1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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"
60 /** \addtogroup osc osc @{ */
63 * State maintained by osc layer for each IO context.
67 struct cl_io_slice oi_cl;
68 /** true if this io is lockless. */
72 struct osc_punch_cbargs {
74 struct completion opc_sync;
79 * State of transfer for osc.
82 struct cl_req_slice or_cl;
86 * State maintained by osc layer for the duration of a system call.
92 struct osc_thread_info {
93 struct ldlm_res_id oti_resname;
94 ldlm_policy_data_t oti_policy;
95 struct cl_lock_descr oti_descr;
96 struct cl_attr oti_attr;
97 struct lustre_handle oti_handle;
98 struct cl_lock_closure oti_closure;
99 struct cl_page_list oti_plist;
103 struct cl_object oo_cl;
104 struct lov_oinfo *oo_oinfo;
106 * True if locking against this stripe got -EUSERS.
109 cfs_time_t oo_contention_time;
110 #ifdef INVARIANT_CHECK
112 * IO context used for invariant checks in osc_lock_has_pages().
114 struct cl_io oo_debug_io;
115 /** Serialization object for osc_object::oo_debug_io. */
116 struct mutex oo_debug_mutex;
119 * List of pages in transfer.
121 struct list_head oo_inflight[CRT_NR];
123 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
124 * locked during take-off and landing.
126 spinlock_t oo_seatbelt;
130 * Lock "micro-states" for osc layer.
132 enum osc_lock_state {
143 * osc-private state of cl_lock.
145 * Interaction with DLM.
147 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
149 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
150 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
152 * This pointer is protected through a reference, acquired by
153 * osc_lock_upcall0(). Also, an additional reference is acquired by
154 * ldlm_lock_addref() call protecting the lock from cancellation, until
155 * osc_lock_unuse() releases it.
157 * Below is a description of how lock references are acquired and released
160 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
161 * - ldlm_lock_create()
162 * - ldlm_lock_new(): initializes a lock with 2 references. One for
163 * the caller (released when reply from the server is received, or on
164 * error), and another for the hash table.
165 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
167 * - When reply is received from the server (osc_enqueue_interpret())
168 * - ldlm_cli_enqueue_fini()
169 * - LDLM_LOCK_PUT(): releases caller reference acquired by
172 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
173 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
175 * - When lock is being cancelled (ldlm_lock_cancel())
176 * - ldlm_lock_destroy()
177 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
180 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
181 * either when lock is cancelled (osc_lock_blocking()), or when locks is
182 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
183 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
187 struct cl_lock_slice ols_cl;
188 /** underlying DLM lock */
189 struct ldlm_lock *ols_lock;
190 /** lock value block */
191 struct ost_lvb ols_lvb;
192 /** DLM flags with which osc_lock::ols_lock was enqueued */
194 /** osc_lock::ols_lock handle */
195 struct lustre_handle ols_handle;
196 struct ldlm_enqueue_info ols_einfo;
197 enum osc_lock_state ols_state;
199 * true, if ldlm_lock_addref() was called against
200 * osc_lock::ols_lock. This is used for sanity checking.
202 * \see osc_lock::ols_has_ref
204 unsigned ols_hold :1,
206 * this is much like osc_lock::ols_hold, except that this bit is
207 * cleared _after_ reference in released in osc_lock_unuse(). This
208 * fine distinction is needed because:
210 * - if ldlm lock still has a reference, osc_ast_data_get() needs
211 * to return associated cl_lock (so that a flag is needed that is
212 * cleared after ldlm_lock_decref() returned), and
214 * - ldlm_lock_decref() can invoke blocking ast (for a
215 * LDLM_FL_CBPENDING lock), and osc_lock functions like
216 * osc_lock_cancel() called from there need to know whether to
217 * release lock reference (so that a flag is needed that is
218 * cleared before ldlm_lock_decref() is called).
222 * inherit the lockless attribute from top level cl_io.
223 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
227 * set by osc_lock_use() to wait until blocking AST enters into
228 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
229 * further synchronization.
233 * If the data of this lock has been flushed to server side.
237 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
238 * the EVAVAIL error as torerable, this will make upper logic happy
239 * to wait all glimpse locks to each OSTs to be completed.
240 * Glimpse lock converts to normal lock if the server lock is
242 * Glimpse lock should be destroyed immediately after use.
