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.
32 * Copyright (c) 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
39 * This file is part of Lustre, http://www.lustre.org/
40 * Lustre is a trademark of Sun Microsystems, Inc.
42 * Internal interfaces of OSC layer.
44 * Author: Nikita Danilov <nikita.danilov@sun.com>
47 #ifndef OSC_CL_INTERNAL_H
48 #define OSC_CL_INTERNAL_H
51 # include <libcfs/libcfs.h>
53 # include <liblustre.h>
57 /* osc_build_res_name() */
59 #include <cl_object.h>
60 #include "osc_internal.h"
67 * State maintained by osc layer for each IO context.
71 struct cl_io_slice oi_cl;
72 /** true if this io is lockless. */
76 struct osc_setattr_cbargs {
78 cfs_completion_t opc_sync;
83 * State of transfer for osc.
86 struct cl_req_slice or_cl;
90 * State maintained by osc layer for the duration of a system call.
96 struct osc_thread_info {
97 struct ldlm_res_id oti_resname;
98 ldlm_policy_data_t oti_policy;
99 struct cl_lock_descr oti_descr;
100 struct cl_attr oti_attr;
101 struct lustre_handle oti_handle;
102 struct cl_page_list oti_plist;
106 struct cl_object oo_cl;
107 struct lov_oinfo *oo_oinfo;
109 * True if locking against this stripe got -EUSERS.
112 cfs_time_t oo_contention_time;
113 #ifdef INVARIANT_CHECK
115 * IO context used for invariant checks in osc_lock_has_pages().
117 struct cl_io oo_debug_io;
118 /** Serialization object for osc_object::oo_debug_io. */
119 cfs_mutex_t oo_debug_mutex;
122 * List of pages in transfer.
124 cfs_list_t oo_inflight[CRT_NR];
126 * Lock, protecting ccc_object::cob_inflight, because a seat-belt is
127 * locked during take-off and landing.
129 cfs_spinlock_t oo_seatbelt;
133 * Lock "micro-states" for osc layer.
135 enum osc_lock_state {
146 * osc-private state of cl_lock.
148 * Interaction with DLM.
150 * CLIO enqueues all DLM locks through ptlrpcd (that is, in "async" mode).
152 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
153 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_lock.
155 * This pointer is protected through a reference, acquired by
156 * osc_lock_upcall0(). Also, an additional reference is acquired by
157 * ldlm_lock_addref() call protecting the lock from cancellation, until
158 * osc_lock_unuse() releases it.
160 * Below is a description of how lock references are acquired and released
163 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
164 * - ldlm_lock_create()
165 * - ldlm_lock_new(): initializes a lock with 2 references. One for
166 * the caller (released when reply from the server is received, or on
167 * error), and another for the hash table.
168 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
170 * - When reply is received from the server (osc_enqueue_interpret())
171 * - ldlm_cli_enqueue_fini()
172 * - LDLM_LOCK_PUT(): releases caller reference acquired by
175 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
176 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
178 * - When lock is being cancelled (ldlm_lock_cancel())
179 * - ldlm_lock_destroy()
180 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
183 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
184 * either when lock is cancelled (osc_lock_blocking()), or when locks is
185 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
186 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
190 struct cl_lock_slice ols_cl;
191 /** underlying DLM lock */
192 struct ldlm_lock *ols_lock;
193 /** lock value block */
194 struct ost_lvb ols_lvb;
195 /** DLM flags with which osc_lock::ols_lock was enqueued */
197 /** osc_lock::ols_lock handle */
198 struct lustre_handle ols_handle;
199 struct ldlm_enqueue_info ols_einfo;
200 enum osc_lock_state ols_state;
203 * How many pages are using this lock for io, currently only used by
204 * read-ahead. If non-zero, the underlying dlm lock won't be cancelled
205 * during recovery to avoid deadlock. see bz16774.
207 * \see osc_page::ops_lock
208 * \see osc_page_addref_lock(), osc_page_putref_lock()
210 cfs_atomic_t ols_pageref;
213 * true, if ldlm_lock_addref() was called against
214 * osc_lock::ols_lock. This is used for sanity checking.
216 * \see osc_lock::ols_has_ref
218 unsigned ols_hold :1,
220 * this is much like osc_lock::ols_hold, except that this bit is
221 * cleared _after_ reference in released in osc_lock_unuse(). This
222 * fine distinction is needed because:
224 * - if ldlm lock still has a reference, osc_ast_data_get() needs
225 * to return associated cl_lock (so that a flag is needed that is
226 * cleared after ldlm_lock_decref() returned), and
228 * - ldlm_lock_decref() can invoke blocking ast (for a
229 * LDLM_FL_CBPENDING lock), and osc_lock functions like
230 * osc_lock_cancel() called from there need to know whether to
231 * release lock reference (so that a flag is needed that is
232 * cleared before ldlm_lock_decref() is called).
236 * inherit the lockless attribute from top level cl_io.
237 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
241 * set by osc_lock_use() to wait until blocking AST enters into
242 * osc_ldlm_blocking_ast0(), so that cl_lock mutex can be used for
243 * further synchronization.
247 * If the data of this lock has been flushed to server side.
251 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
252 * the EVAVAIL error as torerable, this will make upper logic happy
253 * to wait all glimpse locks to each OSTs to be completed.
254 * Glimpse lock converts to normal lock if the server lock is
256 * Glimpse lock should be destroyed immediately after use.
260 * For async glimpse lock.
