1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * (visit-tags-table FILE)
3 * vim:expandtab:shiftwidth=8:tabstop=8:
6 #ifndef _LUSTRE_DLM_H__
7 #define _LUSTRE_DLM_H__
10 #include <linux/lustre_dlm.h>
11 #elif defined(__APPLE__)
12 #include <darwin/lustre_dlm.h>
13 #elif defined(__WINNT__)
14 #include <winnt/lustre_dlm.h>
16 #error Unsupported operating system.
19 #include <lustre_lib.h>
20 #include <lustre_net.h>
21 #include <lustre_import.h>
22 #include <lustre_handles.h>
23 #include <lustre_export.h> /* for obd_export, for LDLM_DEBUG */
28 #define OBD_LDLM_DEVICENAME "ldlm"
30 #define LDLM_DEFAULT_LRU_SIZE (100 * smp_num_cpus)
35 ELDLM_LOCK_CHANGED = 300,
36 ELDLM_LOCK_ABORTED = 301,
37 ELDLM_LOCK_REPLACED = 302,
38 ELDLM_NO_LOCK_DATA = 303,
40 ELDLM_NAMESPACE_EXISTS = 400,
41 ELDLM_BAD_NAMESPACE = 401
44 #define LDLM_NAMESPACE_SERVER 0
45 #define LDLM_NAMESPACE_CLIENT 1
47 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
49 /* If the server returns one of these flags, then the lock was put on that list.
50 * If the client sends one of these flags (during recovery ONLY!), it wants the
51 * lock added to the specified list, no questions asked. -p */
52 #define LDLM_FL_BLOCK_GRANTED 0x000002
53 #define LDLM_FL_BLOCK_CONV 0x000004
54 #define LDLM_FL_BLOCK_WAIT 0x000008
56 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
57 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was sent */
58 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
59 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
61 /* Lock is being replayed. This could probably be implied by the fact that one
62 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
63 #define LDLM_FL_REPLAY 0x000100
65 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
66 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
68 /* don't run the cancel callback under ldlm_cli_cancel_unused */
69 #define LDLM_FL_FAILED 0x000800
71 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
72 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
73 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
74 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
75 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
77 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
80 /* file & record locking */
81 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
82 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
84 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
85 * the LVB filling happens _after_ the lock has been granted, so another thread
86 * can match before the LVB has been updated. As a dirty hack, we set
87 * LDLM_FL_CAN_MATCH only after we've done the LVB poop.
89 * The proper fix is to do the granting inside of the completion AST, which can
90 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
91 * pretty high-risk, though, and would need a lot more testing. */
92 #define LDLM_FL_CAN_MATCH 0x100000
94 /* A lock contributes to the kms calculation until it has finished the part
95 * of it's cancelation that performs write back on its dirty pages. It
96 * can remain on the granted list during this whole time. Threads racing
97 * to update the kms after performing their writeback need to know to
98 * exclude each others locks from the calculation as they walk the granted
100 #define LDLM_FL_KMS_IGNORE 0x200000
102 /* Don't drop lock covering mmapped file in LRU */
103 #define LDLM_FL_NO_LRU 0x400000
105 /* Immediatelly cancel such locks when they block some other locks. Send
106 cancel notification to original lock holder, but expect no reply. */
107 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
109 /* Flags flags inherited from parent lock when doing intents. */
110 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
112 /* These are flags that are mapped into the flags and ASTs of blocking locks */
113 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
114 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
115 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
117 /* completion ast to be executed */
118 #define LDLM_FL_CP_REQD 0x1000000
120 /* cleanup_resource has already handled the lock */
121 #define LDLM_FL_CLEANED 0x2000000
123 /* optimization hint: LDLM can run blocking callback from current context
124 * w/o involving separate thread. in order to decrease cs rate */
125 #define LDLM_FL_ATOMIC_CB 0x4000000
127 /* while this flag is set, the lock can't change resource */
128 #define LDLM_FL_LOCK_PROTECT 0x8000000
129 #define LDLM_FL_LOCK_PROTECT_BIT 27
132 /* The blocking callback is overloaded to perform two functions. These flags
133 * indicate which operation should be performed. */
134 #define LDLM_CB_BLOCKING 1
135 #define LDLM_CB_CANCELING 2
137 /* compatibility matrix */
138 #define LCK_COMPAT_EX LCK_NL
139 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
140 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
141 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
142 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
143 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX)
144 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
146 extern ldlm_mode_t lck_compat_array[];
148 static inline void lockmode_verify(ldlm_mode_t mode)
150 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
153 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
155 return (lck_compat_array[exist] & new);
160 * cluster name spaces
164 #define DLM_OST_NAMESPACE 1
165 #define DLM_MDS_NAMESPACE 2
168 - do we just separate this by security domains and use a prefix for
169 multiple namespaces in the same domain?
