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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/include/lustre_osc.h
35 * OSC layer structures and methods common for both OSC and MDC.
37 * This file contains OSC interfaces used by OSC and MDC. Most of them
38 * were just moved from lustre/osc/osc_cl_internal.h for Data-on-MDT
41 * Author: Nikita Danilov <nikita.danilov@sun.com>
42 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
43 * Author: Mikhail Pershin <mike.pershin@intel.com>
49 #include <libcfs/libcfs.h>
51 #include <cl_object.h>
57 struct osc_quota_info {
58 /** linkage for quota hash table */
59 struct hlist_node oqi_hash;
64 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
65 page is added to an rpc */
66 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
67 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
68 to give the caller a chance to update
69 or cancel the size of the io */
73 struct osc_async_page {
75 unsigned short oap_cmd;
77 struct list_head oap_pending_item;
78 struct list_head oap_rpc_item;
81 unsigned oap_page_off;
82 enum async_flags oap_async_flags;
84 struct brw_page oap_brw_page;
86 struct ptlrpc_request *oap_request;
87 struct client_obd *oap_cli;
88 struct osc_object *oap_obj;
93 #define oap_page oap_brw_page.pg
94 #define oap_count oap_brw_page.count
95 #define oap_brw_flags oap_brw_page.flag
97 static inline struct osc_async_page *brw_page2oap(struct brw_page *pga)
99 return container_of(pga, struct osc_async_page, oap_brw_page);
103 struct cl_device od_cl;
104 struct obd_export *od_exp;
106 /* Write stats is actually protected by client_obd's lock. */
108 uint64_t os_lockless_writes; /* by bytes */
109 uint64_t os_lockless_reads; /* by bytes */
110 uint64_t os_lockless_truncates; /* by times */
113 /* configuration item(s) */
114 time64_t od_contention_time;
115 int od_lockless_truncate;
121 * State maintained by osc layer for each IO context.
125 struct cl_io_slice oi_cl;
126 /** true if this io is lockless. */
127 unsigned int oi_lockless:1,
128 /** true if this io is counted as active IO */
130 /** true if this io has CAP_SYS_RESOURCE */
131 oi_cap_sys_resource:1;
132 /** how many LRU pages are reserved for this IO */
133 unsigned long oi_lru_reserved;
135 /** active extents, we know how many bytes is going to be written,
136 * so having an active extent will prevent it from being fragmented */
137 struct osc_extent *oi_active;
138 /** partially truncated extent, we need to hold this extent to prevent
139 * page writeback from happening. */
140 struct osc_extent *oi_trunc;
141 /** write osc_lock for this IO, used by osc_extent_find(). */
142 struct osc_lock *oi_write_osclock;
144 struct osc_async_cbargs {
147 struct completion opc_sync;
152 * State maintained by osc layer for the duration of a system call.
158 #define OTI_PVEC_SIZE 256
159 struct osc_thread_info {
160 struct ldlm_res_id oti_resname;
161 union ldlm_policy_data oti_policy;
162 struct cl_attr oti_attr;
164 struct pagevec oti_pagevec;
165 void *oti_pvec[OTI_PVEC_SIZE];
167 * Fields used by cl_lock_discard_pages().
169 pgoff_t oti_next_index;
170 pgoff_t oti_fn_index; /* first non-overlapped index */
171 struct cl_sync_io oti_anchor;
172 struct cl_req_attr oti_req_attr;
173 struct lu_buf oti_ladvise_buf;
176 static inline __u64 osc_enq2ldlm_flags(__u32 enqflags)
180 CDEBUG(D_DLMTRACE, "flags: %x\n", enqflags);
182 LASSERT((enqflags & ~CEF_MASK) == 0);
184 if (enqflags & CEF_NONBLOCK)
185 result |= LDLM_FL_BLOCK_NOWAIT;
186 if (enqflags & CEF_GLIMPSE)
187 result |= LDLM_FL_HAS_INTENT;
188 if (enqflags & CEF_DISCARD_DATA)
189 result |= LDLM_FL_AST_DISCARD_DATA;
190 if (enqflags & CEF_PEEK)
191 result |= LDLM_FL_TEST_LOCK;
192 if (enqflags & CEF_LOCK_MATCH)
193 result |= LDLM_FL_MATCH_LOCK;
194 if (enqflags & CEF_LOCK_NO_EXPAND)
195 result |= LDLM_FL_NO_EXPANSION;
196 if (enqflags & CEF_SPECULATIVE)
197 result |= LDLM_FL_SPECULATIVE;
201 typedef int (*osc_enqueue_upcall_f)(void *cookie, struct lustre_handle *lockh,
204 struct osc_enqueue_args {
205 struct obd_export *oa_exp;
206 enum ldlm_type oa_type;
207 enum ldlm_mode oa_mode;
209 osc_enqueue_upcall_f oa_upcall;
211 struct ost_lvb *oa_lvb;
212 struct lustre_handle oa_lockh;
217 * Bit flags for osc_dlm_lock_at_pageoff().
