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;
76 unsigned short oap_interrupted:1;
78 struct list_head oap_pending_item;
79 struct list_head oap_rpc_item;
82 unsigned oap_page_off;
83 enum async_flags oap_async_flags;
85 struct brw_page oap_brw_page;
87 struct ptlrpc_request *oap_request;
88 struct client_obd *oap_cli;
89 struct osc_object *oap_obj;
94 #define oap_page oap_brw_page.pg
95 #define oap_count oap_brw_page.count
96 #define oap_brw_flags oap_brw_page.flag
98 static inline struct osc_async_page *brw_page2oap(struct brw_page *pga)
100 return container_of(pga, struct osc_async_page, oap_brw_page);
103 struct osc_cache_waiter {
104 struct list_head ocw_entry;
105 wait_queue_head_t ocw_waitq;
106 struct osc_async_page *ocw_oap;
112 struct cl_device od_cl;
113 struct obd_export *od_exp;
115 /* Write stats is actually protected by client_obd's lock. */
117 uint64_t os_lockless_writes; /* by bytes */
118 uint64_t os_lockless_reads; /* by bytes */
119 uint64_t os_lockless_truncates; /* by times */
122 /* configuration item(s) */
123 time64_t od_contention_time;
124 int od_lockless_truncate;
130 * State maintained by osc layer for each IO context.
134 struct cl_io_slice oi_cl;
135 /** true if this io is lockless. */
136 unsigned int oi_lockless:1,
137 /** true if this io is counted as active IO */
139 /** true if this io has CAP_SYS_RESOURCE */
140 oi_cap_sys_resource:1;
141 /** how many LRU pages are reserved for this IO */
142 unsigned long oi_lru_reserved;
144 /** active extents, we know how many bytes is going to be written,
145 * so having an active extent will prevent it from being fragmented */
146 struct osc_extent *oi_active;
147 /** partially truncated extent, we need to hold this extent to prevent
148 * page writeback from happening. */
149 struct osc_extent *oi_trunc;
150 /** write osc_lock for this IO, used by osc_extent_find(). */
151 struct osc_lock *oi_write_osclock;
153 struct osc_async_cbargs {
156 struct completion opc_sync;
161 * State maintained by osc layer for the duration of a system call.
167 #define OTI_PVEC_SIZE 256
168 struct osc_thread_info {
169 struct ldlm_res_id oti_resname;
170 union ldlm_policy_data oti_policy;
171 struct cl_lock_descr oti_descr;
172 struct cl_attr oti_attr;
173 struct lustre_handle oti_handle;
174 struct cl_page_list oti_plist;
176 struct pagevec oti_pagevec;
177 void *oti_pvec[OTI_PVEC_SIZE];
179 * Fields used by cl_lock_discard_pages().
181 pgoff_t oti_next_index;
182 pgoff_t oti_fn_index; /* first non-overlapped index */
183 struct cl_sync_io oti_anchor;
184 struct cl_req_attr oti_req_attr;
185 struct lu_buf oti_ladvise_buf;
188 static inline __u64 osc_enq2ldlm_flags(__u32 enqflags)
192 CDEBUG(D_DLMTRACE, "flags: %x\n", enqflags);
194 LASSERT((enqflags & ~CEF_MASK) == 0);
196 if (enqflags & CEF_NONBLOCK)
197 result |= LDLM_FL_BLOCK_NOWAIT;
198 if (enqflags & CEF_GLIMPSE)
199 result |= LDLM_FL_HAS_INTENT;
200 if (enqflags & CEF_DISCARD_DATA)
201 result |= LDLM_FL_AST_DISCARD_DATA;
202 if (enqflags & CEF_PEEK)
203 result |= LDLM_FL_TEST_LOCK;
204 if (enqflags & CEF_LOCK_MATCH)
205 result |= LDLM_FL_MATCH_LOCK;
206 if (enqflags & CEF_LOCK_NO_EXPAND)
207 result |= LDLM_FL_NO_EXPANSION;
208 if (enqflags & CEF_SPECULATIVE)
209 result |= LDLM_FL_SPECULATIVE;
213 typedef int (*osc_enqueue_upcall_f)(void *cookie, struct lustre_handle *lockh,
216 struct osc_enqueue_args {
217 struct obd_export *oa_exp;
218 enum ldlm_type oa_type;
219 enum ldlm_mode oa_mode;
221 osc_enqueue_upcall_f oa_upcall;
223 struct ost_lvb *oa_lvb;
224 struct lustre_handle oa_lockh;
229 * Bit flags for osc_dlm_lock_at_pageoff().
