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"
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_punch_cbargs {
76 struct completion 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_lock_closure oti_closure;
101 struct cl_page_list oti_plist;
105 struct cl_object oo_cl;
106 struct lov_oinfo *oo_oinfo;
108 * True if locking against this stripe got -EUSERS.
111 cfs_time_t oo_contention_time;
112 #ifdef INVARIANT_CHECK
114 * IO context used for invariant checks in osc_lock_has_pages().
116 struct cl_io oo_debug_io;
117 /** Serialization object for osc_object::oo_debug_io. */
118 struct mutex oo_debug_mutex;
121 * List of pages in transfer.
123 struct list_head oo_inflight[CRT_NR];
125 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
126 * locked during take-off and landing.
128 spinlock_t oo_seatbelt;
132 * Lock "micro-states" for osc layer.
134 enum osc_lock_state {
145 * osc-private state of cl_lock.
147 * Interaction with DLM.
149 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
151 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
152 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
154 * This pointer is protected through a reference, acquired by
155 * osc_lock_upcall0(). Also, an additional reference is acquired by
156 * ldlm_lock_addref() call protecting the lock from cancellation, until
157 * osc_lock_unuse() releases it.
159 * Below is a description of how lock references are acquired and released
162 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
163 * - ldlm_lock_create()
164 * - ldlm_lock_new(): initializes a lock with 2 references. One for
165 * the caller (released when reply from the server is received, or on
166 * error), and another for the hash table.
167 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
169 * - When reply is received from the server (osc_enqueue_interpret())
170 * - ldlm_cli_enqueue_fini()
171 * - LDLM_LOCK_PUT(): releases caller reference acquired by
174 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
175 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
177 * - When lock is being cancelled (ldlm_lock_cancel())
178 * - ldlm_lock_destroy()
179 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
182 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
183 * either when lock is cancelled (osc_lock_blocking()), or when locks is
184 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
185 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
189 struct cl_lock_slice ols_cl;
190 /** underlying DLM lock */
191 struct ldlm_lock *ols_lock;
192 /** lock value block */
193 struct ost_lvb ols_lvb;
194 /** DLM flags with which osc_lock::ols_lock was enqueued */
196 /** osc_lock::ols_lock handle */
197 struct lustre_handle ols_handle;
198 struct ldlm_enqueue_info ols_einfo;
199 enum osc_lock_state ols_state;
201 * true, if ldlm_lock_addref() was called against
202 * osc_lock::ols_lock. This is used for sanity checking.
204 * \see osc_lock::ols_has_ref
206 unsigned ols_hold :1,
208 * this is much like osc_lock::ols_hold, except that this bit is
209 * cleared _after_ reference in released in osc_lock_unuse(). This
210 * fine distinction is needed because:
212 * - if ldlm lock still has a reference, osc_ast_data_get() needs
213 * to return associated cl_lock (so that a flag is needed that is
214 * cleared after ldlm_lock_decref() returned), and
216 * - ldlm_lock_decref() can invoke blocking ast (for a
217 * LDLM_FL_CBPENDING lock), and osc_lock functions like
218 * osc_lock_cancel() called from there need to know whether to
219 * release lock reference (so that a flag is needed that is
220 * cleared before ldlm_lock_decref() is called).
224 * inherit the lockless attribute from top level cl_io.
225 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
229 * set by osc_lock_use() to wait until blocking AST enters into
230 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
231 * further synchronization.
235 * If the data of this lock has been flushed to server side.
239 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
240 * the EVAVAIL error as torerable, this will make upper logic happy
241 * to wait all glimpse locks to each OSTs to be completed.
242 * Glimpse lock converts to normal lock if the server lock is
244 * Glimpse lock should be destroyed immediately after use.
248 * IO that owns this lock. This field is used for a dead-lock
249 * avoidance by osc_lock_enqueue().
251 * \see osc_deadlock_is_possible()
253 struct osc_io *ols_owner;
258 * Page state private for osc layer.
261 struct cl_page_slice ops_cl;
263 * Page queues used by osc to detect when RPC can be formed.
