4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 * This file is part of Lustre, http://www.lustre.org/
38 * Lustre is a trademark of Sun Microsystems, Inc.
40 * Internal interfaces of OSC layer.
42 * Author: Nikita Danilov <nikita.danilov@sun.com>
45 #ifndef OSC_CL_INTERNAL_H
46 #define OSC_CL_INTERNAL_H
49 # include <libcfs/libcfs.h>
51 # include <liblustre.h>
55 /* osc_build_res_name() */
57 #include <cl_object.h>
58 #include "osc_internal.h"
65 * State maintained by osc layer for each IO context.
69 struct cl_io_slice oi_cl;
70 /** true if this io is lockless. */
73 struct obd_info oi_info;
75 struct osc_async_cbargs {
77 cfs_completion_t opc_sync;
82 * State of transfer for osc.
85 struct cl_req_slice or_cl;
89 * State maintained by osc layer for the duration of a system call.
95 struct osc_thread_info {
96 struct ldlm_res_id oti_resname;
97 ldlm_policy_data_t oti_policy;
98 struct cl_lock_descr oti_descr;
99 struct cl_attr oti_attr;
100 struct lustre_handle oti_handle;
101 struct cl_page_list oti_plist;
105 * Manage osc_async_page
107 struct osc_oap_pages {
108 cfs_list_t oop_pending;
109 cfs_list_t oop_urgent;
113 static inline void osc_oap_pages_init(struct osc_oap_pages *list)
115 CFS_INIT_LIST_HEAD(&list->oop_pending);
116 CFS_INIT_LIST_HEAD(&list->oop_urgent);
117 list->oop_num_pending = 0;
121 struct cl_object oo_cl;
122 struct lov_oinfo *oo_oinfo;
124 * True if locking against this stripe got -EUSERS.
127 cfs_time_t oo_contention_time;
128 #ifdef INVARIANT_CHECK
130 * IO context used for invariant checks in osc_lock_has_pages().
132 struct cl_io oo_debug_io;
133 /** Serialization object for osc_object::oo_debug_io. */
134 cfs_mutex_t oo_debug_mutex;
137 * List of pages in transfer.
139 cfs_list_t oo_inflight[CRT_NR];
141 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
142 * locked during take-off and landing.
144 cfs_spinlock_t oo_seatbelt;
147 * used by the osc to keep track of what objects to build into rpcs
149 struct osc_oap_pages oo_read_pages;
150 struct osc_oap_pages oo_write_pages;
151 cfs_list_t oo_ready_item;
152 cfs_list_t oo_hp_ready_item;
153 cfs_list_t oo_write_item;
154 cfs_list_t oo_read_item;
158 * Lock "micro-states" for osc layer.
160 enum osc_lock_state {
171 * osc-private state of cl_lock.
173 * Interaction with DLM.
175 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
177 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
178 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
180 * This pointer is protected through a reference, acquired by
181 * osc_lock_upcall0(). Also, an additional reference is acquired by
182 * ldlm_lock_addref() call protecting the lock from cancellation, until
183 * osc_lock_unuse() releases it.
185 * Below is a description of how lock references are acquired and released
188 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
189 * - ldlm_lock_create()
190 * - ldlm_lock_new(): initializes a lock with 2 references. One for
191 * the caller (released when reply from the server is received, or on
192 * error), and another for the hash table.
193 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
195 * - When reply is received from the server (osc_enqueue_interpret())
196 * - ldlm_cli_enqueue_fini()
197 * - LDLM_LOCK_PUT(): releases caller reference acquired by
200 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
201 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
203 * - When lock is being cancelled (ldlm_lock_cancel())
204 * - ldlm_lock_destroy()
205 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
208 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
209 * either when lock is cancelled (osc_lock_blocking()), or when locks is
210 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
211 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
215 struct cl_lock_slice ols_cl;
216 /** underlying DLM lock */
217 struct ldlm_lock *ols_lock;
218 /** lock value block */
219 struct ost_lvb ols_lvb;
220 /** DLM flags with which osc_lock::ols_lock was enqueued */
222 /** osc_lock::ols_lock handle */
223 struct lustre_handle ols_handle;
224 struct ldlm_enqueue_info ols_einfo;
225 enum osc_lock_state ols_state;
228 * How many pages are using this lock for io, currently only used by
229 * read-ahead. If non-zero, the underlying dlm lock won't be cancelled
230 * during recovery to avoid deadlock. see bz16774.
232 * \see osc_page::ops_lock
233 * \see osc_page_addref_lock(), osc_page_putref_lock()
235 cfs_atomic_t ols_pageref;
238 * true, if ldlm_lock_addref() was called against
239 * osc_lock::ols_lock. This is used for sanity checking.
241 * \see osc_lock::ols_has_ref
243 unsigned ols_hold :1,
245 * this is much like osc_lock::ols_hold, except that this bit is
246 * cleared _after_ reference in released in osc_lock_unuse(). This
247 * fine distinction is needed because:
249 * - if ldlm lock still has a reference, osc_ast_data_get() needs
250 * to return associated cl_lock (so that a flag is needed that is
251 * cleared after ldlm_lock_decref() returned), and
253 * - ldlm_lock_decref() can invoke blocking ast (for a
254 * LDLM_FL_CBPENDING lock), and osc_lock functions like
255 * osc_lock_cancel() called from there need to know whether to
256 * release lock reference (so that a flag is needed that is
257 * cleared before ldlm_lock_decref() is called).
261 * inherit the lockless attribute from top level cl_io.
262 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
266 * set by osc_lock_use() to wait until blocking AST enters into
267 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
268 * further synchronization.
272 * If the data of this lock has been flushed to server side.
