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/proc_fs.h>
13 #include <linux/lustre_lib.h>
14 #include <linux/lustre_net.h>
15 #include <linux/lustre_import.h>
16 #include <linux/lustre_handles.h>
17 #include <linux/lustre_export.h> /* for obd_export, for LDLM_DEBUG */
22 #define OBD_LDLM_DEVICENAME "ldlm"
24 #define LDLM_DEFAULT_LRU_SIZE 100
29 ELDLM_LOCK_CHANGED = 300,
30 ELDLM_LOCK_ABORTED = 301,
31 ELDLM_LOCK_REPLACED = 302,
33 ELDLM_NAMESPACE_EXISTS = 400,
34 ELDLM_BAD_NAMESPACE = 401
37 #define LDLM_NAMESPACE_SERVER 0
38 #define LDLM_NAMESPACE_CLIENT 1
40 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
42 /* If the server returns one of these flags, then the lock was put on that list.
43 * If the client sends one of these flags (during recovery ONLY!), it wants the
44 * lock added to the specified list, no questions asked. -p */
45 #define LDLM_FL_BLOCK_GRANTED 0x000002
46 #define LDLM_FL_BLOCK_CONV 0x000004
47 #define LDLM_FL_BLOCK_WAIT 0x000008
49 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
50 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was sent */
51 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
52 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
54 /* Lock is being replayed. This could probably be implied by the fact that one
55 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
56 #define LDLM_FL_REPLAY 0x000100
58 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
59 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
61 /* don't run the cancel callback under ldlm_cli_cancel_unused */
62 #define LDLM_FL_NO_CALLBACK 0x000800
64 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
65 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
66 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
67 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
68 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
70 /* file & record locking */
71 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
72 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
74 /* These are flags that are mapped into the flags and ASTs of blocking locks */
75 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
76 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
77 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
79 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
80 * the LVB filling happens _after_ the lock has been granted, so another thread
81 * can match before the LVB has been updated. As a dirty hack, we set
82 * LDLM_FL_CAN_MATCH only after we've done the LVB poop.
84 * The proper fix is to do the granting inside of the completion AST, which can
85 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
86 * pretty high-risk, though, and would need a lot more testing. */
87 #define LDLM_FL_CAN_MATCH 0x100000
89 /* The blocking callback is overloaded to perform two functions. These flags
90 * indicate which operation should be performed. */
91 #define LDLM_CB_BLOCKING 1
92 #define LDLM_CB_CANCELING 2
94 #define L2B(c) (1 << c)
96 /* compatibility matrix */
97 #define LCK_COMPAT_EX L2B(LCK_NL)
98 #define LCK_COMPAT_PW (LCK_COMPAT_EX | L2B(LCK_CR))
99 #define LCK_COMPAT_PR (LCK_COMPAT_PW | L2B(LCK_PR))
100 #define LCK_COMPAT_CW (LCK_COMPAT_PW | L2B(LCK_CW))
101 #define LCK_COMPAT_CR (LCK_COMPAT_CW | L2B(LCK_PR) | L2B(LCK_PW))
102 #define LCK_COMPAT_NL (LCK_COMPAT_CR | L2B(LCK_EX))
104 static ldlm_mode_t lck_compat_array[] = {
105 [LCK_EX] LCK_COMPAT_EX,
106 [LCK_PW] LCK_COMPAT_PW,
107 [LCK_PR] LCK_COMPAT_PR,
108 [LCK_CW] LCK_COMPAT_CW,
109 [LCK_CR] LCK_COMPAT_CR,
110 [LCK_NL] LCK_COMPAT_NL
113 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
115 LASSERT(exist >= LCK_EX && exist <= LCK_NL);
116 LASSERT(new >= LCK_EX && new <= LCK_NL);
118 return (lck_compat_array[exist] & L2B(new));
123 * cluster name spaces
127 #define DLM_OST_NAMESPACE 1
128 #define DLM_MDS_NAMESPACE 2
131 - do we just separate this by security domains and use a prefix for
132 multiple namespaces in the same domain?
