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 /* A lock contributes to the kms calculation until it has finished the part
94 * of it's cancelation that performs write back on its dirty pages. It
95 * can remain on the granted list during this whole time. Threads racing
96 * to update the kms after performing their writeback need to know to
97 * exclude each others locks from the calculation as they walk the granted
99 #define LDLM_FL_KMS_IGNORE 0x200000
101 /* The blocking callback is overloaded to perform two functions. These flags
102 * indicate which operation should be performed. */
103 #define LDLM_CB_BLOCKING 1
104 #define LDLM_CB_CANCELING 2
106 /* compatibility matrix */
107 #define LCK_COMPAT_EX LCK_NL
108 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
109 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
110 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
111 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
112 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX)
113 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
115 static ldlm_mode_t lck_compat_array[] = {
116 [LCK_EX] LCK_COMPAT_EX,
117 [LCK_PW] LCK_COMPAT_PW,
118 [LCK_PR] LCK_COMPAT_PR,
119 [LCK_CW] LCK_COMPAT_CW,
120 [LCK_CR] LCK_COMPAT_CR,
121 [LCK_NL] LCK_COMPAT_NL,
122 [LCK_GROUP] LCK_COMPAT_GROUP
125 static inline void lockmode_verify(ldlm_mode_t mode)
127 LASSERT(mode >= LCK_EX && mode <= LCK_GROUP);
130 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
132 return (lck_compat_array[exist] & new);
137 * cluster name spaces
141 #define DLM_OST_NAMESPACE 1
142 #define DLM_MDS_NAMESPACE 2
145 - do we just separate this by security domains and use a prefix for
146 multiple namespaces in the same domain?
151 struct ldlm_resource;
152 struct ldlm_namespace;
154 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
155 void *req_cookie, ldlm_mode_t mode, int flags,
158 struct ldlm_valblock_ops {
159 int (*lvbo_init)(struct ldlm_resource *res);
160 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
161 int buf_idx, int increase);
164 struct ldlm_namespace {
166 __u32 ns_client; /* is this a client-side lock tree? */
167 struct list_head *ns_hash; /* hash table for ns */
168 __u32 ns_refcount; /* count of resources in the hash */
169 struct list_head ns_root_list; /* all root resources in ns */
170 struct lustre_lock ns_lock; /* protects hash, refcount, list */
171 struct list_head ns_list_chain; /* position in global NS list */
173 struct proc_dir_entry *ns_proc_dir;
176 struct list_head ns_unused_list; /* all root resources in ns */
178 unsigned int ns_max_unused;
180 spinlock_t ns_counter_lock;
183 ldlm_res_policy ns_policy;
184 struct ldlm_valblock_ops *ns_lvbo;
190 * Resource hash table
194 #define RES_HASH_BITS 10
195 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
196 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
200 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
201 struct ldlm_lock_desc *new, void *data,
203 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
205 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
208 struct portals_handle l_handle; // must be first in the structure
210 struct ldlm_resource *l_resource;
211 struct ldlm_lock *l_parent;
212 struct list_head l_children;
213 struct list_head l_childof;
214 struct list_head l_lru;
215 struct list_head l_res_link; // position in one of three res lists
216 struct list_head l_export_chain; // per-export chain of locks
218 ldlm_mode_t l_req_mode;
219 ldlm_mode_t l_granted_mode;
221 ldlm_completion_callback l_completion_ast;
222 ldlm_blocking_callback l_blocking_ast;
223 ldlm_glimpse_callback l_glimpse_ast;
225 struct obd_export *l_export;
226 struct obd_export *l_conn_export;
228 struct lustre_handle l_remote_handle;
229 ldlm_policy_data_t l_policy_data;
235 /* If the lock is granted, a process sleeps on this waitq to learn when
236 * it's no longer in use. If the lock is not granted, a process sleeps
237 * on this waitq to learn when it becomes granted. */
238 wait_queue_head_t l_waitq;
239 struct timeval l_enqueued_time;
241 unsigned long l_last_used; /* jiffies */
242 struct ldlm_extent l_req_extent;
244 /* Client-side-only members */
245 __u32 l_lvb_len; /* temporary storage for */
246 void *l_lvb_data; /* an LVB received during */
247 void *l_lvb_swabber; /* an enqueue */
250 /* Server-side-only members */
251 struct list_head l_pending_chain; /* callbacks pending */
252 unsigned long l_callback_timeout;
255 #define LDLM_PLAIN 10
256 #define LDLM_EXTENT 11
257 #define LDLM_FLOCK 12
259 #define LDLM_MIN_TYPE 10
260 #define LDLM_MAX_TYPE 12
262 struct ldlm_resource {
263 struct ldlm_namespace *lr_namespace;
264 struct list_head lr_hash;
265 struct ldlm_resource *lr_parent; /* 0 for a root resource */
266 struct list_head lr_children; /* list head for child resources */
267 struct list_head lr_childof; /* part of ns_root_list if root res,