276 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
277 * the EVAVAIL error as torerable, this will make upper logic happy
278 * to wait all glimpse locks to each OSTs to be completed.
279 * Glimpse lock converts to normal lock if the server lock is
281 * Glimpse lock should be destroyed immediately after use.
285 * For async glimpse lock.
289 * IO that owns this lock. This field is used for a dead-lock
290 * avoidance by osc_lock_enqueue_wait().
292 * XXX: unfortunately, the owner of a osc_lock is not unique,
293 * the lock may have multiple users, if the lock is granted and
296 struct osc_io *ols_owner;
301 * Page state private for osc layer.
304 struct cl_page_slice ops_cl;
306 * Page queues used by osc to detect when RPC can be formed.
308 struct osc_async_page ops_oap;
310 * An offset within page from which next transfer starts. This is used
311 * by cl_page_clip() to submit partial page transfers.
315 * An offset within page at which next transfer ends.
317 * \see osc_page::ops_from.
321 * Boolean, true iff page is under transfer. Used for sanity checking.
323 unsigned ops_transfer_pinned:1,
325 * True for a `temporary page' created by read-ahead code, probably
326 * outside of any DLM lock.
330 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
334 * Linkage into a per-osc_object list of pages in flight. For
337 cfs_list_t ops_inflight;
339 * Thread that submitted this page for transfer. For debugging.
341 cfs_task_t *ops_submitter;
343 * Submit time - the time when the page is starting RPC. For debugging.
345 cfs_time_t ops_submit_time;
348 * A lock of which we hold a reference covers this page. Only used by
349 * read-ahead: for a readahead page, we hold it's covering lock to
350 * prevent it from being canceled during recovery.
352 * \see osc_lock::ols_pageref
353 * \see osc_page_addref_lock(), osc_page_putref_lock().
355 struct cl_lock *ops_lock;
358 extern cfs_mem_cache_t *osc_page_kmem;
359 extern cfs_mem_cache_t *osc_lock_kmem;
360 extern cfs_mem_cache_t *osc_object_kmem;
361 extern cfs_mem_cache_t *osc_thread_kmem;
362 extern cfs_mem_cache_t *osc_session_kmem;
363 extern cfs_mem_cache_t *osc_req_kmem;
365 extern struct lu_device_type osc_device_type;
366 extern struct lu_context_key osc_key;
367 extern struct lu_context_key osc_session_key;
369 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
371 int osc_lock_init(const struct lu_env *env,
372 struct cl_object *obj, struct cl_lock *lock,
373 const struct cl_io *io);
374 int osc_io_init (const struct lu_env *env,
375 struct cl_object *obj, struct cl_io *io);
376 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
378 struct lu_object *osc_object_alloc(const struct lu_env *env,
379 const struct lu_object_header *hdr,
380 struct lu_device *dev);
381 struct cl_page *osc_page_init (const struct lu_env *env,
382 struct cl_object *obj,
383 struct cl_page *page, cfs_page_t *vmpage);
385 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
386 struct ldlm_res_id *resname);
387 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
388 pgoff_t start, pgoff_t end);
389 int osc_lvb_print (const struct lu_env *env, void *cookie,
390 lu_printer_t p, const struct ost_lvb *lvb);
392 void osc_io_submit_page(const struct lu_env *env,
393 struct osc_io *oio, struct osc_page *opg,
394 enum cl_req_type crt);
395 void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
396 struct obdo *oa, struct osc_async_page *oap,
398 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops);
399 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
400 obd_flag async_flags);
401 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
402 cfs_page_t *page, loff_t offset);
403 int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops);
404 int osc_teardown_async_page(struct osc_object *obj,
405 struct osc_page *ops);
406 int osc_queue_sync_page(const struct lu_env *env, struct osc_page *ops,
407 int cmd, int brw_flags);
408 void osc_io_unplug(const struct lu_env *env, struct client_obd *cli,
409 struct osc_object *osc, pdl_policy_t pol);
411 void osc_object_set_contended (struct osc_object *obj);
412 void osc_object_clear_contended(struct osc_object *obj);
413 int osc_object_is_contended (struct osc_object *obj);
415 int osc_lock_is_lockless (const struct osc_lock *olck);
417 /*****************************************************************************
423 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
425 struct osc_thread_info *info;
427 info = lu_context_key_get(&env->le_ctx, &osc_key);
428 LASSERT(info != NULL);
432 static inline struct osc_session *osc_env_session(const struct lu_env *env)
434 struct osc_session *ses;
436 ses = lu_context_key_get(env->le_ses, &osc_session_key);
437 LASSERT(ses != NULL);
441 static inline struct osc_io *osc_env_io(const struct lu_env *env)
443 return &osc_env_session(env)->os_io;
446 static inline int osc_is_object(const struct lu_object *obj)
448 return obj->lo_dev->ld_type == &osc_device_type;
451 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
453 LINVRNT(d->ld_type == &osc_device_type);
454 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
457 static inline struct obd_export *osc_export(const struct osc_object *obj)
459 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
462 static inline struct osc_object *cl2osc(const struct cl_object *obj)
464 LINVRNT(osc_is_object(&obj->co_lu));
465 return container_of0(obj, struct osc_object, oo_cl);
468 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
470 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
471 if (mode == CLM_READ)
473 else if (mode == CLM_WRITE)
479 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
481 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
484 else if (mode == LCK_PW)
490 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
492 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
493 return container_of0(slice, struct osc_page, ops_cl);
496 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
498 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
499 return container_of0(slice, struct osc_lock, ols_cl);
502 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
504 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
507 static inline int osc_io_srvlock(struct osc_io *oio)
509 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
514 #endif /* OSC_CL_INTERNAL_H */