221 * Just check if the desired lock exists, it won't hold reference
224 OSC_DAP_FL_TEST_LOCK = 1 << 0,
226 * Return the lock even if it is being canceled.
228 OSC_DAP_FL_CANCELING = 1 << 1
232 * The set of operations which are different for MDC and OSC objects
234 struct osc_object_operations {
235 void (*oto_build_res_name)(struct osc_object *osc,
236 struct ldlm_res_id *resname);
237 struct ldlm_lock* (*oto_dlmlock_at_pgoff)(const struct lu_env *env,
238 struct osc_object *obj,
240 enum osc_dap_flags dap_flags);
244 struct cl_object oo_cl;
245 struct lov_oinfo *oo_oinfo;
247 * True if locking against this stripe got -EUSERS.
250 ktime_t oo_contention_time;
251 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
253 * IO context used for invariant checks in osc_lock_has_pages().
255 struct cl_io oo_debug_io;
256 /** Serialization object for osc_object::oo_debug_io. */
257 struct mutex oo_debug_mutex;
260 * used by the osc to keep track of what objects to build into rpcs.
261 * Protected by client_obd->cli_loi_list_lock.
263 struct list_head oo_ready_item;
264 struct list_head oo_hp_ready_item;
265 struct list_head oo_write_item;
266 struct list_head oo_read_item;
269 * extent is a red black tree to manage (async) dirty pages.
271 struct rb_root oo_root;
273 * Manage write(dirty) extents.
275 struct list_head oo_hp_exts; /* list of hp extents */
276 struct list_head oo_urgent_exts; /* list of writeback extents */
277 struct list_head oo_full_exts;
279 struct list_head oo_reading_exts;
281 atomic_t oo_nr_reads;
282 atomic_t oo_nr_writes;
284 /** Protect extent tree. Will be used to protect
285 * oo_{read|write}_pages soon. */
289 * Radix tree for caching pages
291 spinlock_t oo_tree_lock;
292 struct radix_tree_root oo_tree;
293 unsigned long oo_npages;
295 /* Protect osc_lock this osc_object has */
296 struct list_head oo_ol_list;
297 spinlock_t oo_ol_spin;
299 /** number of active IOs of this object */
301 wait_queue_head_t oo_io_waitq;
303 const struct osc_object_operations *oo_obj_ops;
307 static inline void osc_build_res_name(struct osc_object *osc,
308 struct ldlm_res_id *resname)
310 return osc->oo_obj_ops->oto_build_res_name(osc, resname);
313 static inline struct ldlm_lock *osc_dlmlock_at_pgoff(const struct lu_env *env,
314 struct osc_object *obj,
316 enum osc_dap_flags flags)
318 return obj->oo_obj_ops->oto_dlmlock_at_pgoff(env, obj, index, flags);
321 static inline void osc_object_lock(struct osc_object *obj)
323 spin_lock(&obj->oo_lock);
326 static inline int osc_object_trylock(struct osc_object *obj)
328 return spin_trylock(&obj->oo_lock);
331 static inline void osc_object_unlock(struct osc_object *obj)
333 spin_unlock(&obj->oo_lock);
336 #define assert_osc_object_is_locked(obj) \
337 assert_spin_locked(&obj->oo_lock)
339 static inline void osc_object_set_contended(struct osc_object *obj)
341 obj->oo_contention_time = ktime_get();
343 obj->oo_contended = 1;
346 static inline void osc_object_clear_contended(struct osc_object *obj)
348 obj->oo_contended = 0;
352 * Lock "micro-states" for osc layer.
354 enum osc_lock_state {
363 * osc-private state of cl_lock.
365 * Interaction with DLM.
367 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
368 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_dlmlock.
370 * This pointer is protected through a reference, acquired by
371 * osc_lock_upcall0(). Also, an additional reference is acquired by
372 * ldlm_lock_addref() call protecting the lock from cancellation, until
373 * osc_lock_unuse() releases it.
