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 (c) 2008, 2010, Oracle and/or its affiliates. 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_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 struct cl_object oo_cl;
105 struct lov_oinfo *oo_oinfo;
107 * True if locking against this stripe got -EUSERS.
110 cfs_time_t oo_contention_time;
111 #ifdef INVARIANT_CHECK
113 * IO context used for invariant checks in osc_lock_has_pages().
115 struct cl_io oo_debug_io;
116 /** Serialization object for osc_object::oo_debug_io. */
117 cfs_mutex_t oo_debug_mutex;
120 * List of pages in transfer.
122 cfs_list_t oo_inflight[CRT_NR];
124 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
125 * locked during take-off and landing.
127 cfs_spinlock_t oo_seatbelt;
131 * Lock "micro-states" for osc layer.
133 enum osc_lock_state {
144 * osc-private state of cl_lock.
146 * Interaction with DLM.
148 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
150 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
151 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
153 * This pointer is protected through a reference, acquired by
154 * osc_lock_upcall0(). Also, an additional reference is acquired by
155 * ldlm_lock_addref() call protecting the lock from cancellation, until
156 * osc_lock_unuse() releases it.
158 * Below is a description of how lock references are acquired and released
161 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
162 * - ldlm_lock_create()
163 * - ldlm_lock_new(): initializes a lock with 2 references. One for
164 * the caller (released when reply from the server is received, or on
165 * error), and another for the hash table.
166 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
168 * - When reply is received from the server (osc_enqueue_interpret())
169 * - ldlm_cli_enqueue_fini()
170 * - LDLM_LOCK_PUT(): releases caller reference acquired by
173 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
174 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
176 * - When lock is being cancelled (ldlm_lock_cancel())
177 * - ldlm_lock_destroy()
178 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
181 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
182 * either when lock is cancelled (osc_lock_blocking()), or when locks is
183 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
184 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
188 struct cl_lock_slice ols_cl;
189 /** underlying DLM lock */
190 struct ldlm_lock *ols_lock;
191 /** lock value block */
192 struct ost_lvb ols_lvb;
193 /** DLM flags with which osc_lock::ols_lock was enqueued */
195 /** osc_lock::ols_lock handle */
196 struct lustre_handle ols_handle;
197 struct ldlm_enqueue_info ols_einfo;
198 enum osc_lock_state ols_state;
201 * How many pages are using this lock for io, currently only used by
202 * read-ahead. If non-zero, the underlying dlm lock won't be cancelled
203 * during recovery to avoid deadlock. see bz16774.
205 * \see osc_page::ops_lock
206 * \see osc_page_addref_lock(), osc_page_putref_lock()
208 cfs_atomic_t ols_pageref;
211 * true, if ldlm_lock_addref() was called against
212 * osc_lock::ols_lock. This is used for sanity checking.
214 * \see osc_lock::ols_has_ref
216 unsigned ols_hold :1,
218 * this is much like osc_lock::ols_hold, except that this bit is
219 * cleared _after_ reference in released in osc_lock_unuse(). This
220 * fine distinction is needed because:
222 * - if ldlm lock still has a reference, osc_ast_data_get() needs
223 * to return associated cl_lock (so that a flag is needed that is
224 * cleared after ldlm_lock_decref() returned), and
226 * - ldlm_lock_decref() can invoke blocking ast (for a
227 * LDLM_FL_CBPENDING lock), and osc_lock functions like
228 * osc_lock_cancel() called from there need to know whether to
229 * release lock reference (so that a flag is needed that is
230 * cleared before ldlm_lock_decref() is called).
234 * inherit the lockless attribute from top level cl_io.
235 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
239 * set by osc_lock_use() to wait until blocking AST enters into
240 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
241 * further synchronization.
245 * If the data of this lock has been flushed to server side.
249 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
250 * the EVAVAIL error as torerable, this will make upper logic happy
251 * to wait all glimpse locks to each OSTs to be completed.
252 * Glimpse lock converts to normal lock if the server lock is
254 * Glimpse lock should be destroyed immediately after use.
258 * IO that owns this lock. This field is used for a dead-lock
259 * avoidance by osc_lock_enqueue_wait().
261 * XXX: unfortunately, the owner of a osc_lock is not unique,
262 * the lock may have multiple users, if the lock is granted and
265 struct osc_io *ols_owner;
270 * Page state private for osc layer.
273 struct cl_page_slice ops_cl;
275 * Page queues used by osc to detect when RPC can be formed.
