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 (1 << 0) /* 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 (1 << 1)
46 #define LDLM_FL_BLOCK_CONV (1 << 2)
47 #define LDLM_FL_BLOCK_WAIT (1 << 3)
49 #define LDLM_FL_CBPENDING (1 << 4) // this lock is being destroyed
50 #define LDLM_FL_AST_SENT (1 << 5) // blocking or cancel packet was sent
51 #define LDLM_FL_WAIT_NOREPROC (1 << 6)// not a real lock flag,not saved in lock
52 #define LDLM_FL_CANCEL (1 << 7) // 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 (1 << 8)
58 #define LDLM_FL_INTENT_ONLY (1 << 9) /* don't grant lock, just do intent */
59 #define LDLM_FL_LOCAL_ONLY (1 << 10) /* see ldlm_cli_cancel_unused */
61 /* don't run the cancel callback under ldlm_cli_cancel_unused */
62 #define LDLM_FL_NO_CALLBACK (1 << 11)
64 #define LDLM_FL_HAS_INTENT (1 << 12) /* lock request has intent */
65 #define LDLM_FL_CANCELING (1 << 13) /* lock cancel has already been sent */
66 #define LDLM_FL_LOCAL (1 << 14) // a local lock (ie, no srv/cli split)
67 #define LDLM_FL_WARN (1 << 15) /* see ldlm_cli_cancel_unused */
68 #define LDLM_FL_MATCH_DATA (1 << 16) /* see ldlm_lock_match */
70 /* The blocking callback is overloaded to perform two functions. These flags
71 * indicate which operation should be performed. */
72 #define LDLM_CB_BLOCKING 1
73 #define LDLM_CB_CANCELING 2
75 #define L2B(c) (1 << c)
77 /* compatibility matrix */
78 #define LCK_COMPAT_EX L2B(LCK_NL)
79 #define LCK_COMPAT_PW (LCK_COMPAT_EX | L2B(LCK_CR))
80 #define LCK_COMPAT_PR (LCK_COMPAT_PW | L2B(LCK_PR))
81 #define LCK_COMPAT_CW (LCK_COMPAT_PW | L2B(LCK_CW))
82 #define LCK_COMPAT_CR (LCK_COMPAT_CW | L2B(LCK_PR) | L2B(LCK_PW))
83 #define LCK_COMPAT_NL (LCK_COMPAT_CR | L2B(LCK_EX))
85 static ldlm_mode_t lck_compat_array[] = {
86 [LCK_EX] LCK_COMPAT_EX,
87 [LCK_PW] LCK_COMPAT_PW,
88 [LCK_PR] LCK_COMPAT_PR,
89 [LCK_CW] LCK_COMPAT_CW,
90 [LCK_CR] LCK_COMPAT_CR,
91 [LCK_NL] LCK_COMPAT_NL
94 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
96 if (exist < LCK_EX || exist > LCK_NL)
98 if (new < LCK_EX || new > LCK_NL)
101 return (lck_compat_array[exist] & L2B(new));
106 * cluster name spaces
110 #define DLM_OST_NAMESPACE 1
111 #define DLM_MDS_NAMESPACE 2
114 - do we just separate this by security domains and use a prefix for
115 multiple namespaces in the same domain?
119 struct ldlm_namespace {
121 __u32 ns_client; /* is this a client-side lock tree? */
122 struct list_head *ns_hash; /* hash table for ns */
123 __u32 ns_refcount; /* count of resources in the hash */
124 struct list_head ns_root_list; /* all root resources in ns */
125 struct lustre_lock ns_lock; /* protects hash, refcount, list */
126 struct list_head ns_list_chain; /* position in global NS list */
128 struct proc_dir_entry *ns_proc_dir;
131 struct list_head ns_unused_list; /* all root resources in ns */
132 unsigned int ns_nr_unused;
133 unsigned int ns_max_unused;
135 spinlock_t ns_counter_lock;
142 * Resource hash table
146 #define RES_HASH_BITS 10
147 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
148 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
152 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
153 struct ldlm_lock_desc *new, void *data,
155 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
159 struct portals_handle l_handle; // must be first in the structure
161 struct ldlm_resource *l_resource;
162 struct ldlm_lock *l_parent;
163 struct list_head l_children;
164 struct list_head l_childof;
165 struct list_head l_lru;
166 struct list_head l_res_link; // position in one of three res lists
167 struct list_head l_export_chain; // per-export chain of locks
168 struct list_head l_pending_chain; // locks with callbacks pending
169 unsigned long l_callback_timeout;
171 ldlm_mode_t l_req_mode;
172 ldlm_mode_t l_granted_mode;
174 ldlm_completion_callback l_completion_ast;
175 ldlm_blocking_callback l_blocking_ast;
177 struct obd_export *l_export;
178 struct lustre_handle *l_connh;
180 struct lustre_handle l_remote_handle;
182 struct ldlm_extent l_extent;
183 __u32 l_version[RES_VERSION_SIZE];
189 /* If the lock is granted, a process sleeps on this waitq to learn when
190 * it's no longer in use. If the lock is not granted, a process sleeps
191 * on this waitq to learn when it becomes granted. */
192 wait_queue_head_t l_waitq;
193 struct timeval l_enqueued_time;
196 typedef int (*ldlm_res_compat)(struct ldlm_lock *child, struct ldlm_lock *new);
197 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
198 void *req_cookie, ldlm_mode_t mode, int flags,
201 #define LDLM_PLAIN 10
202 #define LDLM_EXTENT 11
204 #define LDLM_MIN_TYPE 10
205 #define LDLM_MAX_TYPE 11