375 * Below is a description of how lock references are acquired and released
378 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
379 * - ldlm_lock_create()
380 * - ldlm_lock_new(): initializes a lock with 2 references. One for
381 * the caller (released when reply from the server is received, or on
382 * error), and another for the hash table.
383 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
385 * - When reply is received from the server (osc_enqueue_interpret())
386 * - ldlm_cli_enqueue_fini()
387 * - LDLM_LOCK_PUT(): releases caller reference acquired by
390 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
391 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
393 * - When lock is being cancelled (ldlm_lock_cancel())
394 * - ldlm_lock_destroy()
395 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
398 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
399 * either when lock is cancelled (osc_lock_blocking()), or when locks is
400 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
401 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
405 struct cl_lock_slice ols_cl;
406 /** Internal lock to protect states, etc. */
408 /** Owner sleeps on this channel for state change */
409 struct cl_sync_io *ols_owner;
410 /** waiting list for this lock to be cancelled */
411 struct list_head ols_waiting_list;
412 /** wait entry of ols_waiting_list */
413 struct list_head ols_wait_entry;
414 /** list entry for osc_object::oo_ol_list */
415 struct list_head ols_nextlock_oscobj;
417 /** underlying DLM lock */
418 struct ldlm_lock *ols_dlmlock;
419 /** DLM flags with which osc_lock::ols_lock was enqueued */
421 /** osc_lock::ols_lock handle */
422 struct lustre_handle ols_handle;
423 struct ldlm_enqueue_info ols_einfo;
424 enum osc_lock_state ols_state;
425 /** lock value block */
426 struct ost_lvb ols_lvb;
427 /** Lockless operations to be used by lockless lock */
428 const struct cl_lock_operations *ols_lockless_ops;
430 * true, if ldlm_lock_addref() was called against
431 * osc_lock::ols_lock. This is used for sanity checking.
433 * \see osc_lock::ols_has_ref
435 unsigned ols_hold :1,
437 * this is much like osc_lock::ols_hold, except that this bit is
438 * cleared _after_ reference in released in osc_lock_unuse(). This
439 * fine distinction is needed because:
441 * - if ldlm lock still has a reference, osc_ast_data_get() needs
442 * to return associated cl_lock (so that a flag is needed that is
443 * cleared after ldlm_lock_decref() returned), and
445 * - ldlm_lock_decref() can invoke blocking ast (for a
446 * LDLM_FL_CBPENDING lock), and osc_lock functions like
447 * osc_lock_cancel() called from there need to know whether to
448 * release lock reference (so that a flag is needed that is
449 * cleared before ldlm_lock_decref() is called).
453 * inherit the lockless attribute from top level cl_io.
454 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
458 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
459 * the EVAVAIL error as torerable, this will make upper logic happy
460 * to wait all glimpse locks to each OSTs to be completed.
461 * Glimpse lock converts to normal lock if the server lock is granted.
462 * Glimpse lock should be destroyed immediately after use.
466 * For async glimpse lock.
470 * for speculative locks - asynchronous glimpse locks and ladvise
471 * lockahead manual lock requests
473 * Used to tell osc layer to not wait for the ldlm reply from the
474 * server, so the osc lock will be short lived - It only exists to
475 * create the ldlm request and is not updated on request completion.
480 static inline int osc_lock_is_lockless(const struct osc_lock *ols)
482 return (ols->ols_cl.cls_ops == ols->ols_lockless_ops);
486 * Page state private for osc layer.
489 struct cl_page_slice ops_cl;
491 * Page queues used by osc to detect when RPC can be formed.
493 struct osc_async_page ops_oap;
495 * An offset within page from which next transfer starts. This is used
496 * by cl_page_clip() to submit partial page transfers.
500 * An offset within page at which next transfer ends.
502 * \see osc_page::ops_from.
506 * Boolean, true iff page is under transfer. Used for sanity checking.
508 unsigned ops_transfer_pinned:1,
514 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
518 * If the page is in osc_object::oo_tree.
522 * lru page list. See osc_lru_{del|use}() in osc_page.c for usage.
524 struct list_head ops_lru;
526 * Submit time - the time when the page is starting RPC. For debugging.