268 * part of lr_children if child */
270 struct list_head lr_granted;
271 struct list_head lr_converting;
272 struct list_head lr_waiting;
273 ldlm_mode_t lr_most_restr;
274 __u32 lr_type; /* LDLM_PLAIN or LDLM_EXTENT */
275 struct ldlm_resource *lr_root;
276 struct ldlm_res_id lr_name;
277 atomic_t lr_refcount;
279 /* Server-side-only lock value block elements */
280 struct semaphore lr_lvb_sem;
284 /* lr_tmp holds a list head temporarily, during the building of a work
285 * queue. see ldlm_add_ast_work_item and ldlm_run_ast_work */
289 struct ldlm_ast_work {
290 struct ldlm_lock *w_lock;
292 struct ldlm_lock_desc w_desc;
293 struct list_head w_list;
299 extern struct obd_ops ldlm_obd_ops;
301 extern char *ldlm_lockname[];
302 extern char *ldlm_typename[];
303 extern char *ldlm_it2str(int it);
305 #define __LDLM_DEBUG(level, lock, format, a...) \
307 if (lock->l_resource == NULL) { \
308 CDEBUG(level, "### " format \
309 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "\
310 "res: \?\? rrc=\?\? type: \?\?\? flags: %x remote: " \
311 LPX64" expref: %d\n" , ## a, lock, \
312 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
313 lock->l_readers, lock->l_writers, \
314 ldlm_lockname[lock->l_granted_mode], \
315 ldlm_lockname[lock->l_req_mode], \
316 lock->l_flags, lock->l_remote_handle.cookie, \
318 atomic_read(&lock->l_export->exp_refcount) : -99); \
321 if (lock->l_resource->lr_type == LDLM_EXTENT) { \
322 CDEBUG(level, "### " format \
323 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
324 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64\
325 "] (req "LPU64"->"LPU64") flags: %x remote: "LPX64 \
326 " expref: %d\n" , ## a, \
327 lock->l_resource->lr_namespace->ns_name, lock, \
328 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
329 lock->l_readers, lock->l_writers, \
330 ldlm_lockname[lock->l_granted_mode], \
331 ldlm_lockname[lock->l_req_mode], \
332 lock->l_resource->lr_name.name[0], \
333 lock->l_resource->lr_name.name[1], \
334 atomic_read(&lock->l_resource->lr_refcount), \
335 ldlm_typename[lock->l_resource->lr_type], \
336 lock->l_policy_data.l_extent.start, \
337 lock->l_policy_data.l_extent.end, \
338 lock->l_req_extent.start, lock->l_req_extent.end, \
339 lock->l_flags, lock->l_remote_handle.cookie, \
341 atomic_read(&lock->l_export->exp_refcount) : -99); \
344 if (lock->l_resource->lr_type == LDLM_FLOCK) { \
345 CDEBUG(level, "### " format \
346 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
347 "res: "LPU64"/"LPU64" rrc: %d type: %s pid: "LPU64" " \
348 "["LPU64"->"LPU64"] flags: %x remote: "LPX64 \
349 " expref: %d\n" , ## a, \
350 lock->l_resource->lr_namespace->ns_name, lock, \
351 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
352 lock->l_readers, lock->l_writers, \
353 ldlm_lockname[lock->l_granted_mode], \
354 ldlm_lockname[lock->l_req_mode], \
355 lock->l_resource->lr_name.name[0], \
356 lock->l_resource->lr_name.name[1], \
357 atomic_read(&lock->l_resource->lr_refcount), \
358 ldlm_typename[lock->l_resource->lr_type], \
359 lock->l_policy_data.l_flock.pid, \
360 lock->l_policy_data.l_flock.start, \
361 lock->l_policy_data.l_flock.end, \
362 lock->l_flags, lock->l_remote_handle.cookie, \
364 atomic_read(&lock->l_export->exp_refcount) : -99); \
368 CDEBUG(level, "### " format \
369 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
370 "res: "LPU64"/"LPU64" rrc: %d type: %s flags: %x " \
371 "remote: "LPX64" expref: %d\n" , ## a, \
372 lock->l_resource->lr_namespace->ns_name, \
373 lock, lock->l_handle.h_cookie, \
374 atomic_read (&lock->l_refc), \
375 lock->l_readers, lock->l_writers, \
376 ldlm_lockname[lock->l_granted_mode], \
377 ldlm_lockname[lock->l_req_mode], \
378 lock->l_resource->lr_name.name[0], \
379 lock->l_resource->lr_name.name[1], \
380 atomic_read(&lock->l_resource->lr_refcount), \
381 ldlm_typename[lock->l_resource->lr_type], \
382 lock->l_flags, lock->l_remote_handle.cookie, \
384 atomic_read(&lock->l_export->exp_refcount) : -99); \
388 #define LDLM_DEBUG(lock, format, a...) __LDLM_DEBUG(D_DLMTRACE, lock, \
390 #define LDLM_ERROR(lock, format, a...) __LDLM_DEBUG(D_ERROR, lock, format, ## a)
392 #define LDLM_DEBUG_NOLOCK(format, a...) \
393 CDEBUG(D_DLMTRACE, "### " format "\n" , ## a)
395 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
396 int first_enq, ldlm_error_t *err);
402 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
403 #define LDLM_ITER_STOP 2 /* stop iterating */
405 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
406 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
408 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
410 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
412 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
413 ldlm_res_iterator_t iter, void *closure);
415 int ldlm_replay_locks(struct obd_import *imp);
416 void ldlm_change_cbdata(struct ldlm_namespace *, struct ldlm_res_id *,
417 ldlm_iterator_t iter, void *data);