264 * IO that owns this lock. This field is used for a dead-lock
265 * avoidance by osc_lock_enqueue_wait().
267 * XXX: unfortunately, the owner of a osc_lock is not unique,
268 * the lock may have multiple users, if the lock is granted and
271 struct osc_io *ols_owner;
276 * Page state private for osc layer.
279 struct cl_page_slice ops_cl;
281 * Page queues used by osc to detect when RPC can be formed.
283 struct osc_async_page ops_oap;
285 * An offset within page from which next transfer starts. This is used
286 * by cl_page_clip() to submit partial page transfers.
290 * An offset within page at which next transfer ends.
292 * \see osc_page::ops_from.
296 * Boolean, true iff page is under transfer. Used for sanity checking.
298 unsigned ops_transfer_pinned:1,
300 * True for a `temporary page' created by read-ahead code, probably
301 * outside of any DLM lock.
305 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
309 * Linkage into a per-osc_object list of pages in flight. For
312 cfs_list_t ops_inflight;
314 * Thread that submitted this page for transfer. For debugging.
316 cfs_task_t *ops_submitter;
318 * Submit time - the time when the page is starting RPC. For debugging.
320 cfs_time_t ops_submit_time;
323 * A lock of which we hold a reference covers this page. Only used by
324 * read-ahead: for a readahead page, we hold it's covering lock to
325 * prevent it from being canceled during recovery.
327 * \see osc_lock::ols_pageref
328 * \see osc_page_addref_lock(), osc_page_putref_lock().
330 struct cl_lock *ops_lock;
333 extern cfs_mem_cache_t *osc_page_kmem;
334 extern cfs_mem_cache_t *osc_lock_kmem;
335 extern cfs_mem_cache_t *osc_object_kmem;
336 extern cfs_mem_cache_t *osc_thread_kmem;
337 extern cfs_mem_cache_t *osc_session_kmem;
338 extern cfs_mem_cache_t *osc_req_kmem;
340 extern struct lu_device_type osc_device_type;
341 extern struct lu_context_key osc_key;
342 extern struct lu_context_key osc_session_key;
344 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
346 int osc_lock_init(const struct lu_env *env,
347 struct cl_object *obj, struct cl_lock *lock,
348 const struct cl_io *io);
349 int osc_io_init (const struct lu_env *env,
350 struct cl_object *obj, struct cl_io *io);
351 int osc_req_init (const struct lu_env *env, struct cl_device *dev,
353 struct lu_object *osc_object_alloc(const struct lu_env *env,
354 const struct lu_object_header *hdr,
355 struct lu_device *dev);
356 struct cl_page *osc_page_init (const struct lu_env *env,
357 struct cl_object *obj,
358 struct cl_page *page, cfs_page_t *vmpage);
360 void osc_lock_build_res(const struct lu_env *env, const struct osc_object *obj,
361 struct ldlm_res_id *resname);
362 void osc_index2policy (ldlm_policy_data_t *policy, const struct cl_object *obj,
363 pgoff_t start, pgoff_t end);
364 int osc_lvb_print (const struct lu_env *env, void *cookie,
365 lu_printer_t p, const struct ost_lvb *lvb);
366 void osc_io_submit_page(const struct lu_env *env,
367 struct osc_io *oio, struct osc_page *opg,
368 enum cl_req_type crt);
370 void osc_object_set_contended (struct osc_object *obj);
371 void osc_object_clear_contended(struct osc_object *obj);
372 int osc_object_is_contended (struct osc_object *obj);
374 int osc_lock_is_lockless (const struct osc_lock *olck);
376 /*****************************************************************************
382 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
384 struct osc_thread_info *info;
386 info = lu_context_key_get(&env->le_ctx, &osc_key);
387 LASSERT(info != NULL);
391 static inline struct osc_session *osc_env_session(const struct lu_env *env)
393 struct osc_session *ses;
395 ses = lu_context_key_get(env->le_ses, &osc_session_key);
396 LASSERT(ses != NULL);
400 static inline struct osc_io *osc_env_io(const struct lu_env *env)
402 return &osc_env_session(env)->os_io;
405 static inline int osc_is_object(const struct lu_object *obj)
407 return obj->lo_dev->ld_type == &osc_device_type;
410 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
412 LINVRNT(d->ld_type == &osc_device_type);
413 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
416 static inline struct obd_export *osc_export(const struct osc_object *obj)
418 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
421 static inline struct osc_object *cl2osc(const struct cl_object *obj)
423 LINVRNT(osc_is_object(&obj->co_lu));
424 return container_of0(obj, struct osc_object, oo_cl);
427 static inline ldlm_mode_t osc_cl_lock2ldlm(enum cl_lock_mode mode)
429 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
430 if (mode == CLM_READ)
432 else if (mode == CLM_WRITE)
438 static inline enum cl_lock_mode osc_ldlm2cl_lock(ldlm_mode_t mode)
440 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
443 else if (mode == LCK_PW)
449 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
451 LINVRNT(osc_is_object(&slice->cpl_obj->co_lu));
452 return container_of0(slice, struct osc_page, ops_cl);
455 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
457 LINVRNT(osc_is_object(&slice->cls_obj->co_lu));
458 return container_of0(slice, struct osc_lock, ols_cl);
461 static inline struct osc_lock *osc_lock_at(const struct cl_lock *lock)
463 return cl2osc_lock(cl_lock_at(lock, &osc_device_type));
466 static inline int osc_io_srvlock(struct osc_io *oio)
468 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
473 #endif /* OSC_CL_INTERNAL_H */