528 ktime_t ops_submit_time;
531 struct osc_brw_async_args {
533 int aa_requested_nob;
537 struct brw_page **aa_ppga;
538 struct client_obd *aa_cli;
539 struct list_head aa_oaps;
540 struct list_head aa_exts;
543 extern struct kmem_cache *osc_lock_kmem;
544 extern struct kmem_cache *osc_object_kmem;
545 extern struct kmem_cache *osc_thread_kmem;
546 extern struct kmem_cache *osc_session_kmem;
547 extern struct kmem_cache *osc_extent_kmem;
548 extern struct kmem_cache *osc_quota_kmem;
549 extern struct kmem_cache *osc_obdo_kmem;
551 extern struct lu_context_key osc_key;
552 extern struct lu_context_key osc_session_key;
554 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
557 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
558 struct cl_page *page, pgoff_t ind);
559 void osc_index2policy(union ldlm_policy_data *policy, const struct cl_object *obj,
560 pgoff_t start, pgoff_t end);
561 void osc_lru_add_batch(struct client_obd *cli, struct list_head *list);
562 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
563 enum cl_req_type crt, int brw_flags);
564 int lru_queue_work(const struct lu_env *env, void *data);
565 long osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
566 long target, bool force);
569 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
571 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
572 struct page *page, loff_t offset);
573 int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
574 struct osc_page *ops, cl_commit_cbt cb);
575 int osc_page_cache_add(const struct lu_env *env, struct osc_page *opg,
576 struct cl_io *io, cl_commit_cbt cb);
577 int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj,
578 struct osc_page *ops);
579 int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
580 struct osc_page *ops);
581 int osc_queue_sync_pages(const struct lu_env *env, const struct cl_io *io,
582 struct osc_object *obj, struct list_head *list,
584 int osc_cache_truncate_start(const struct lu_env *env, struct osc_object *obj,
585 __u64 size, struct osc_extent **extp);
586 void osc_cache_truncate_end(const struct lu_env *env, struct osc_extent *ext);
587 int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
588 pgoff_t start, pgoff_t end, int hp, int discard);
589 int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
590 pgoff_t start, pgoff_t end);
591 int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
592 struct osc_object *osc, int async);
593 static inline void osc_wake_cache_waiters(struct client_obd *cli)
595 wake_up(&cli->cl_cache_waiters);
598 static inline int osc_io_unplug_async(const struct lu_env *env,
599 struct client_obd *cli,
600 struct osc_object *osc)
602 return osc_io_unplug0(env, cli, osc, 1);
605 static inline void osc_io_unplug(const struct lu_env *env,
606 struct client_obd *cli,
607 struct osc_object *osc)
609 (void)osc_io_unplug0(env, cli, osc, 0);
612 typedef bool (*osc_page_gang_cbt)(const struct lu_env *, struct cl_io *,
613 struct osc_page *, void *);
614 bool osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io,
615 struct osc_object *osc, pgoff_t start, pgoff_t end,
616 osc_page_gang_cbt cb, void *cbdata);
617 bool osc_discard_cb(const struct lu_env *env, struct cl_io *io,
618 struct osc_page *ops, void *cbdata);
621 int osc_device_init(const struct lu_env *env, struct lu_device *d,
622 const char *name, struct lu_device *next);
623 struct lu_device *osc_device_fini(const struct lu_env *env,
624 struct lu_device *d);
625 struct lu_device *osc_device_free(const struct lu_env *env,
626 struct lu_device *d);
629 int osc_object_init(const struct lu_env *env, struct lu_object *obj,
630 const struct lu_object_conf *conf);
631 void osc_object_free(const struct lu_env *env, struct lu_object *obj);
632 int osc_lvb_print(const struct lu_env *env, void *cookie,
633 lu_printer_t p, const struct ost_lvb *lvb);
634 int osc_object_print(const struct lu_env *env, void *cookie,
635 lu_printer_t p, const struct lu_object *obj);
636 int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
637 struct cl_attr *attr);
638 int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
639 const struct cl_attr *attr, unsigned valid);
640 int osc_object_glimpse(const struct lu_env *env, const struct cl_object *obj,
641 struct ost_lvb *lvb);
642 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc);
643 int osc_object_is_contended(struct osc_object *obj);
644 int osc_object_find_cbdata(const struct lu_env *env, struct cl_object *obj,
645 ldlm_iterator_t iter, void *data);
646 int osc_object_prune(const struct lu_env *env, struct cl_object *obj);
649 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd);
650 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg);
651 int osc_precleanup_common(struct obd_device *obd);