277 struct osc_async_page ops_oap;
279 * An offset within page from which next transfer starts. This is used
280 * by cl_page_clip() to submit partial page transfers.
284 * An offset within page at which next transfer ends.
286 * \see osc_page::ops_from.
290 * Boolean, true iff page is under transfer. Used for sanity checking.
292 unsigned ops_transfer_pinned:1,
294 * True for a `temporary page' created by read-ahead code, probably
295 * outside of any DLM lock.
299 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
303 * Linkage into a per-osc_object list of pages in flight. For
306 cfs_list_t ops_inflight;
308 * Thread that submitted this page for transfer. For debugging.
310 cfs_task_t *ops_submitter;
312 * Submit time - the time when the page is starting RPC. For debugging.
314 cfs_time_t ops_submit_time;
317 * A lock of which we hold a reference covers this page. Only used by
318 * read-ahead: for a readahead page, we hold it's covering lock to
319 * prevent it from being canceled during recovery.
321 * \see osc_lock::ols_pageref
322 * \see osc_page_addref_lock(), osc_page_putref_lock().
324 struct cl_lock *ops_lock;
327 extern cfs_mem_cache_t *osc_page_kmem;
328 extern cfs_mem_cache_t *osc_lock_kmem;
329 extern cfs_mem_cache_t *osc_object_kmem;
330 extern cfs_mem_cache_t *osc_thread_kmem;
331 extern cfs_mem_cache_t *osc_session_kmem;
332 extern cfs_mem_cache_t *osc_req_kmem;
334 extern struct lu_device_type osc_device_type;
335 extern struct lu_context_key osc_key;
336 extern struct lu_context_key osc_session_key;
338 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
340 int osc_lock_init(const struct lu_env *env,
341 struct cl_object *obj, struct cl_lock *lock,
342 const struct cl_io *io);
343 int osc_io_init (const struct lu_env *env,
344 struct cl_object *obj, struct cl_io *io);
345 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
347 struct lu_object *osc_object_alloc(const struct lu_env *env,
348 const struct lu_object_header *hdr,
349 struct lu_device *dev);
350 struct cl_page *osc_page_init (const struct lu_env *env,
351 struct cl_object *obj,
352 struct cl_page *page, cfs_page_t *vmpage);
354 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
355 struct ldlm_res_id *resname);
356 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
357 pgoff_t start, pgoff_t end);
358 int osc_lvb_print (const struct lu_env *env, void *cookie,
359 lu_printer_t p, const struct ost_lvb *lvb);
360 void osc_io_submit_page(const struct lu_env *env,
361 struct osc_io *oio, struct osc_page *opg,
362 enum cl_req_type crt);
364 void osc_object_set_contended (struct osc_object *obj);
365 void osc_object_clear_contended(struct osc_object *obj);
366 int osc_object_is_contended (struct osc_object *obj);
368 int osc_lock_is_lockless (const struct osc_lock *olck);
370 /*****************************************************************************
376 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
378 struct osc_thread_info *info;
380 info = lu_context_key_get(&env->le_ctx, &osc_key);
381 LASSERT(info != NULL);
385 static inline struct osc_session *osc_env_session(const struct lu_env *env)
387 struct osc_session *ses;
389 ses = lu_context_key_get(env->le_ses, &osc_session_key);
390 LASSERT(ses != NULL);
394 static inline struct osc_io *osc_env_io(const struct lu_env *env)
396 return &osc_env_session(env)->os_io;
399 static inline int osc_is_object(const struct lu_object *obj)
401 return obj->lo_dev->ld_type == &osc_device_type;
404 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
406 LINVRNT(d->ld_type == &osc_device_type);
407 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
410 static inline struct obd_export *osc_export(const struct osc_object *obj)
412 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
415 static inline struct osc_object *cl2osc(const struct cl_object *obj)
417 LINVRNT(osc_is_object(&obj->co_lu));
418 return container_of0(obj, struct osc_object, oo_cl);
421 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
423 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
424 if (mode == CLM_READ)
426 else if (mode == CLM_WRITE)
432 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
434 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
437 else if (mode == LCK_PW)
443 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
445 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
446 return container_of0(slice, struct osc_page, ops_cl);
449 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
451 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
452 return container_of0(slice, struct osc_lock, ols_cl);
455 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
457 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
460 static inline int osc_io_srvlock(struct osc_io *oio)
462 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
467 #endif /* OSC_CL_INTERNAL_H */