_rate;
326 * Grant speed (GR-CR) per T.
328 atomic_t pl_grant_speed;
330 * Server lock volume. Protected by pl_lock.
332 __u64 pl_server_lock_volume;
334 * Current biggest client lock volume. Protected by pl_lock.
336 __u64 pl_client_lock_volume;
338 * Lock volume factor. SLV on client is calculated as following:
339 * server_slv * lock_volume_factor.
341 atomic_t pl_lock_volume_factor;
343 * Time when last slv from server was obtained.
345 time_t pl_recalc_time;
347 * Recalc period for pool.
349 time_t pl_recalc_period;
351 * Recalc and shrink ops.
353 struct ldlm_pool_ops *pl_ops;
355 * Number of planned locks for next period.
361 struct lprocfs_stats *pl_stats;
364 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
365 void *req_cookie, ldlm_mode_t mode, int flags,
368 struct ldlm_valblock_ops {
369 int (*lvbo_init)(struct ldlm_resource *res);
370 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
371 int buf_idx, int increase);
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 @ns_contented_locks found, the resource considered
463 unsigned ns_contended_locks;
466 * The resource remembers contended state during @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 */
549 * Must be first in the structure.
551 struct portals_handle l_handle;
553 * Lock reference count.
557 * Internal spinlock protects l_resource. we should hold this lock
558 * first before grabbing res_lock.
562 * ldlm_lock_change_resource() can change this.
564 struct ldlm_resource *l_resource;
566 * Protected by ns_hash_lock. List item for client side lru list.
568 struct list_head l_lru;
570 * Protected by lr_lock, linkage to resource's lock queues.
572 struct list_head l_res_link;
574 * Tree node for ldlm_extent.
576 struct ldlm_interval *l_tree_node;
578 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
581 struct hlist_node l_exp_hash;
583 * Protected by lr_lock. Requested mode.
585 ldlm_mode_t l_req_mode;
587 * Granted mode, also protected by lr_lock.
589 ldlm_mode_t l_granted_mode;
591 * Lock enqueue completion handler.
593 ldlm_completion_callback l_completion_ast;
595 * Lock blocking ast handler.
597 ldlm_blocking_callback l_blocking_ast;
599 * Lock glimpse handler.
601 ldlm_glimpse_callback l_glimpse_ast;
602 ldlm_weigh_callback l_weigh_ast;
607 struct obd_export *l_export;
609 * Lock connection export.
611 struct obd_export *l_conn_export;
614 * Remote lock handle.
616 struct lustre_handle l_remote_handle;
618 ldlm_policy_data_t l_policy_data;
621 * Protected by lr_lock. Various counters: readers, writers, etc.
627 * Set for locks that were removed from class hash table and will be
628 * destroyed when last reference to them is released. Set by
629 * ldlm_lock_destroy_internal().
631 * Protected by lock and resource locks.
636 * If the lock is granted, a process sleeps on this waitq to learn when
637 * it's no longer in use. If the lock is not granted, a process sleeps
638 * on this waitq to learn when it becomes granted.
643 * Seconds. it will be updated if there is any activity related to
644 * the lock, e.g. enqueue the lock or send block AST.
646 cfs_time_t l_last_activity;
649 * Jiffies. Should be converted to time if needed.
651 cfs_time_t l_last_used;
653 struct ldlm_extent l_req_extent;
656 * Client-side-only members.
660 * Temporary storage for an LVB received during an enqueue operation.
667 spinlock_t l_extents_list_lock;
668 struct list_head l_extents_list;
670 struct list_head l_cache_locks_list;
673 * Server-side-only members.
676 /** connection cookie for the client originated the operation. */
677 __u64 l_client_cookie;
680 * Protected by elt_lock. Callbacks pending.
682 struct list_head l_pending_chain;
684 cfs_time_t l_callback_timeout;
687 * Pid which created this lock.
692 * For ldlm_add_ast_work_item().
694 struct list_head l_bl_ast;
696 * For ldlm_add_ast_work_item().
698 struct list_head l_cp_ast;
700 * For ldlm_add_ast_work_item().