179 * waiting_locks_spinlock
193 struct ldlm_resource;
194 struct ldlm_namespace;
196 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
197 void *req_cookie, ldlm_mode_t mode, int flags,
200 struct ldlm_valblock_ops {
201 int (*lvbo_init)(struct ldlm_resource *res);
202 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
203 int buf_idx, int increase);
206 struct ldlm_namespace {
208 __u32 ns_client; /* is this a client-side lock tree? */
209 struct list_head *ns_hash; /* hash table for ns */
210 spinlock_t ns_hash_lock;
211 __u32 ns_refcount; /* count of resources in the hash */
212 struct list_head ns_root_list; /* all root resources in ns */
213 struct list_head ns_list_chain; /* position in global NS list */
215 struct list_head ns_unused_list; /* all root resources in ns */
217 spinlock_t ns_unused_lock;
219 unsigned int ns_max_unused;
220 cfs_time_t ns_next_dump; /* next debug dump, jiffies */
224 ldlm_res_policy ns_policy;
225 struct ldlm_valblock_ops *ns_lvbo;
227 cfs_waitq_t ns_waitq;
232 * Resource hash table
236 #define RES_HASH_BITS 10
237 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
238 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
242 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
243 struct ldlm_lock_desc *new, void *data,
245 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
247 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
250 struct portals_handle l_handle; // must be first in the structure
253 /* ldlm_lock_change_resource() can change this */
254 struct ldlm_resource *l_resource;
256 /* set once, no need to protect it */
257 struct ldlm_lock *l_parent;
259 /* protected by ns_hash_lock */
260 struct list_head l_children;
261 struct list_head l_childof;
263 /* protected by ns_hash_lock. FIXME */
264 struct list_head l_lru;
266 /* protected by lr_lock */
267 struct list_head l_res_link; // position in one of three res lists
269 /* protected by led_lock */
270 struct list_head l_export_chain; // per-export chain of locks
272 /* protected by lr_lock */
273 ldlm_mode_t l_req_mode;
274 ldlm_mode_t l_granted_mode;
276 ldlm_completion_callback l_completion_ast;
277 ldlm_blocking_callback l_blocking_ast;
278 ldlm_glimpse_callback l_glimpse_ast;
280 struct obd_export *l_export;
281 struct obd_export *l_conn_export;
283 /* protected by lr_lock */
286 struct lustre_handle l_remote_handle;
287 ldlm_policy_data_t l_policy_data;
289 /* protected by lr_lock */
294 /* If the lock is granted, a process sleeps on this waitq to learn when
295 * it's no longer in use. If the lock is not granted, a process sleeps
296 * on this waitq to learn when it becomes granted. */
298 struct timeval l_enqueued_time;
300 cfs_time_t l_last_used; /* jiffies */
301 struct ldlm_extent l_req_extent;
303 /* Client-side-only members */
304 __u32 l_lvb_len; /* temporary storage for */
305 void *l_lvb_data; /* an LVB received during */
306 void *l_lvb_swabber; /* an enqueue */
309 /* Server-side-only members */
311 /* protected by elt_lock */
312 struct list_head l_pending_chain; /* callbacks pending */
313 cfs_time_t l_callback_timeout; /* jiffies */
315 __u32 l_pid; /* pid which created this lock */
316 __u32 l_pidb; /* who holds LOCK_PROTECT_BIT */
318 struct list_head l_tmp;
320 /* for ldlm_add_ast_work_item() */
321 struct list_head l_bl_ast;
322 struct list_head l_cp_ast;
323 struct ldlm_lock *l_blocking_lock;
327 struct ldlm_resource {
328 struct ldlm_namespace *lr_namespace;
330 /* protected by ns_hash_lock */
331 struct list_head lr_hash;
332 struct ldlm_resource *lr_parent; /* 0 for a root resource */
333 struct list_head lr_children; /* list head for child resources */
334 struct list_head lr_childof; /* part of ns_root_list if root res,
335 * part of lr_children if child */
338 /* protected by lr_lock */
339 struct list_head lr_granted;
340 struct list_head lr_converting;
341 struct list_head lr_waiting;
342 ldlm_mode_t lr_most_restr;
343 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
344 struct ldlm_res_id lr_name;
345 atomic_t lr_refcount;
347 /* Server-side-only lock value block elements */
348 struct semaphore lr_lvb_sem;
352 /* lr_tmp holds a list head temporarily, during the building of a work
353 * queue. see ldlm_add_ast_work_item and ldlm_run_ast_work */
357 struct ldlm_ast_work {
358 struct ldlm_lock *w_lock;
360 struct ldlm_lock_desc w_desc;