137 struct ldlm_resource;
138 struct ldlm_namespace;
140 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
141 void *req_cookie, ldlm_mode_t mode, int flags,
144 struct ldlm_valblock_ops {
145 int (*lvbo_init)(struct ldlm_resource *res);
146 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
150 struct ldlm_namespace {
152 __u32 ns_client; /* is this a client-side lock tree? */
153 struct list_head *ns_hash; /* hash table for ns */
154 __u32 ns_refcount; /* count of resources in the hash */
155 struct list_head ns_root_list; /* all root resources in ns */
156 struct lustre_lock ns_lock; /* protects hash, refcount, list */
157 struct list_head ns_list_chain; /* position in global NS list */
159 struct proc_dir_entry *ns_proc_dir;
162 struct list_head ns_unused_list; /* all root resources in ns */
164 unsigned int ns_max_unused;
166 spinlock_t ns_counter_lock;
169 ldlm_res_policy ns_policy;
170 struct ldlm_valblock_ops *ns_lvbo;
176 * Resource hash table
180 #define RES_HASH_BITS 10
181 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
182 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
186 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
187 struct ldlm_lock_desc *new, void *data,
189 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
191 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
194 struct portals_handle l_handle; // must be first in the structure
196 struct ldlm_resource *l_resource;
197 struct ldlm_lock *l_parent;
198 struct list_head l_children;
199 struct list_head l_childof;
200 struct list_head l_lru;
201 struct list_head l_res_link; // position in one of three res lists
202 struct list_head l_export_chain; // per-export chain of locks
203 struct list_head l_pending_chain; // locks with callbacks pending
204 unsigned long l_callback_timeout;
206 ldlm_mode_t l_req_mode;
207 ldlm_mode_t l_granted_mode;
209 ldlm_completion_callback l_completion_ast;
210 ldlm_blocking_callback l_blocking_ast;
211 ldlm_glimpse_callback l_glimpse_ast;
214 struct obd_export *l_export;
215 /* XXX phil can fix this, I'm sure */
216 struct obd_export *l_conn_export;
217 // struct lustre_handle *l_connh;
219 struct lustre_handle l_remote_handle;
220 ldlm_policy_data_t l_policy_data;
222 /* This LVB is used only on the client side, as temporary storage for
223 * a lock value block received during an enqueue */
232 /* If the lock is granted, a process sleeps on this waitq to learn when
233 * it's no longer in use. If the lock is not granted, a process sleeps
234 * on this waitq to learn when it becomes granted. */
235 wait_queue_head_t l_waitq;
236 struct timeval l_enqueued_time;
237 unsigned long l_last_used; /* jiffies */
241 #define LDLM_PLAIN 10
242 #define LDLM_EXTENT 11
243 #define LDLM_FLOCK 12
245 #define LDLM_MIN_TYPE 10
246 #define LDLM_MAX_TYPE 12
248 struct ldlm_resource {
249 struct ldlm_namespace *lr_namespace;
250 struct list_head lr_hash;
251 struct ldlm_resource *lr_parent; /* 0 for a root resource */
252 struct list_head lr_children; /* list head for child resources */
253 struct list_head lr_childof; /* part of ns_root_list if root res,
254 * part of lr_children if child */
256 struct list_head lr_granted;
257 struct list_head lr_converting;
258 struct list_head lr_waiting;
259 ldlm_mode_t lr_most_restr;
260 __u32 lr_type; /* LDLM_PLAIN or LDLM_EXTENT */
261 struct ldlm_resource *lr_root;
262 struct ldlm_res_id lr_name;
263 atomic_t lr_refcount;
265 /* Server-side-only lock value block elements */
266 struct semaphore lr_lvb_sem;
270 /* lr_tmp holds a list head temporarily, during the building of a work
271 * queue. see ldlm_add_ast_work_item and ldlm_run_ast_work */
275 struct ldlm_ast_work {
276 struct ldlm_lock *w_lock;
278 struct ldlm_lock_desc w_desc;
279 struct list_head w_list;
285 extern struct obd_ops ldlm_obd_ops;
287 extern char *ldlm_lockname[];
288 extern char *ldlm_typename[];
289 extern char *ldlm_it2str(int it);
291 #define __LDLM_DEBUG(level, lock, format, a...) \
293 if (lock->l_resource == NULL) { \
294 CDEBUG(level, "### " format \
295 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "\
296 "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: " \
297 LPX64" expref: %d\n" , ## a, lock, \
298 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
299 lock->l_readers, lock->l_writers, \
300 ldlm_lockname[lock->l_granted_mode], \