702 struct list_head l_rk_ast;
704 struct ldlm_lock *l_blocking_lock;
708 * Protected by lr_lock, linkages to "skip lists".
710 struct list_head l_sl_mode;
711 struct list_head l_sl_policy;
712 struct lu_ref l_reference;
715 struct ldlm_resource {
716 struct ldlm_namespace *lr_namespace;
718 /* protected by ns_hash_lock */
719 struct list_head lr_hash;
720 struct ldlm_resource *lr_parent; /* 0 for a root resource */
721 struct list_head lr_children; /* list head for child resources */
722 struct list_head lr_childof; /* part of ns_root_list if root res,
723 * part of lr_children if child */
726 /* protected by lr_lock */
727 struct list_head lr_granted;
728 struct list_head lr_converting;
729 struct list_head lr_waiting;
730 ldlm_mode_t lr_most_restr;
731 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
732 struct ldlm_res_id lr_name;
733 atomic_t lr_refcount;
735 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
737 /* Server-side-only lock value block elements */
738 struct semaphore lr_lvb_sem;
742 /* when the resource was considered as contended */
743 cfs_time_t lr_contention_time;
745 * List of references to this resource. For debugging.
747 struct lu_ref lr_reference;
750 struct ldlm_ast_work {
751 struct ldlm_lock *w_lock;
753 struct ldlm_lock_desc w_desc;
754 struct list_head w_list;
760 /* ldlm_enqueue parameters common */
761 struct ldlm_enqueue_info {
762 __u32 ei_type; /* Type of the lock being enqueued. */
763 __u32 ei_mode; /* Mode of the lock being enqueued. */
764 void *ei_cb_bl; /* blocking lock callback */
765 void *ei_cb_cp; /* lock completion callback */
766 void *ei_cb_gl; /* lock glimpse callback */
767 void *ei_cb_wg; /* lock weigh callback */
768 void *ei_cbdata; /* Data to be passed into callbacks. */
769 short ei_async:1; /* async request */
772 extern struct obd_ops ldlm_obd_ops;
774 extern char *ldlm_lockname[];
775 extern char *ldlm_typename[];
776 extern char *ldlm_it2str(int it);
778 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
781 if (((level) & D_CANTMASK) != 0 || \
782 ((libcfs_debug & (level)) != 0 && \
783 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
784 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
785 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
787 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
792 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
793 struct libcfs_debug_msg_data *data, const char *fmt,
795 __attribute__ ((format (printf, 4, 5)));
797 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
798 static cfs_debug_limit_state_t _ldlm_cdls; \
799 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
800 __FILE__, __FUNCTION__, __LINE__, \
804 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
805 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
807 #define LDLM_DEBUG(lock, fmt, a...) do { \
808 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
809 __FILE__, __FUNCTION__, __LINE__, \
812 #else /* !LIBCFS_DEBUG */
813 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
814 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
815 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
819 #define LDLM_DEBUG_NOLOCK(format, a...) \
820 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
822 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
823 int first_enq, ldlm_error_t *err,
824 struct list_head *work_list);
830 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
831 #define LDLM_ITER_STOP 2 /* stop iterating */
833 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
834 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
836 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
838 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
840 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
841 ldlm_res_iterator_t iter, void *closure);
843 int ldlm_replay_locks(struct obd_import *imp);
844 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
845 ldlm_iterator_t iter, void *data);
848 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
851 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
855 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
856 void *data, int flag);
857 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
858 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
859 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
860 ldlm_blocking_callback, ldlm_glimpse_callback);
861 int ldlm_handle_convert(struct ptlrpc_request *req);
862 int ldlm_handle_cancel(struct ptlrpc_request *req);
863 int ldlm_request_cancel(struct ptlrpc_request *req,
864 const struct ldlm_request *dlm_req, int first);
865 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
866 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
867 void ldlm_revoke_export_locks(struct obd_export *exp);
868 int ldlm_get_ref(void);
869 void ldlm_put_ref(void);
870 int ldlm_init_export(struct obd_export *exp);
871 void ldlm_destroy_export(struct obd_export *exp);
874 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
875 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
876 void ldlm_lock2handle(const struct ldlm_lock *lock,
877 struct lustre_handle *lockh);
878 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
879 void ldlm_cancel_callback(struct ldlm_lock *);
880 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
882 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
884 return __ldlm_handle2lock(h, 0);
887 #define LDLM_LOCK_REF_DEL(lock) \
888 lu_ref_del(&lock->l_reference, "handle", cfs_current())
890 static inline struct ldlm_lock *
891 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
893 struct ldlm_lock *lock;
895 lock = __ldlm_handle2lock(h, flags);
897 LDLM_LOCK_REF_DEL(lock);
901 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
902 struct lustre_msg *m, int buf_idx,
905 if (res->lr_namespace->ns_lvbo &&
906 res->lr_namespace->ns_lvbo->lvbo_update) {
907 return res->lr_namespace->ns_lvbo->lvbo_update(res, m, buf_idx,
913 int ldlm_error2errno(ldlm_error_t error);
914 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
915 * confuses user-space. */
918 * Release a temporary lock reference obtained by ldlm_handle2lock() or
919 * __ldlm_handle2lock().