420 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
423 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
427 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
428 void *data, int flag);
429 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
430 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
431 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
432 ldlm_blocking_callback, ldlm_glimpse_callback);
433 int ldlm_handle_convert(struct ptlrpc_request *req);
434 int ldlm_handle_cancel(struct ptlrpc_request *req);
435 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
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(struct ldlm_lock *lock, struct lustre_handle *lockh);
443 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *, int flags);
444 void ldlm_cancel_callback(struct ldlm_lock *);
445 int ldlm_lock_set_data(struct lustre_handle *, void *data);
446 void ldlm_lock_remove_from_lru(struct ldlm_lock *);
447 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
448 struct lustre_handle *);
450 static inline struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *h)
452 return __ldlm_handle2lock(h, 0);
455 #define LDLM_LOCK_PUT(lock) \
457 /*LDLM_DEBUG((lock), "put");*/ \
458 ldlm_lock_put(lock); \
461 #define LDLM_LOCK_GET(lock) \
463 ldlm_lock_get(lock); \
464 /*LDLM_DEBUG((lock), "get");*/ \
468 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
469 void ldlm_lock_put(struct ldlm_lock *lock);
470 void ldlm_lock_destroy(struct ldlm_lock *lock);
471 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
472 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
473 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
474 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
475 void ldlm_lock_allow_match(struct ldlm_lock *lock);
476 int ldlm_lock_match(struct ldlm_namespace *ns, int flags, struct ldlm_res_id *,
477 __u32 type, ldlm_policy_data_t *, ldlm_mode_t mode,
478 struct lustre_handle *);
479 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
481 void ldlm_lock_cancel(struct ldlm_lock *lock);
482 void ldlm_cancel_locks_for_export(struct obd_export *export);
483 void ldlm_reprocess_all(struct ldlm_resource *res);
484 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
485 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
486 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
489 int ldlm_test(struct obd_device *device, struct lustre_handle *connh);
490 int ldlm_regression_start(struct obd_device *obddev,
491 struct lustre_handle *connh,
492 unsigned int threads, unsigned int max_locks_in,
493 unsigned int num_resources_in,
494 unsigned int num_extents_in);
495 int ldlm_regression_stop(void);
499 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 local);
500 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
501 int ldlm_namespace_free(struct ldlm_namespace *ns, int force);
502 int ldlm_proc_setup(void);
503 void ldlm_proc_cleanup(void);
505 /* resource.c - internal */
506 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
507 struct ldlm_resource *parent,
508 struct ldlm_res_id, __u32 type,
510 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
511 int ldlm_resource_putref(struct ldlm_resource *res);
512 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
513 struct ldlm_lock *lock);
514 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
515 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
516 void ldlm_dump_all_namespaces(void);
517 void ldlm_namespace_dump(struct ldlm_namespace *);
518 void ldlm_resource_dump(struct ldlm_resource *);
519 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
523 int ldlm_expired_completion_wait(void *data);
524 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
525 int ldlm_cli_enqueue(struct obd_export *exp,
526 struct ptlrpc_request *req,
527 struct ldlm_namespace *ns,
530 ldlm_policy_data_t *,
533 ldlm_blocking_callback blocking,
534 ldlm_completion_callback completion,
535 ldlm_glimpse_callback glimpse,
540 struct lustre_handle *lockh);
541 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
542 void *data, __u32 data_len);
543 int ldlm_cli_convert(struct lustre_handle *, int new_mode, int *flags);
544 int ldlm_cli_cancel(struct lustre_handle *lockh);
545 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
546 int flags, void *opaque);
549 /* This has to be here because recursive inclusion sucks. */
550 int intent_disposition(struct ldlm_reply *rep, int flag);
551 void intent_set_disposition(struct ldlm_reply *rep, int flag);
552 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
553 void *data, int flag);
556 /* ioctls for trying requests */
557 #define IOC_LDLM_TYPE 'f'
558 #define IOC_LDLM_MIN_NR 40
560 #define IOC_LDLM_TEST _IOWR('f', 40, long)
561 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
562 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
563 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
564 #define IOC_LDLM_MAX_NR 43