301 ldlm_lockname[lock->l_req_mode], \
302 lock->l_flags, lock->l_remote_handle.cookie, \
304 atomic_read(&lock->l_export->exp_refcount) : -99); \
307 if (lock->l_resource->lr_type == LDLM_EXTENT) { \
308 CDEBUG(level, "### " format \
309 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
310 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64\
311 "] flags: %x remote: "LPX64" expref: %d\n" , ## a, \
312 lock->l_resource->lr_namespace->ns_name, lock, \
313 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
314 lock->l_readers, lock->l_writers, \
315 ldlm_lockname[lock->l_granted_mode], \
316 ldlm_lockname[lock->l_req_mode], \
317 lock->l_resource->lr_name.name[0], \
318 lock->l_resource->lr_name.name[1], \
319 atomic_read(&lock->l_resource->lr_refcount), \
320 ldlm_typename[lock->l_resource->lr_type], \
321 lock->l_policy_data.l_extent.start, \
322 lock->l_policy_data.l_extent.end, \
323 lock->l_flags, lock->l_remote_handle.cookie, \
325 atomic_read(&lock->l_export->exp_refcount) : -99); \
328 if (lock->l_resource->lr_type == LDLM_FLOCK) { \
329 CDEBUG(level, "### " format \
330 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
331 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d " \
332 "["LPU64"->"LPU64"] flags: %x remote: "LPX64 \
333 " expref: %d\n" , ## a, \
334 lock->l_resource->lr_namespace->ns_name, lock, \
335 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
336 lock->l_readers, lock->l_writers, \
337 ldlm_lockname[lock->l_granted_mode], \
338 ldlm_lockname[lock->l_req_mode], \
339 lock->l_resource->lr_name.name[0], \
340 lock->l_resource->lr_name.name[1], \
341 atomic_read(&lock->l_resource->lr_refcount), \
342 ldlm_typename[lock->l_resource->lr_type], \
343 lock->l_policy_data.l_flock.pid, \
344 lock->l_policy_data.l_flock.start, \
345 lock->l_policy_data.l_flock.end, \
346 lock->l_flags, lock->l_remote_handle.cookie, \
348 atomic_read(&lock->l_export->exp_refcount) : -99); \
352 CDEBUG(level, "### " format \
353 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
354 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x " \
355 "remote: "LPX64" expref: %d\n" , ## a, \
356 lock->l_resource->lr_namespace->ns_name, \
357 lock, lock->l_handle.h_cookie, \
358 atomic_read (&lock->l_refc), \
359 lock->l_readers, lock->l_writers, \
360 ldlm_lockname[lock->l_granted_mode], \
361 ldlm_lockname[lock->l_req_mode], \
362 lock->l_resource->lr_name.name[0], \
363 lock->l_resource->lr_name.name[1], \
364 atomic_read(&lock->l_resource->lr_refcount), \
365 ldlm_typename[lock->l_resource->lr_type], \
366 lock->l_flags, lock->l_remote_handle.cookie, \
368 atomic_read(&lock->l_export->exp_refcount) : -99); \
372 #define LDLM_DEBUG(lock, format, a...) __LDLM_DEBUG(D_DLMTRACE, lock, \
374 #define LDLM_ERROR(lock, format, a...) __LDLM_DEBUG(D_ERROR, lock, format, ## a)
376 #define LDLM_DEBUG_NOLOCK(format, a...) \
377 CDEBUG(D_DLMTRACE, "### " format "\n" , ## a)
379 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
380 int first_enq, ldlm_error_t *err);
386 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
387 #define LDLM_ITER_STOP 2 /* stop iterating */
389 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
390 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
392 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
394 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
396 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
397 ldlm_res_iterator_t iter, void *closure);
399 int ldlm_replay_locks(struct obd_import *imp);
400 void ldlm_change_cbdata(struct ldlm_namespace *, struct ldlm_res_id *,
401 ldlm_iterator_t iter, void *data);
404 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
407 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
411 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
412 void *data, int flag);
413 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
414 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
415 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
416 ldlm_blocking_callback, ldlm_glimpse_callback);
417 int ldlm_handle_convert(struct ptlrpc_request *req);
418 int ldlm_handle_cancel(struct ptlrpc_request *req);
419 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
420 int ldlm_get_ref(void);
421 void ldlm_put_ref(int force);
424 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