921 #define LDLM_LOCK_PUT(lock) \
923 LDLM_LOCK_REF_DEL(lock); \
924 /*LDLM_DEBUG((lock), "put");*/ \
925 ldlm_lock_put(lock); \
929 * Release a lock reference obtained by some other means (see
932 #define LDLM_LOCK_RELEASE(lock) \
934 /*LDLM_DEBUG((lock), "put");*/ \
935 ldlm_lock_put(lock); \
938 #define LDLM_LOCK_GET(lock) \
940 ldlm_lock_get(lock); \
941 /*LDLM_DEBUG((lock), "get");*/ \
945 #define ldlm_lock_list_put(head, member, count) \
947 struct ldlm_lock *_lock, *_next; \
949 list_for_each_entry_safe(_lock, _next, head, member) { \
952 list_del_init(&_lock->member); \
953 LDLM_LOCK_RELEASE(_lock); \
958 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
959 void ldlm_lock_put(struct ldlm_lock *lock);
960 void ldlm_lock_destroy(struct ldlm_lock *lock);
961 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
962 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
963 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
964 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
965 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
966 void ldlm_lock_allow_match(struct ldlm_lock *lock);
967 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
968 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
969 const struct ldlm_res_id *, ldlm_type_t type,
970 ldlm_policy_data_t *, ldlm_mode_t mode,
971 struct lustre_handle *, int unref);
972 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
974 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
975 void ldlm_lock_cancel(struct ldlm_lock *lock);
976 void ldlm_cancel_locks_for_export(struct obd_export *export);
977 void ldlm_reprocess_all(struct ldlm_resource *res);
978 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
979 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
980 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
981 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
984 struct ldlm_namespace *
985 ldlm_namespace_new(struct obd_device *obd, char *name,
986 ldlm_side_t client, ldlm_appetite_t apt);
987 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
988 void ldlm_namespace_free(struct ldlm_namespace *ns,
989 struct obd_import *imp, int force);
990 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
991 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
992 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
993 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
994 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
995 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
996 void ldlm_namespace_get(struct ldlm_namespace *ns);
997 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
998 int ldlm_proc_setup(void);
1000 void ldlm_proc_cleanup(void);
1002 static inline void ldlm_proc_cleanup(void) {}
1005 /* resource.c - internal */
1006 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1007 struct ldlm_resource *parent,
1008 const struct ldlm_res_id *,
1009 ldlm_type_t type, int create);
1010 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1011 int ldlm_resource_putref(struct ldlm_resource *res);
1012 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
1013 struct ldlm_lock *lock);
1014 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1015 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1016 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1017 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1018 void ldlm_resource_dump(int level, struct ldlm_resource *);
1019 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1020 const struct ldlm_res_id *);
1022 #define LDLM_RESOURCE_ADDREF(res) do { \
1023 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1026 #define LDLM_RESOURCE_DELREF(res) do { \
1027 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1030 struct ldlm_callback_suite {
1031 ldlm_completion_callback lcs_completion;
1032 ldlm_blocking_callback lcs_blocking;
1033 ldlm_glimpse_callback lcs_glimpse;
1034 ldlm_weigh_callback lcs_weigh;
1037 /* ldlm_request.c */
1038 int ldlm_expired_completion_wait(void *data);
1039 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1040 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1041 void *data, int flag);
1042 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1043 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1044 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1045 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1046 struct ldlm_enqueue_info *einfo,
