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. */
74 struct osc_setattr_cbargs {
76 cfs_completion_t opc_sync;
81 * State of transfer for osc.
84 struct cl_req_slice or_cl;
88 * State maintained by osc layer for the duration of a system call.
94 struct osc_thread_info {
95 struct ldlm_res_id oti_resname;
96 ldlm_policy_data_t oti_policy;
97 struct cl_lock_descr oti_descr;
98 struct cl_attr oti_attr;
99 struct lustre_handle oti_handle;
100 struct cl_page_list oti_plist;
104 * Manage osc_async_page
106 struct osc_oap_pages {
107 cfs_list_t oop_pending;
108 cfs_list_t oop_urgent;
112 static inline void osc_oap_pages_init(struct osc_oap_pages *list)
114 CFS_INIT_LIST_HEAD(&list->oop_pending);
115 CFS_INIT_LIST_HEAD(&list->oop_urgent);
116 list->oop_num_pending = 0;
120 struct cl_object oo_cl;
121 struct lov_oinfo *oo_oinfo;
123 * True if locking against this stripe got -EUSERS.
126 cfs_time_t oo_contention_time;
127 #ifdef INVARIANT_CHECK
129 * IO context used for invariant checks in osc_lock_has_pages().
131 struct cl_io oo_debug_io;
132 /** Serialization object for osc_object::oo_debug_io. */
133 cfs_mutex_t oo_debug_mutex;
136 * List of pages in transfer.
138 cfs_list_t oo_inflight[CRT_NR];
140 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
141 * locked during take-off and landing.
143 cfs_spinlock_t oo_seatbelt;
146 * used by the osc to keep track of what objects to build into rpcs
148 struct osc_oap_pages oo_read_pages;
149 struct osc_oap_pages oo_write_pages;
150 cfs_list_t oo_ready_item;
151 cfs_list_t oo_hp_ready_item;
152 cfs_list_t oo_write_item;
153 cfs_list_t oo_read_item;
157 * Lock "micro-states" for osc layer.
159 enum osc_lock_state {
170 * osc-private state of cl_lock.
172 * Interaction with DLM.
174 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
176 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
177 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
179 * This pointer is protected through a reference, acquired by
180 * osc_lock_upcall0(). Also, an additional reference is acquired by
181 * ldlm_lock_addref() call protecting the lock from cancellation, until
182 * osc_lock_unuse() releases it.
184 * Below is a description of how lock references are acquired and released
187 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
188 * - ldlm_lock_create()
189 * - ldlm_lock_new(): initializes a lock with 2 references. One for
190 * the caller (released when reply from the server is received, or on
191 * error), and another for the hash table.
192 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
194 * - When reply is received from the server (osc_enqueue_interpret())
195 * - ldlm_cli_enqueue_fini()
196 * - LDLM_LOCK_PUT(): releases caller reference acquired by
199 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
200 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
202 * - When lock is being cancelled (ldlm_lock_cancel())
203 * - ldlm_lock_destroy()
204 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
207 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
208 * either when lock is cancelled (osc_lock_blocking()), or when locks is
209 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
210 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
214 struct cl_lock_slice ols_cl;
215 /** underlying DLM lock */
216 struct ldlm_lock *ols_lock;
217 /** lock value block */
218 struct ost_lvb ols_lvb;
219 /** DLM flags with which osc_lock::ols_lock was enqueued */
221 /** osc_lock::ols_lock handle */
222 struct lustre_handle ols_handle;
223 struct ldlm_enqueue_info ols_einfo;
224 enum osc_lock_state ols_state;
227 * How many pages are using this lock for io, currently only used by
228 * read-ahead. If non-zero, the underlying dlm lock won't be cancelled
229 * during recovery to avoid deadlock. see bz16774.
231 * \see osc_page::ops_lock
232 * \see osc_page_addref_lock(), osc_page_putref_lock()
234 cfs_atomic_t ols_pageref;
237 * true, if ldlm_lock_addref() was called against
238 * osc_lock::ols_lock. This is used for sanity checking.
240 * \see osc_lock::ols_has_ref
242 unsigned ols_hold :1,
244 * this is much like osc_lock::ols_hold, except that this bit is
245 * cleared _after_ reference in released in osc_lock_unuse(). This
246 * fine distinction is needed because:
248 * - if ldlm lock still has a reference, osc_ast_data_get() needs
249 * to return associated cl_lock (so that a flag is needed that is
250 * cleared after ldlm_lock_decref() returned), and
252 * - ldlm_lock_decref() can invoke blocking ast (for a
253 * LDLM_FL_CBPENDING lock), and osc_lock functions like
254 * osc_lock_cancel() called from there need to know whether to
255 * release lock reference (so that a flag is needed that is
256 * cleared before ldlm_lock_decref() is called).
260 * inherit the lockless attribute from top level cl_io.
261 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
265 * set by osc_lock_use() to wait until blocking AST enters into
266 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
267 * further synchronization.
271 * If the data of this lock has been flushed to server side.
275 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
276 * the EVAVAIL error as torerable, this will make upper logic happy
277 * to wait all glimpse locks to each OSTs to be completed.
278 * Glimpse lock converts to normal lock if the server lock is
280 * Glimpse lock should be destroyed immediately after use.
284 * For async glimpse lock.
288 * IO that owns this lock. This field is used for a dead-lock
289 * avoidance by osc_lock_enqueue_wait().
