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,
32 ELDLM_NO_LOCK_DATA = 303,
34 ELDLM_NAMESPACE_EXISTS = 400,
35 ELDLM_BAD_NAMESPACE = 401
38 #define LDLM_NAMESPACE_SERVER 0
39 #define LDLM_NAMESPACE_CLIENT 1
41 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
43 /* If the server returns one of these flags, then the lock was put on that list.
44 * If the client sends one of these flags (during recovery ONLY!), it wants the
45 * lock added to the specified list, no questions asked. -p */
46 #define LDLM_FL_BLOCK_GRANTED 0x000002
47 #define LDLM_FL_BLOCK_CONV 0x000004
48 #define LDLM_FL_BLOCK_WAIT 0x000008
50 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
51 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was sent */
52 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
53 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
55 /* Lock is being replayed. This could probably be implied by the fact that one
56 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
57 #define LDLM_FL_REPLAY 0x000100
59 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
60 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
62 /* don't run the cancel callback under ldlm_cli_cancel_unused */
63 #define LDLM_FL_FAILED 0x000800
65 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
66 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
67 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
68 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
69 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
71 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
74 /* file & record locking */
75 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
76 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
78 /* These are flags that are mapped into the flags and ASTs of blocking locks */
79 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
80 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
81 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
83 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
84 * the LVB filling happens _after_ the lock has been granted, so another thread
85 * can match before the LVB has been updated. As a dirty hack, we set
86 * LDLM_FL_CAN_MATCH only after we've done the LVB poop.
88 * The proper fix is to do the granting inside of the completion AST, which can
89 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
90 * pretty high-risk, though, and would need a lot more testing. */
91 #define LDLM_FL_CAN_MATCH 0x100000
93 /* The blocking callback is overloaded to perform two functions. These flags
94 * indicate which operation should be performed. */
95 #define LDLM_CB_BLOCKING 1
96 #define LDLM_CB_CANCELING 2
98 /* compatibility matrix */
99 #define LCK_COMPAT_EX LCK_NL
100 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
101 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
102 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
103 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
104 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX)
105 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
107 static ldlm_mode_t lck_compat_array[] = {
108 [LCK_EX] LCK_COMPAT_EX,
109 [LCK_PW] LCK_COMPAT_PW,
110 [LCK_PR] LCK_COMPAT_PR,
111 [LCK_CW] LCK_COMPAT_CW,
112 [LCK_CR] LCK_COMPAT_CR,
113 [LCK_NL] LCK_COMPAT_NL,
114 [LCK_GROUP] LCK_COMPAT_GROUP
117 static inline void lockmode_verify(ldlm_mode_t mode)
119 LASSERT(mode >= LCK_EX && mode <= LCK_GROUP);
122 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
124 return (lck_compat_array[exist] & new);
129 * cluster name spaces
133 #define DLM_OST_NAMESPACE 1
134 #define DLM_MDS_NAMESPACE 2
137 - do we just separate this by security domains and use a prefix for
138 multiple namespaces in the same domain?