425 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
426 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh);
427 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *, int flags);
428 void ldlm_cancel_callback(struct ldlm_lock *);
429 int ldlm_lock_set_data(struct lustre_handle *, void *data);
430 void ldlm_lock_remove_from_lru(struct ldlm_lock *);
431 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
432 struct lustre_handle *);
434 static inline struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *h)
436 return __ldlm_handle2lock(h, 0);
439 #define LDLM_LOCK_PUT(lock) \
441 /*LDLM_DEBUG((lock), "put");*/ \
442 ldlm_lock_put(lock); \
445 #define LDLM_LOCK_GET(lock) \
447 ldlm_lock_get(lock); \
448 /*LDLM_DEBUG((lock), "get");*/ \
452 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
453 void ldlm_lock_put(struct ldlm_lock *lock);
454 void ldlm_lock_destroy(struct ldlm_lock *lock);
455 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
456 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
457 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
458 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
459 void ldlm_lock_allow_match(struct ldlm_lock *lock);
460 int ldlm_lock_match(struct ldlm_namespace *ns, int flags, struct ldlm_res_id *,
461 __u32 type, ldlm_policy_data_t *, ldlm_mode_t mode,
462 struct lustre_handle *);
463 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
465 void ldlm_lock_cancel(struct ldlm_lock *lock);
466 void ldlm_cancel_locks_for_export(struct obd_export *export);
467 void ldlm_reprocess_all(struct ldlm_resource *res);
468 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
469 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
470 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
473 int ldlm_test(struct obd_device *device, struct lustre_handle *connh);
474 int ldlm_regression_start(struct obd_device *obddev,
475 struct lustre_handle *connh,
476 unsigned int threads, unsigned int max_locks_in,
477 unsigned int num_resources_in,
478 unsigned int num_extents_in);
479 int ldlm_regression_stop(void);
483 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 local);
484 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
485 int ldlm_namespace_free(struct ldlm_namespace *ns, int force);
486 int ldlm_proc_setup(void);
487 void ldlm_proc_cleanup(void);
489 /* resource.c - internal */
490 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
491 struct ldlm_resource *parent,
492 struct ldlm_res_id, __u32 type,
494 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
495 int ldlm_resource_putref(struct ldlm_resource *res);
496 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
497 struct ldlm_lock *lock);
498 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
499 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
500 void ldlm_dump_all_namespaces(void);
501 void ldlm_namespace_dump(struct ldlm_namespace *);
502 void ldlm_resource_dump(struct ldlm_resource *);
503 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
507 int ldlm_expired_completion_wait(void *data);
508 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
509 int ldlm_cli_enqueue(struct obd_export *exp,
510 struct ptlrpc_request *req,
511 struct ldlm_namespace *ns,
514 ldlm_policy_data_t *,
517 ldlm_blocking_callback blocking,
518 ldlm_completion_callback completion,
519 ldlm_glimpse_callback glimpse,
524 struct lustre_handle *lockh);
525 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
526 void *data, __u32 data_len);
527 int ldlm_cli_convert(struct lustre_handle *, int new_mode, int *flags);
528 int ldlm_cli_cancel(struct lustre_handle *lockh);
529 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
530 int flags, void *opaque);
533 /* This has to be here because recursive inclusion sucks. */
534 int intent_disposition(struct ldlm_reply *rep, int flag);
535 void intent_set_disposition(struct ldlm_reply *rep, int flag);
536 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
537 void *data, int flag);
540 /* ioctls for trying requests */
541 #define IOC_LDLM_TYPE 'f'
542 #define IOC_LDLM_MIN_NR 40
544 #define IOC_LDLM_TEST _IOWR('f', 40, long)
545 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
546 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
547 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
548 #define IOC_LDLM_MAX_NR 43