652 int osc_cleanup_common(struct obd_device *obd);
653 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
654 u32 keylen, void *key, u32 vallen, void *val,
655 struct ptlrpc_request_set *set);
656 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
657 struct hlist_node *hnode, void *arg);
658 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
659 struct obd_device *obd, struct obd_uuid *cluuid,
660 struct obd_connect_data *data, void *localdata);
661 int osc_disconnect(struct obd_export *exp);
662 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
663 obd_enqueue_update_f upcall, void *cookie);
666 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
667 enum cl_req_type crt, struct cl_2queue *queue);
668 int osc_io_commit_async(const struct lu_env *env,
669 const struct cl_io_slice *ios,
670 struct cl_page_list *qin, int from, int to,
672 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios);
673 void osc_io_iter_fini(const struct lu_env *env,
674 const struct cl_io_slice *ios);
675 int osc_io_rw_iter_init(const struct lu_env *env,
676 const struct cl_io_slice *ios);
677 void osc_io_rw_iter_fini(const struct lu_env *env,
678 const struct cl_io_slice *ios);
679 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios);
680 void osc_io_setattr_end(const struct lu_env *env,
681 const struct cl_io_slice *slice);
682 int osc_io_read_start(const struct lu_env *env,
683 const struct cl_io_slice *slice);
684 int osc_io_write_start(const struct lu_env *env,
685 const struct cl_io_slice *slice);
686 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice);
687 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
688 struct cl_fsync_io *fio);
689 void osc_io_fsync_end(const struct lu_env *env,
690 const struct cl_io_slice *slice);
691 void osc_read_ahead_release(const struct lu_env *env, void *cbdata);
694 void osc_lock_to_lockless(const struct lu_env *env, struct osc_lock *ols,
696 void osc_lock_wake_waiters(const struct lu_env *env, struct osc_object *osc,
697 struct osc_lock *oscl);
698 int osc_lock_enqueue_wait(const struct lu_env *env, struct osc_object *obj,
699 struct osc_lock *oscl);
700 void osc_lock_set_writer(const struct lu_env *env, const struct cl_io *io,
701 struct cl_object *obj, struct osc_lock *oscl);
702 int osc_lock_print(const struct lu_env *env, void *cookie,
703 lu_printer_t p, const struct cl_lock_slice *slice);
704 void osc_lock_cancel(const struct lu_env *env,
705 const struct cl_lock_slice *slice);
706 void osc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice);
707 int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data);
708 unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock);
710 /*****************************************************************************
712 * Accessors and type conversions.
715 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
717 struct osc_thread_info *info;
719 info = lu_context_key_get(&env->le_ctx, &osc_key);
720 LASSERT(info != NULL);
724 static inline struct osc_session *osc_env_session(const struct lu_env *env)
726 struct osc_session *ses;
728 ses = lu_context_key_get(env->le_ses, &osc_session_key);
729 LASSERT(ses != NULL);
733 static inline struct osc_io *osc_env_io(const struct lu_env *env)
735 return &osc_env_session(env)->os_io;
738 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
740 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
743 static inline struct obd_export *osc_export(const struct osc_object *obj)
745 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
748 static inline struct client_obd *osc_cli(const struct osc_object *obj)
750 return &osc_export(obj)->exp_obd->u.cli;
753 static inline struct osc_object *cl2osc(const struct cl_object *obj)
755 return container_of0(obj, struct osc_object, oo_cl);
758 static inline struct cl_object *osc2cl(const struct osc_object *obj)
760 return (struct cl_object *)&obj->oo_cl;
763 static inline struct osc_device *obd2osc_dev(const struct obd_device *obd)
765 return container_of0(obd->obd_lu_dev, struct osc_device,
769 static inline struct lu_device *osc2lu_dev(struct osc_device *osc)
771 return &osc->od_cl.cd_lu_dev;
774 static inline struct lu_object *osc2lu(struct osc_object *osc)
776 return &osc->oo_cl.co_lu;
779 static inline struct osc_object *lu2osc(const struct lu_object *obj)
781 return container_of0(obj, struct osc_object, oo_cl.co_lu);
784 static inline struct osc_io *cl2osc_io(const struct lu_env *env,
785 const struct cl_io_slice *slice)
787 struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
789 LINVRNT(oio == osc_env_io(env));
793 static inline enum ldlm_mode osc_cl_lock2ldlm(enum cl_lock_mode mode)
795 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
796 if (mode == CLM_READ)
798 if (mode == CLM_WRITE)