143 struct ldlm_resource;
144 struct ldlm_namespace;
146 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
147 void *req_cookie, ldlm_mode_t mode, int flags,
150 struct ldlm_valblock_ops {
151 int (*lvbo_init)(struct ldlm_resource *res);
152 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
153 int buf_idx, int increase);
156 struct ldlm_namespace {
158 __u32 ns_client; /* is this a client-side lock tree? */
159 struct list_head *ns_hash; /* hash table for ns */
160 __u32 ns_refcount; /* count of resources in the hash */
161 struct list_head ns_root_list; /* all root resources in ns */
162 struct lustre_lock ns_lock; /* protects hash, refcount, list */
163 struct list_head ns_list_chain; /* position in global NS list */
165 struct proc_dir_entry *ns_proc_dir;
168 struct list_head ns_unused_list; /* all root resources in ns */
170 unsigned int ns_max_unused;
172 spinlock_t ns_counter_lock;
175 ldlm_res_policy ns_policy;
176 struct ldlm_valblock_ops *ns_lvbo;
182 * Resource hash table
186 #define RES_HASH_BITS 10
187 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
188 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
192 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
193 struct ldlm_lock_desc *new, void *data,
195 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
197 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
200 struct portals_handle l_handle; // must be first in the structure
202 struct ldlm_resource *l_resource;
203 struct ldlm_lock *l_parent;
204 struct list_head l_children;
205 struct list_head l_childof;
206 struct list_head l_lru;
207 struct list_head l_res_link; // position in one of three res lists
208 struct list_head l_export_chain; // per-export chain of locks
210 ldlm_mode_t l_req_mode;
211 ldlm_mode_t l_granted_mode;
213 ldlm_completion_callback l_completion_ast;
214 ldlm_blocking_callback l_blocking_ast;
215 ldlm_glimpse_callback l_glimpse_ast;
217 struct obd_export *l_export;
218 struct obd_export *l_conn_export;
220 struct lustre_handle l_remote_handle;
221 ldlm_policy_data_t l_policy_data;
227 /* If the lock is granted, a process sleeps on this waitq to learn when
228 * it's no longer in use. If the lock is not granted, a process sleeps
229 * on this waitq to learn when it becomes granted. */
230 wait_queue_head_t l_waitq;
231 struct timeval l_enqueued_time;
233 unsigned long l_last_used; /* jiffies */
234 struct ldlm_extent l_req_extent;
236 /* Client-side-only members */
237 __u32 l_lvb_len; /* temporary storage for */
238 void *l_lvb_data; /* an LVB received during */
239 void *l_lvb_swabber; /* an enqueue */
242 /* Server-side-only members */
243 struct list_head l_pending_chain; /* callbacks pending */
244 unsigned long l_callback_timeout;
247 #define LDLM_PLAIN 10
248 #define LDLM_EXTENT 11
249 #define LDLM_FLOCK 12
251 #define LDLM_MIN_TYPE 10
252 #define LDLM_MAX_TYPE 12
254 struct ldlm_resource {
255 struct ldlm_namespace *lr_namespace;
256 struct list_head lr_hash;
257 struct ldlm_resource *lr_parent; /* 0 for a root resource */
258 struct list_head lr_children; /* list head for child resources */
259 struct list_head lr_childof; /* part of ns_root_list if root res,
260 * part of lr_children if child */
262 struct list_head lr_granted;
263 struct list_head lr_converting;
264 struct list_head lr_waiting;
265 ldlm_mode_t lr_most_restr;
266 __u32 lr_type; /* LDLM_PLAIN or LDLM_EXTENT */
267 struct ldlm_resource *lr_root;
268 struct ldlm_res_id lr_name;
269 atomic_t lr_refcount;
271 /* Server-side-only lock value block elements */
272 struct semaphore lr_lvb_sem;
276 /* lr_tmp holds a list head temporarily, during the building of a work
277 * queue. see ldlm_add_ast_work_item and ldlm_run_ast_work */
281 struct ldlm_ast_work {
282 struct ldlm_lock *w_lock;
284 struct ldlm_lock_desc w_desc;
285 struct list_head w_list;
291 extern struct obd_ops ldlm_obd_ops;
293 extern char *ldlm_lockname[];
294 extern char *ldlm_typename[];
295 extern char *ldlm_it2str(int it);
297 #define __LDLM_DEBUG(level, lock, format, a...) \
299 if (lock->l_resource == NULL) { \
300 CDEBUG(level, "### " format \
301 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "\
302 "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: " \
303 LPX64" expref: %d\n" , ## a, lock, \
304 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
305 lock->l_readers, lock->l_writers, \
306 ldlm_lockname[lock->l_granted_mode], \