265 struct osc_async_page ops_oap;
267 * An offset within page from which next transfer starts. This is used
268 * by cl_page_clip() to submit partial page transfers.
272 * An offset within page at which next transfer ends.
274 * \see osc_page::ops_from.
278 * Boolean, true iff page is under transfer. Used for sanity checking.
280 unsigned ops_transfer_pinned:1,
282 * True for a `temporary page' created by read-ahead code, probably
283 * outside of any DLM lock.
287 * Linkage into a per-osc_object list of pages in flight. For
290 struct list_head ops_inflight;
292 * Thread that submitted this page for transfer. For debugging.
294 cfs_task_t *ops_submitter;
296 * Submit time - the time when the page is starting RPC. For debugging.
298 cfs_time_t ops_submit_time;
301 extern cfs_mem_cache_t *osc_page_kmem;
302 extern cfs_mem_cache_t *osc_lock_kmem;
303 extern cfs_mem_cache_t *osc_object_kmem;
304 extern cfs_mem_cache_t *osc_thread_kmem;
305 extern cfs_mem_cache_t *osc_session_kmem;
306 extern cfs_mem_cache_t *osc_req_kmem;
308 extern struct lu_device_type osc_device_type;
309 extern struct lu_context_key osc_key;
310 extern struct lu_context_key osc_session_key;
312 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
314 int osc_lock_init(const struct lu_env *env,
315 struct cl_object *obj, struct cl_lock *lock,
316 const struct cl_io *io);
317 int osc_io_init (const struct lu_env *env,
318 struct cl_object *obj, struct cl_io *io);
319 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
321 struct lu_object *osc_object_alloc(const struct lu_env *env,
322 const struct lu_object_header *hdr,
323 struct lu_device *dev);
324 struct cl_page *osc_page_init (const struct lu_env *env,
325 struct cl_object *obj,
326 struct cl_page *page, cfs_page_t *vmpage);
328 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
329 struct ldlm_res_id *resname);
330 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
331 pgoff_t start, pgoff_t end);
332 int osc_lvb_print (const struct lu_env *env, void *cookie,
333 lu_printer_t p, const struct ost_lvb *lvb);
334 void osc_io_submit_page(const struct lu_env *env,
335 struct osc_io *oio, struct osc_page *opg,
336 enum cl_req_type crt);
338 void osc_object_set_contended (struct osc_object *obj);
339 void osc_object_clear_contended(struct osc_object *obj);
340 int osc_object_is_contended (struct osc_object *obj);
342 int osc_lock_is_lockless (const struct osc_lock *olck);
344 /*****************************************************************************
350 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
352 struct osc_thread_info *info;
354 info = lu_context_key_get(&env->le_ctx, &osc_key);
355 LASSERT(info != NULL);
359 static inline struct osc_session *osc_env_session(const struct lu_env *env)
361 struct osc_session *ses;
363 ses = lu_context_key_get(env->le_ses, &osc_session_key);
364 LASSERT(ses != NULL);
368 static inline struct osc_io *osc_env_io(const struct lu_env *env)
370 return &osc_env_session(env)->os_io;
373 static inline int osc_is_object(const struct lu_object *obj)
375 return obj->lo_dev->ld_type == &osc_device_type;
378 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
380 LINVRNT(d->ld_type == &osc_device_type);
381 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
384 static inline struct obd_export *osc_export(const struct osc_object *obj)
386 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
389 static inline struct osc_object *cl2osc(const struct cl_object *obj)
391 LINVRNT(osc_is_object(&obj->co_lu));
392 return container_of0(obj, struct osc_object, oo_cl);
395 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
397 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
398 if (mode == CLM_READ)
400 else if (mode == CLM_WRITE)
406 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
408 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
411 else if (mode == LCK_PW)
417 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
419 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
420 return container_of0(slice, struct osc_page, ops_cl);
423 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
425 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
426 return container_of0(slice, struct osc_lock, ols_cl);
429 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
431 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
434 static inline int osc_io_srvlock(struct osc_io *oio)
436 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
441 #endif /* OSC_CL_INTERNAL_H */