1047 const struct ldlm_res_id *res_id,
1048 ldlm_policy_data_t *policy, int *flags,
1049 void *lvb, __u32 lvb_len, void *lvb_swabber,
1050 struct lustre_handle *lockh, int async);
1051 int ldlm_prep_enqueue_req(struct obd_export *exp,
1052 struct ptlrpc_request *req,
1053 struct list_head *cancels,
1055 int ldlm_prep_elc_req(struct obd_export *exp,
1056 struct ptlrpc_request *req,
1057 int version, int opc, int canceloff,
1058 struct list_head *cancels, int count);
1059 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1060 const struct ldlm_request *dlm_req,
1061 const struct ldlm_callback_suite *cbs);
1062 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1063 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1064 int *flags, void *lvb, __u32 lvb_len,
1065 void *lvb_swabber, struct lustre_handle *lockh,
1067 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1068 const struct ldlm_res_id *res_id,
1069 ldlm_type_t type, ldlm_policy_data_t *policy,
1070 ldlm_mode_t mode, int *flags,
1071 ldlm_blocking_callback blocking,
1072 ldlm_completion_callback completion,
1073 ldlm_glimpse_callback glimpse,
1074 void *data, __u32 lvb_len, void *lvb_swabber,
1075 const __u64 *client_cookie,
1076 struct lustre_handle *lockh);
1077 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1078 void *data, __u32 data_len);
1079 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1080 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1081 int ldlm_handle_convert0(struct ptlrpc_request *req,
1082 const struct ldlm_request *dlm_req);
1083 int ldlm_cli_cancel(struct lustre_handle *lockh);
1084 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1085 int flags, void *opaque);
1086 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1087 const struct ldlm_res_id *res_id,
1088 ldlm_policy_data_t *policy,
1089 ldlm_mode_t mode, int flags, void *opaque);
1090 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1091 int count, int flags);
1092 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1093 struct list_head *cancels,
1094 ldlm_policy_data_t *policy,
1095 ldlm_mode_t mode, int lock_flags,
1096 int cancel_flags, void *opaque);
1097 int ldlm_cli_cancel_list(struct list_head *head, int count,
1098 struct ptlrpc_request *req, int flags);
1101 /* This has to be here because recursive inclusion sucks. */
1102 int intent_disposition(struct ldlm_reply *rep, int flag);
1103 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1106 /* ioctls for trying requests */
1107 #define IOC_LDLM_TYPE 'f'
1108 #define IOC_LDLM_MIN_NR 40
1110 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1111 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1112 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1113 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1114 #define IOC_LDLM_MAX_NR 43
1117 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1118 * than one lock_res is dead-lock safe.
1120 enum lock_res_type {
1125 static inline void lock_res(struct ldlm_resource *res)
1127 spin_lock(&res->lr_lock);
1130 static inline void lock_res_nested(struct ldlm_resource *res,
1131 enum lock_res_type mode)
1133 spin_lock_nested(&res->lr_lock, mode);
1137 static inline void unlock_res(struct ldlm_resource *res)
1139 spin_unlock(&res->lr_lock);
1142 static inline void check_res_locked(struct ldlm_resource *res)
1144 LASSERT_SPIN_LOCKED(&res->lr_lock);
1147 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1148 void unlock_res_and_lock(struct ldlm_lock *lock);
1151 void ldlm_pools_recalc(ldlm_side_t client);
1152 int ldlm_pools_init(void);
1153 void ldlm_pools_fini(void);
1155 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1156 int idx, ldlm_side_t client);
1157 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1158 unsigned int gfp_mask);
1159 void ldlm_pool_fini(struct ldlm_pool *pl);
1160 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1161 int ldlm_pool_recalc(struct ldlm_pool *pl);
1162 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1163 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1164 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1165 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1166 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1167 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1168 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1169 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1170 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);