361 struct list_head w_list;
367 extern struct obd_ops ldlm_obd_ops;
369 extern char *ldlm_lockname[];
370 extern char *ldlm_typename[];
371 extern char *ldlm_it2str(int it);
373 void ldlm_lock_debug(cfs_debug_limit_state_t *cdls,
374 __u32 level, struct ldlm_lock *lock,
375 const char *file, const char *func, const int line,
378 #define LDLM_DEBUG(lock, fmt, a...) ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
379 __FILE__, __func__, __LINE__, "### " fmt, ## a)
381 #define LDLM_ERROR(lock, fmt, a...) \
383 static cfs_debug_limit_state_t cdls; \
384 ldlm_lock_debug(&cdls, D_ERROR, lock, \
385 __FILE__, __func__, __LINE__, "### " fmt, ## a); \
388 #define LDLM_DEBUG_NOLOCK(format, a...) \
389 CDEBUG(D_DLMTRACE, "### " format "\n" , ## a)
391 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
392 int first_enq, ldlm_error_t *err,
393 struct list_head *work_list);
399 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
400 #define LDLM_ITER_STOP 2 /* stop iterating */
402 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
403 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
405 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
407 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
409 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
410 ldlm_res_iterator_t iter, void *closure);
412 int ldlm_replay_locks(struct obd_import *imp);
413 void ldlm_resource_iterate(struct ldlm_namespace *, struct ldlm_res_id *,
414 ldlm_iterator_t iter, void *data);
418 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
421 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
425 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
426 void *data, int flag);
427 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
428 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
429 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
430 ldlm_blocking_callback, ldlm_glimpse_callback);
431 int ldlm_handle_convert(struct ptlrpc_request *req);
432 int ldlm_handle_cancel(struct ptlrpc_request *req);
433 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
434 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock);
435 void ldlm_revoke_export_locks(struct obd_export *exp);
436 int ldlm_get_ref(void);
437 void ldlm_put_ref(int force);
440 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
441 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
442 void ldlm_lock2handle(const struct ldlm_lock *lock,
443 struct lustre_handle *lockh);
444 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
445 void ldlm_cancel_callback(struct ldlm_lock *);
446 int ldlm_lock_set_data(struct lustre_handle *, void *data);
447 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
448 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
449 const struct lustre_handle *);
451 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
453 return __ldlm_handle2lock(h, 0);
456 #define LDLM_LOCK_PUT(lock) \
458 /*LDLM_DEBUG((lock), "put");*/ \
459 ldlm_lock_put(lock); \
462 #define LDLM_LOCK_GET(lock) \
464 ldlm_lock_get(lock); \
465 /*LDLM_DEBUG((lock), "get");*/ \
469 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
470 void ldlm_lock_put(struct ldlm_lock *lock);
471 void ldlm_lock_destroy(struct ldlm_lock *lock);
472 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
473 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
474 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
475 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
476 void ldlm_lock_allow_match(struct ldlm_lock *lock);
477 int ldlm_lock_match(struct ldlm_namespace *ns, int flags, struct ldlm_res_id *,
478 ldlm_type_t type, ldlm_policy_data_t *, ldlm_mode_t mode,
479 struct lustre_handle *);
480 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
482 void ldlm_lock_cancel(struct ldlm_lock *lock);
483 void ldlm_cancel_locks_for_export(struct obd_export *export);
484 void ldlm_reprocess_all(struct ldlm_resource *res);
485 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
486 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
487 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
490 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 local);
491 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
492 int ldlm_namespace_free(struct ldlm_namespace *ns, int force);
493 int ldlm_proc_setup(void);