207 struct ldlm_resource {
208 struct ldlm_namespace *lr_namespace;
209 struct list_head lr_hash;
210 struct ldlm_resource *lr_parent; /* 0 for a root resource */
211 struct list_head lr_children; /* list head for child resources */
212 struct list_head lr_childof; /* part of ns_root_list if root res,
213 * part of lr_children if child */
215 struct list_head lr_granted;
216 struct list_head lr_converting;
217 struct list_head lr_waiting;
218 ldlm_mode_t lr_most_restr;
219 __u32 lr_type; /* LDLM_PLAIN or LDLM_EXTENT */
220 struct ldlm_resource *lr_root;
221 struct ldlm_res_id lr_name;
222 __u32 lr_version[RES_VERSION_SIZE];
223 atomic_t lr_refcount;
225 /* lr_tmp holds a list head temporarily, during the building of a work
226 * queue. see ldlm_add_ast_work_item and ldlm_run_ast_work */
230 struct ldlm_ast_work {
231 struct ldlm_lock *w_lock;
233 struct ldlm_lock_desc w_desc;
234 struct list_head w_list;
240 extern struct obd_ops ldlm_obd_ops;
242 extern char *ldlm_lockname[];
243 extern char *ldlm_typename[];
244 extern char *ldlm_it2str(int it);
246 #define __LDLM_DEBUG(level, lock, format, a...) \
248 if (lock->l_resource == NULL) { \
249 CDEBUG(level, "### " format \
250 " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s "\
251 "res: \?\? rrc=\?\? type: \?\?\? remote: " \
252 LPX64"\n" , ## a, lock, lock->l_handle.h_cookie, \
253 atomic_read(&lock->l_refc), \
254 lock->l_readers, lock->l_writers, \
255 ldlm_lockname[lock->l_granted_mode], \
256 ldlm_lockname[lock->l_req_mode], \
257 lock->l_remote_handle.cookie); \
260 if (lock->l_resource->lr_type == LDLM_EXTENT) { \
261 CDEBUG(level, "### " format \
262 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
263 "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64\
264 "] remote: "LPX64"\n" , ## a, \
265 lock->l_resource->lr_namespace->ns_name, lock, \
266 lock->l_handle.h_cookie, atomic_read(&lock->l_refc), \
267 lock->l_readers, lock->l_writers, \
268 ldlm_lockname[lock->l_granted_mode], \
269 ldlm_lockname[lock->l_req_mode], \
270 lock->l_resource->lr_name.name[0], \
271 lock->l_resource->lr_name.name[1], \
272 atomic_read(&lock->l_resource->lr_refcount), \
273 ldlm_typename[lock->l_resource->lr_type], \
274 lock->l_extent.start, lock->l_extent.end, \
275 lock->l_remote_handle.cookie); \
279 CDEBUG(level, "### " format \
280 " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " \
281 "res: "LPU64"/"LPU64" rrc: %d type: %s remote: "LPX64 \
283 lock->l_resource->lr_namespace->ns_name, \
284 lock, lock->l_handle.h_cookie, \
285 atomic_read (&lock->l_refc), \
286 lock->l_readers, lock->l_writers, \
287 ldlm_lockname[lock->l_granted_mode], \
288 ldlm_lockname[lock->l_req_mode], \
289 lock->l_resource->lr_name.name[0], \
290 lock->l_resource->lr_name.name[1], \
291 atomic_read(&lock->l_resource->lr_refcount), \
292 ldlm_typename[lock->l_resource->lr_type], \
293 lock->l_remote_handle.cookie); \
297 #define LDLM_DEBUG(lock, format, a...) __LDLM_DEBUG(D_DLMTRACE, lock, \
299 #define LDLM_ERROR(lock, format, a...) __LDLM_DEBUG(D_ERROR, lock, format, ## a)
301 #define LDLM_DEBUG_NOLOCK(format, a...) \
302 CDEBUG(D_DLMTRACE, "### " format "\n" , ## a)
308 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
309 #define LDLM_ITER_STOP 0 /* stop iterating */
311 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
312 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
314 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
316 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
318 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
319 ldlm_res_iterator_t iter, void *closure);
321 int ldlm_replay_locks(struct obd_import *imp);
322 void ldlm_change_cbdata(struct ldlm_namespace *, struct ldlm_res_id *,
323 ldlm_iterator_t iter, void *data);
326 int ldlm_extent_compat(struct ldlm_lock *, struct ldlm_lock *);
327 int ldlm_extent_policy(struct ldlm_namespace *, struct ldlm_lock **, void *,
328 ldlm_mode_t, int flags, void *);
331 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
332 void *data, int flag);
333 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
334 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
335 ldlm_blocking_callback);
336 int ldlm_handle_convert(struct ptlrpc_request *req);
337 int ldlm_handle_cancel(struct ptlrpc_request *req);
338 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
341 void ldlm_register_intent(ldlm_res_policy arg);
342 void ldlm_unregister_intent(void);
343 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh);
344 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *, int flags);
345 void ldlm_cancel_callback(struct ldlm_lock *);
346 int ldlm_lock_set_data(struct lustre_handle *, void *data);