246 * IO that owns this lock. This field is used for a dead-lock
247 * avoidance by osc_lock_enqueue().
249 * \see osc_deadlock_is_possible()
251 struct osc_io *ols_owner;
256 * Page state private for osc layer.
259 struct cl_page_slice ops_cl;
261 * Page queues used by osc to detect when RPC can be formed.
263 struct osc_async_page ops_oap;
265 * An offset within page from which next transfer starts. This is used
266 * by cl_page_clip() to submit partial page transfers.
270 * An offset within page at which next transfer ends.
272 * \see osc_page::ops_from.
276 * Boolean, true iff page is under transfer. Used for sanity checking.
278 unsigned ops_transfer_pinned:1,
280 * True for a `temporary page' created by read-ahead code, probably
281 * outside of any DLM lock.
285 * Linkage into a per-osc_object list of pages in flight. For
288 struct list_head ops_inflight;
290 * Thread that submitted this page for transfer. For debugging.
292 cfs_task_t *ops_submitter;
295 extern cfs_mem_cache_t *osc_page_kmem;
296 extern cfs_mem_cache_t *osc_lock_kmem;
297 extern cfs_mem_cache_t *osc_object_kmem;
298 extern cfs_mem_cache_t *osc_thread_kmem;
299 extern cfs_mem_cache_t *osc_session_kmem;
300 extern cfs_mem_cache_t *osc_req_kmem;
302 extern struct lu_device_type osc_device_type;
303 extern struct lu_context_key osc_key;
304 extern struct lu_context_key osc_session_key;
306 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
308 int osc_lock_init(const struct lu_env *env,
309 struct cl_object *obj, struct cl_lock *lock,
310 const struct cl_io *io);
311 int osc_io_init (const struct lu_env *env,
312 struct cl_object *obj, struct cl_io *io);
313 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
315 struct lu_object *osc_object_alloc(const struct lu_env *env,
316 const struct lu_object_header *hdr,
317 struct lu_device *dev);
318 struct cl_page *osc_page_init (const struct lu_env *env,
319 struct cl_object *obj,
320 struct cl_page *page, cfs_page_t *vmpage);
322 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
323 struct ldlm_res_id *resname);
324 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
325 pgoff_t start, pgoff_t end);
326 int osc_lvb_print (const struct lu_env *env, void *cookie,
327 lu_printer_t p, const struct ost_lvb *lvb);
328 void osc_io_submit_page(const struct lu_env *env,
329 struct osc_io *oio, struct osc_page *opg,
330 enum cl_req_type crt);
332 void osc_object_set_contended (struct osc_object *obj);
333 void osc_object_clear_contended(struct osc_object *obj);
334 int osc_object_is_contended (struct osc_object *obj);
336 int osc_lock_is_lockless (const struct osc_lock *olck);
338 /*****************************************************************************
344 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
346 struct osc_thread_info *info;
348 info = lu_context_key_get(&env->le_ctx, &osc_key);
349 LASSERT(info != NULL);
353 static inline struct osc_session *osc_env_session(const struct lu_env *env)
355 struct osc_session *ses;
357 ses = lu_context_key_get(env->le_ses, &osc_session_key);
358 LASSERT(ses != NULL);
362 static inline struct osc_io *osc_env_io(const struct lu_env *env)
364 return &osc_env_session(env)->os_io;
367 static inline int osc_is_object(const struct lu_object *obj)
369 return obj->lo_dev->ld_type == &osc_device_type;
372 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
374 LINVRNT(d->ld_type == &osc_device_type);
375 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
378 static inline struct obd_export *osc_export(const struct osc_object *obj)
380 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
383 static inline struct osc_object *cl2osc(const struct cl_object *obj)
385 LINVRNT(osc_is_object(&obj->co_lu));
386 return container_of0(obj, struct osc_object, oo_cl);
389 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
391 LASSERT(mode == CLM_READ || mode == CLM_WRITE);
392 return mode == CLM_READ ? LCK_PR : LCK_PW;
395 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
397 LASSERT(mode == LCK_PR || mode == LCK_PW);
398 return mode == LCK_PR ? CLM_READ : CLM_WRITE;
401 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
403 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
404 return container_of0(slice, struct osc_page, ops_cl);
407 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
409 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
410 return container_of0(slice, struct osc_lock, ols_cl);
413 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
415 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
420 #endif /* OSC_CL_INTERNAL_H */