307 ldlm_lockname[lock->l_req_mode], \
308 lock->l_flags, lock->l_remote_handle.cookie, \
310 atomic_read(&lock->l_export->exp_refcount) : -99); \
313 if (lock->l_resource->lr_type == LDLM_EXTENT) { \
314 CDEBUG(level, "### " format \
315 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
316 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64\
317 "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64 \
318 " expref: %d\n" , ## a, \
319 lock->l_resource->lr_namespace->ns_name, lock, \
320 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
321 lock->l_readers, lock->l_writers, \
322 ldlm_lockname[lock->l_granted_mode], \
323 ldlm_lockname[lock->l_req_mode], \
324 lock->l_resource->lr_name.name[0], \
325 lock->l_resource->lr_name.name[1], \
326 atomic_read(&lock->l_resource->lr_refcount), \
327 ldlm_typename[lock->l_resource->lr_type], \
328 lock->l_policy_data.l_extent.start, \
329 lock->l_policy_data.l_extent.end, \
330 lock->l_req_extent.start, lock->l_req_extent.end, \
331 lock->l_flags, lock->l_remote_handle.cookie, \
333 atomic_read(&lock->l_export->exp_refcount) : -99); \
336 if (lock->l_resource->lr_type == LDLM_FLOCK) { \
337 CDEBUG(level, "### " format \
338 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
339 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: "LPU64" " \
340 "["LPU64"->"LPU64"] flags: %x remote: "LPX64 \
341 " expref: %d\n" , ## a, \
342 lock->l_resource->lr_namespace->ns_name, lock, \
343 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
344 lock->l_readers, lock->l_writers, \
345 ldlm_lockname[lock->l_granted_mode], \
346 ldlm_lockname[lock->l_req_mode], \
347 lock->l_resource->lr_name.name[0], \
348 lock->l_resource->lr_name.name[1], \
349 atomic_read(&lock->l_resource->lr_refcount), \
350 ldlm_typename[lock->l_resource->lr_type], \
351 lock->l_policy_data.l_flock.pid, \
352 lock->l_policy_data.l_flock.start, \
353 lock->l_policy_data.l_flock.end, \
354 lock->l_flags, lock->l_remote_handle.cookie, \
356 atomic_read(&lock->l_export->exp_refcount) : -99); \
360 CDEBUG(level, "### " format \
361 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
362 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x " \
363 "remote: "LPX64" expref: %d\n" , ## a, \
364 lock->l_resource->lr_namespace->ns_name, \
365 lock, lock->l_handle.h_cookie, \
366 atomic_read (&lock->l_refc), \
367 lock->l_readers, lock->l_writers, \
368 ldlm_lockname[lock->l_granted_mode], \
369 ldlm_lockname[lock->l_req_mode], \
370 lock->l_resource->lr_name.name[0], \
371 lock->l_resource->lr_name.name[1], \
372 atomic_read(&lock->l_resource->lr_refcount), \
373 ldlm_typename[lock->l_resource->lr_type], \
374 lock->l_flags, lock->l_remote_handle.cookie, \
376 atomic_read(&lock->l_export->exp_refcount) : -99); \
380 #define LDLM_DEBUG(lock, format, a...) __LDLM_DEBUG(D_DLMTRACE, lock, \
382 #define LDLM_ERROR(lock, format, a...) __LDLM_DEBUG(D_ERROR, lock, format, ## a)
384 #define LDLM_DEBUG_NOLOCK(format, a...) \
385 CDEBUG(D_DLMTRACE, "### " format "\n" , ## a)
387 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
388 int first_enq, ldlm_error_t *err);
394 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
395 #define LDLM_ITER_STOP 2 /* stop iterating */
397 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
398 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
400 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
402 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
404 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
405 ldlm_res_iterator_t iter, void *closure);
407 int ldlm_replay_locks(struct obd_import *imp);
408 void ldlm_change_cbdata(struct ldlm_namespace *, struct ldlm_res_id *,
409 ldlm_iterator_t iter, void *data);
412 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
415 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
419 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
420 void *data, int flag);
421 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
422 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
423 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
424 ldlm_blocking_callback, ldlm_glimpse_callback);
425 int ldlm_handle_convert(struct ptlrpc_request *req);
426 int ldlm_handle_cancel(struct ptlrpc_request *req);
427 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
428 int ldlm_get_ref(void);
429 void ldlm_put_ref(int force);