291 * XXX: unfortunately, the owner of a osc_lock is not unique,
292 * the lock may have multiple users, if the lock is granted and
295 struct osc_io *ols_owner;
300 * Page state private for osc layer.
303 struct cl_page_slice ops_cl;
305 * Page queues used by osc to detect when RPC can be formed.
307 struct osc_async_page ops_oap;
309 * An offset within page from which next transfer starts. This is used
310 * by cl_page_clip() to submit partial page transfers.
314 * An offset within page at which next transfer ends.
316 * \see osc_page::ops_from.
320 * Boolean, true iff page is under transfer. Used for sanity checking.
322 unsigned ops_transfer_pinned:1,
324 * True for a `temporary page' created by read-ahead code, probably
325 * outside of any DLM lock.
329 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
333 * Linkage into a per-osc_object list of pages in flight. For
336 cfs_list_t ops_inflight;
338 * Thread that submitted this page for transfer. For debugging.
340 cfs_task_t *ops_submitter;
342 * Submit time - the time when the page is starting RPC. For debugging.
344 cfs_time_t ops_submit_time;
347 * A lock of which we hold a reference covers this page. Only used by
348 * read-ahead: for a readahead page, we hold it's covering lock to
349 * prevent it from being canceled during recovery.
351 * \see osc_lock::ols_pageref
352 * \see osc_page_addref_lock(), osc_page_putref_lock().
354 struct cl_lock *ops_lock;
357 extern cfs_mem_cache_t *osc_page_kmem;
358 extern cfs_mem_cache_t *osc_lock_kmem;
359 extern cfs_mem_cache_t *osc_object_kmem;
360 extern cfs_mem_cache_t *osc_thread_kmem;
361 extern cfs_mem_cache_t *osc_session_kmem;
362 extern cfs_mem_cache_t *osc_req_kmem;
364 extern struct lu_device_type osc_device_type;
365 extern struct lu_context_key osc_key;
366 extern struct lu_context_key osc_session_key;
368 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
370 int osc_lock_init(const struct lu_env *env,
371 struct cl_object *obj, struct cl_lock *lock,
372 const struct cl_io *io);
373 int osc_io_init (const struct lu_env *env,
374 struct cl_object *obj, struct cl_io *io);
375 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
377 struct lu_object *osc_object_alloc(const struct lu_env *env,
378 const struct lu_object_header *hdr,
379 struct lu_device *dev);
380 struct cl_page *osc_page_init (const struct lu_env *env,
381 struct cl_object *obj,
382 struct cl_page *page, cfs_page_t *vmpage);
384 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
385 struct ldlm_res_id *resname);
386 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
387 pgoff_t start, pgoff_t end);
388 int osc_lvb_print (const struct lu_env *env, void *cookie,
389 lu_printer_t p, const struct ost_lvb *lvb);
391 void osc_io_submit_page(const struct lu_env *env,
392 struct osc_io *oio, struct osc_page *opg,
393 enum cl_req_type crt);
394 void osc_ap_completion(const struct lu_env *env, struct client_obd *cli,
395 struct obdo *oa, struct osc_async_page *oap,
397 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops);
398 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
399 obd_flag async_flags);
400 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
401 cfs_page_t *page, loff_t offset);
402 int osc_queue_async_io(const struct lu_env *env, struct osc_page *ops);
403 int osc_teardown_async_page(struct osc_object *obj,
404 struct osc_page *ops);
405 int osc_queue_sync_page(const struct lu_env *env, struct osc_page *ops,
406 int cmd, int brw_flags);
407 void osc_io_unplug(const struct lu_env *env, struct client_obd *cli,
408 struct osc_object *osc, pdl_policy_t pol);
410 void osc_object_set_contended (struct osc_object *obj);
411 void osc_object_clear_contended(struct osc_object *obj);
412 int osc_object_is_contended (struct osc_object *obj);
414 int osc_lock_is_lockless (const struct osc_lock *olck);
416 /*****************************************************************************
422 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
424 struct osc_thread_info *info;
426 info = lu_context_key_get(&env->le_ctx, &osc_key);
427 LASSERT(info != NULL);
431 static inline struct osc_session *osc_env_session(const struct lu_env *env)
433 struct osc_session *ses;
435 ses = lu_context_key_get(env->le_ses, &osc_session_key);
436 LASSERT(ses != NULL);
440 static inline struct osc_io *osc_env_io(const struct lu_env *env)
442 return &osc_env_session(env)->os_io;
445 static inline int osc_is_object(const struct lu_object *obj)
447 return obj->lo_dev->ld_type == &osc_device_type;
450 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
452 LINVRNT(d->ld_type == &osc_device_type);
453 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
456 static inline struct obd_export *osc_export(const struct osc_object *obj)
458 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
461 static inline struct osc_object *cl2osc(const struct cl_object *obj)
463 LINVRNT(osc_is_object(&obj->co_lu));
464 return container_of0(obj, struct osc_object, oo_cl);
467 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
469 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
470 if (mode == CLM_READ)
472 else if (mode == CLM_WRITE)
478 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
480 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
483 else if (mode == LCK_PW)
489 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
491 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
492 return container_of0(slice, struct osc_page, ops_cl);
495 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
497 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
498 return container_of0(slice, struct osc_lock, ols_cl);
501 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
503 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
506 static inline int osc_io_srvlock(struct osc_io *oio)
508 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
513 #endif /* OSC_CL_INTERNAL_H */