347 void ldlm_lock_remove_from_lru(struct ldlm_lock *);
348 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
349 struct lustre_handle *);
351 static inline struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *h)
353 return __ldlm_handle2lock(h, 0);
356 #define LDLM_LOCK_PUT(lock) \
358 /*LDLM_DEBUG((lock), "put");*/ \
359 ldlm_lock_put(lock); \
362 #define LDLM_LOCK_GET(lock) \
364 ldlm_lock_get(lock); \
365 /*LDLM_DEBUG((lock), "get");*/ \
369 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
370 void ldlm_lock_put(struct ldlm_lock *lock);
371 void ldlm_lock_destroy(struct ldlm_lock *lock);
372 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
373 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
374 void ldlm_lock_addref_internal(struct ldlm_lock *, __u32 mode);
375 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
376 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
377 void ldlm_grant_lock(struct ldlm_lock *lock, void *data, int datalen);
378 int ldlm_lock_match(struct ldlm_namespace *ns, int flags, struct ldlm_res_id *,
379 __u32 type, void *cookie, int cookielen, ldlm_mode_t mode,
380 void *data, struct lustre_handle *);
382 ldlm_lock_create(struct ldlm_namespace *ns,
383 struct lustre_handle *parent_lock_handle, struct ldlm_res_id,
384 __u32 type, ldlm_mode_t, ldlm_blocking_callback,
386 ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *, struct ldlm_lock **,
387 void *cookie, int cookie_len, int *flags,
388 ldlm_completion_callback completion);
389 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
391 void ldlm_lock_cancel(struct ldlm_lock *lock);
392 void ldlm_cancel_locks_for_export(struct obd_export *export);
393 int ldlm_run_ast_work(struct list_head *rpc_list);
394 void ldlm_reprocess_all(struct ldlm_resource *res);
395 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
396 void ldlm_lock_dump(int level, struct ldlm_lock *lock);
397 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
400 int ldlm_test(struct obd_device *device, struct lustre_handle *connh);
401 int ldlm_regression_start(struct obd_device *obddev,
402 struct lustre_handle *connh,
403 unsigned int threads, unsigned int max_locks_in,
404 unsigned int num_resources_in,
405 unsigned int num_extents_in);
406 int ldlm_regression_stop(void);
410 struct ldlm_namespace *ldlm_namespace_new(char *name, __u32 local);
411 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int local_only);
412 int ldlm_namespace_free(struct ldlm_namespace *ns);
413 int ldlm_proc_setup(struct obd_device *obd);
414 void ldlm_proc_cleanup(struct obd_device *obd);
416 /* resource.c - internal */
417 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
418 struct ldlm_resource *parent,
419 struct ldlm_res_id, __u32 type,
421 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
422 int ldlm_resource_putref(struct ldlm_resource *res);
423 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
424 struct ldlm_lock *lock);
425 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
426 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
427 void ldlm_dump_all_namespaces(void);
428 void ldlm_namespace_dump(struct ldlm_namespace *);
429 void ldlm_resource_dump(struct ldlm_resource *);
430 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
434 int ldlm_expired_completion_wait(void *data);
435 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
436 int ldlm_cli_enqueue(struct lustre_handle *conn,
437 struct ptlrpc_request *req,
438 struct ldlm_namespace *ns,
439 struct lustre_handle *parent_lock_handle,
442 void *cookie, int cookielen,
445 ldlm_completion_callback completion,
446 ldlm_blocking_callback callback,
448 struct lustre_handle *lockh);
449 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
450 void *data, __u32 data_len);
451 int ldlm_cli_convert(struct lustre_handle *, int new_mode, int *flags);
452 int ldlm_cli_cancel(struct lustre_handle *lockh);
453 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
454 int flags, void *opaque);
457 /* This has to be here because recurisve inclusion sucks. */
458 int intent_disposition(struct ldlm_reply *rep, int flag);
459 void intent_set_disposition(struct ldlm_reply *rep, int flag);
460 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
461 void *data, int flag);
464 /* ioctls for trying requests */
465 #define IOC_LDLM_TYPE 'f'
466 #define IOC_LDLM_MIN_NR 40
468 #define IOC_LDLM_TEST _IOWR('f', 40, long)
469 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
470 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
471 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
472 #define IOC_LDLM_MAX_NR 43