432 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
433 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
434 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh);
435 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *, int flags);
436 void ldlm_cancel_callback(struct ldlm_lock *);
437 int ldlm_lock_set_data(struct lustre_handle *, void *data);
438 void ldlm_lock_remove_from_lru(struct ldlm_lock *);
439 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
440 struct lustre_handle *);
442 static inline struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *h)
444 return __ldlm_handle2lock(h, 0);
447 #define LDLM_LOCK_PUT(lock) \
449 /*LDLM_DEBUG((lock), "put");*/ \
450 ldlm_lock_put(lock); \
453 #define LDLM_LOCK_GET(lock) \
455 ldlm_lock_get(lock); \
456 /*LDLM_DEBUG((lock), "get");*/ \
460 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
461 void ldlm_lock_put(struct ldlm_lock *lock);
462 void ldlm_lock_destroy(struct ldlm_lock *lock);
463 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
464 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
465 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
466 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
467 void ldlm_lock_allow_match(struct ldlm_lock *lock);
468 int ldlm_lock_match(struct ldlm_namespace *ns, int flags, struct ldlm_res_id *,
469 __u32 type, ldlm_policy_data_t *, ldlm_mode_t mode,
470 struct lustre_handle *);
471 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
473 void ldlm_lock_cancel(struct ldlm_lock *lock);
474 void ldlm_cancel_locks_for_export(struct obd_export *export);
475 void ldlm_reprocess_all(struct ldlm_resource *res);
476 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
477 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
478 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
481 int ldlm_test(struct obd_device *device, struct lustre_handle *connh);
482 int ldlm_regression_start(struct obd_device *obddev,
483 struct lustre_handle *connh,
484 unsigned int threads, unsigned int max_locks_in,
485 unsigned int num_resources_in,
486 unsigned int num_extents_in);
487 int ldlm_regression_stop(void);
491 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 local);
492 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
493 int ldlm_namespace_free(struct ldlm_namespace *ns, int force);
494 int ldlm_proc_setup(void);
495 void ldlm_proc_cleanup(void);
497 /* resource.c - internal */
498 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
499 struct ldlm_resource *parent,
500 struct ldlm_res_id, __u32 type,
502 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
503 int ldlm_resource_putref(struct ldlm_resource *res);
504 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
505 struct ldlm_lock *lock);
506 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
507 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
508 void ldlm_dump_all_namespaces(void);
509 void ldlm_namespace_dump(struct ldlm_namespace *);
510 void ldlm_resource_dump(struct ldlm_resource *);
511 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
515 int ldlm_expired_completion_wait(void *data);
516 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
517 int ldlm_cli_enqueue(struct obd_export *exp,
518 struct ptlrpc_request *req,
519 struct ldlm_namespace *ns,
522 ldlm_policy_data_t *,
525 ldlm_blocking_callback blocking,
526 ldlm_completion_callback completion,
527 ldlm_glimpse_callback glimpse,
532 struct lustre_handle *lockh);
533 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
534 void *data, __u32 data_len);
535 int ldlm_cli_convert(struct lustre_handle *, int new_mode, int *flags);
536 int ldlm_cli_cancel(struct lustre_handle *lockh);
537 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
538 int flags, void *opaque);
541 /* This has to be here because recursive inclusion sucks. */
542 int intent_disposition(struct ldlm_reply *rep, int flag);
543 void intent_set_disposition(struct ldlm_reply *rep, int flag);
544 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
545 void *data, int flag);
548 /* ioctls for trying requests */
549 #define IOC_LDLM_TYPE 'f'
550 #define IOC_LDLM_MIN_NR 40
552 #define IOC_LDLM_TEST _IOWR('f', 40, long)
553 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
554 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
555 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
556 #define IOC_LDLM_MAX_NR 43