495 void ldlm_proc_cleanup(void);
497 static inline void ldlm_proc_cleanup(void) {}
500 /* resource.c - internal */
501 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
502 struct ldlm_resource *parent,
503 struct ldlm_res_id, ldlm_type_t type,
505 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
506 int ldlm_resource_putref(struct ldlm_resource *res);
507 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
508 struct ldlm_lock *lock);
509 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
510 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
511 void ldlm_dump_all_namespaces(int level);
512 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
513 void ldlm_resource_dump(int level, struct ldlm_resource *);
514 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
517 struct ldlm_callback_suite {
518 ldlm_completion_callback lcs_completion;
519 ldlm_blocking_callback lcs_blocking;
520 ldlm_glimpse_callback lcs_glimpse;
524 int ldlm_expired_completion_wait(void *data);
525 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
526 void *data, int flag);
527 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
528 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
529 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
530 struct ldlm_res_id res_id,
531 ldlm_type_t type, ldlm_policy_data_t *policy,
532 ldlm_mode_t mode, int *flags,
533 ldlm_blocking_callback blocking,
534 ldlm_completion_callback completion,
535 ldlm_glimpse_callback glimpse,
536 void *data, void *lvb, __u32 lvb_len, void *lvb_swabber,
537 struct lustre_handle *lockh, int async);
538 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
539 const struct ldlm_request *dlm_req,
540 const struct ldlm_callback_suite *cbs);
541 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
542 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
543 int *flags, void *lvb, __u32 lvb_len,
544 void *lvb_swabber, struct lustre_handle *lockh,
546 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns, struct ldlm_res_id res_id,
547 ldlm_type_t type, ldlm_policy_data_t *policy,
548 ldlm_mode_t mode, int *flags,
549 ldlm_blocking_callback blocking,
550 ldlm_completion_callback completion,
551 ldlm_glimpse_callback glimpse,
552 void *data, __u32 lvb_len, void *lvb_swabber,
553 struct lustre_handle *lockh);
554 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
555 void *data, __u32 data_len);
556 int ldlm_cli_convert(struct lustre_handle *, int new_mode, int *flags);
557 int ldlm_handle_convert0(struct ptlrpc_request *req,
558 const struct ldlm_request *dlm_req);
559 int ldlm_cli_cancel(struct lustre_handle *lockh);
560 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
561 int flags, void *opaque);
562 int ldlm_cli_join_lru(struct ldlm_namespace *, struct ldlm_res_id *, int join);
565 /* This has to be here because recursive inclusion sucks. */
566 int intent_disposition(struct ldlm_reply *rep, int flag);
567 void intent_set_disposition(struct ldlm_reply *rep, int flag);
570 /* ioctls for trying requests */
571 #define IOC_LDLM_TYPE 'f'
572 #define IOC_LDLM_MIN_NR 40
574 #define IOC_LDLM_TEST _IOWR('f', 40, long)
575 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
576 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
577 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
578 #define IOC_LDLM_MAX_NR 43
580 static inline void lock_res(struct ldlm_resource *res)
582 spin_lock(&res->lr_lock);
585 static inline void unlock_res(struct ldlm_resource *res)
587 spin_unlock(&res->lr_lock);
590 static inline void check_res_locked(struct ldlm_resource *res)
592 LASSERT_SPIN_LOCKED(&res->lr_lock);
595 static inline void lock_bitlock(struct ldlm_lock *lock)
597 bit_spin_lock(LDLM_FL_LOCK_PROTECT_BIT, (void *) &lock->l_flags);
598 LASSERT(lock->l_pidb == 0);
599 lock->l_pidb = current->pid;
602 static inline void unlock_bitlock(struct ldlm_lock *lock)
604 LASSERT(lock->l_pidb == current->pid);
606 bit_spin_unlock(LDLM_FL_LOCK_PROTECT_BIT, (void *) &lock->l_flags);
609 static inline void lock_bitlock(struct ldlm_lock *lock)
611 lock->l_flags |= 1 << LDLM_FL_LOCK_PROTECT_BIT;
614 static inline void unlock_bitlock(struct ldlm_lock *lock)
616 lock->l_flags &= ~(1 << LDLM_FL_LOCK_PROTECT_BIT);
619 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
620 void unlock_res_and_lock(struct ldlm_lock *lock);