803 static inline enum cl_lock_mode osc_ldlm2cl_lock(enum ldlm_mode mode)
805 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
813 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
815 return container_of0(slice, struct osc_page, ops_cl);
818 static inline struct osc_page *oap2osc(struct osc_async_page *oap)
820 return container_of0(oap, struct osc_page, ops_oap);
823 static inline pgoff_t osc_index(struct osc_page *opg)
825 return opg->ops_cl.cpl_page->cp_osc_index;
828 static inline struct cl_page *oap2cl_page(struct osc_async_page *oap)
830 return oap2osc(oap)->ops_cl.cpl_page;
833 static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
835 return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
838 static inline struct osc_page *
839 osc_cl_page_osc(struct cl_page *page, struct osc_object *osc)
841 const struct cl_page_slice *slice;
843 LASSERT(osc != NULL);
844 slice = cl_object_page_slice(&osc->oo_cl, page);
845 return cl2osc_page(slice);
848 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
850 return container_of0(slice, struct osc_lock, ols_cl);
853 static inline int osc_io_srvlock(struct osc_io *oio)
855 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
858 enum osc_extent_state {
859 OES_INV = 0, /** extent is just initialized or destroyed */
860 OES_ACTIVE = 1, /** process is using this extent */
861 OES_CACHE = 2, /** extent is ready for IO */
862 OES_LOCKING = 3, /** locking page to prepare IO */
863 OES_LOCK_DONE = 4, /** locking finished, ready to send */
864 OES_RPC = 5, /** in RPC */
865 OES_TRUNC = 6, /** being truncated */
870 * osc_extent data to manage dirty pages.
871 * osc_extent has the following attributes:
872 * 1. all pages in the same must be in one RPC in write back;
873 * 2. # of pages must be less than max_pages_per_rpc - implied by 1;
874 * 3. must be covered by only 1 osc_lock;
875 * 4. exclusive. It's impossible to have overlapped osc_extent.
877 * The lifetime of an extent is from when the 1st page is dirtied to when
878 * all pages inside it are written out.
882 * page lock -> client_obd_list_lock -> object lock(osc_object::oo_lock)
885 /** red-black tree node */
886 struct rb_node oe_node;
887 /** osc_object of this extent */
888 struct osc_object *oe_obj;
889 /** refcount, removed from red-black tree if reaches zero. */
891 /** busy if non-zero */
893 /** link list of osc_object's oo_{hp|urgent|locking}_exts. */
894 struct list_head oe_link;
895 /** state of this extent */
896 enum osc_extent_state oe_state;
897 /** flags for this extent. */
898 /** 0 is write, 1 is read */
899 unsigned int oe_rw:1,
900 /** sync extent, queued by osc_queue_sync_pages() */
902 /** set if this extent has partial, sync pages.
903 * Extents with partial page(s) can't merge with others in RPC */
907 /** an ACTIVE extent is going to be truncated, so when this extent
908 * is released, it will turn into TRUNC state instead of CACHE. */
910 /** this extent should be written asap and someone may wait for the
911 * write to finish. This bit is usually set along with urgent if
912 * the extent was CACHE state.
913 * fsync_wait extent can't be merged because new extent region may
914 * exceed fsync range. */
916 /** covering lock is being canceled */
918 /** this extent should be written back asap. set if one of pages is
919 * called by page WB daemon, or sync write or reading requests. */
921 /** Non-delay RPC should be used for this extent. */
923 /** direct IO pages */
925 /** how many grants allocated for this extent.
926 * Grant allocated for this extent. There is no grant allocated
927 * for reading extents and sync write extents. */
928 unsigned int oe_grants;
929 /** # of dirty pages in this extent */
930 unsigned int oe_nr_pages;
931 /** list of pending oap pages. Pages in this list are NOT sorted. */
932 struct list_head oe_pages;
933 /** start and end index of this extent, include start and end
934 * themselves. Page offset here is the page index of osc_pages.
935 * oe_start is used as keyword for red-black tree. */
938 /** maximum ending index of this extent, this is limited by
939 * max_pages_per_rpc, lock extent and chunk size. */
941 /** waitqueue - for those who want to be notified if this extent's
942 * state has changed. */
943 wait_queue_head_t oe_waitq;
944 /** lock covering this extent */
945 struct ldlm_lock *oe_dlmlock;
946 /** terminator of this extent. Must be true if this extent is in IO. */
947 struct task_struct *oe_owner;
948 /** return value of writeback. If somebody is waiting for this extent,
949 * this value can be known by outside world. */
951 /** max pages per rpc when this extent was created */
952 unsigned int oe_mppr;
953 /** FLR: layout version when this osc_extent is publised */
954 __u32 oe_layout_version;
959 #endif /* LUSTRE_OSC_H */