233 * Just check if the desired lock exists, it won't hold reference
236 OSC_DAP_FL_TEST_LOCK = 1 << 0,
238 * Return the lock even if it is being canceled.
240 OSC_DAP_FL_CANCELING = 1 << 1
244 * The set of operations which are different for MDC and OSC objects
246 struct osc_object_operations {
247 void (*oto_build_res_name)(struct osc_object *osc,
248 struct ldlm_res_id *resname);
249 struct ldlm_lock* (*oto_dlmlock_at_pgoff)(const struct lu_env *env,
250 struct osc_object *obj,
252 enum osc_dap_flags dap_flags);
256 struct cl_object oo_cl;
257 struct lov_oinfo *oo_oinfo;
259 * True if locking against this stripe got -EUSERS.
262 ktime_t oo_contention_time;
263 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
265 * IO context used for invariant checks in osc_lock_has_pages().
267 struct cl_io oo_debug_io;
268 /** Serialization object for osc_object::oo_debug_io. */
269 struct mutex oo_debug_mutex;
272 * used by the osc to keep track of what objects to build into rpcs.
273 * Protected by client_obd->cli_loi_list_lock.
275 struct list_head oo_ready_item;
276 struct list_head oo_hp_ready_item;
277 struct list_head oo_write_item;
278 struct list_head oo_read_item;
281 * extent is a red black tree to manage (async) dirty pages.
283 struct rb_root oo_root;
285 * Manage write(dirty) extents.
287 struct list_head oo_hp_exts; /* list of hp extents */
288 struct list_head oo_urgent_exts; /* list of writeback extents */
289 struct list_head oo_full_exts;
291 struct list_head oo_reading_exts;
293 atomic_t oo_nr_reads;
294 atomic_t oo_nr_writes;
296 /** Protect extent tree. Will be used to protect
297 * oo_{read|write}_pages soon. */
301 * Radix tree for caching pages
303 spinlock_t oo_tree_lock;
304 struct radix_tree_root oo_tree;
305 unsigned long oo_npages;
307 /* Protect osc_lock this osc_object has */
308 struct list_head oo_ol_list;
309 spinlock_t oo_ol_spin;
311 /** number of active IOs of this object */
313 wait_queue_head_t oo_io_waitq;
315 const struct osc_object_operations *oo_obj_ops;
319 static inline void osc_build_res_name(struct osc_object *osc,
320 struct ldlm_res_id *resname)
322 return osc->oo_obj_ops->oto_build_res_name(osc, resname);
325 static inline struct ldlm_lock *osc_dlmlock_at_pgoff(const struct lu_env *env,
326 struct osc_object *obj,
328 enum osc_dap_flags flags)
330 return obj->oo_obj_ops->oto_dlmlock_at_pgoff(env, obj, index, flags);
333 static inline void osc_object_lock(struct osc_object *obj)
335 spin_lock(&obj->oo_lock);
338 static inline int osc_object_trylock(struct osc_object *obj)
340 return spin_trylock(&obj->oo_lock);
343 static inline void osc_object_unlock(struct osc_object *obj)
345 spin_unlock(&obj->oo_lock);
348 static inline int osc_object_is_locked(struct osc_object *obj)
350 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
351 return spin_is_locked(&obj->oo_lock);
354 * It is not perfect to return true all the time.
355 * But since this function is only used for assertion
356 * and checking, it seems OK.
362 static inline void osc_object_set_contended(struct osc_object *obj)
364 obj->oo_contention_time = ktime_get();
366 obj->oo_contended = 1;
369 static inline void osc_object_clear_contended(struct osc_object *obj)
371 obj->oo_contended = 0;
375 * Lock "micro-states" for osc layer.
377 enum osc_lock_state {
386 * osc-private state of cl_lock.
388 * Interaction with DLM.
390 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
391 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_dlmlock.
393 * This pointer is protected through a reference, acquired by
394 * osc_lock_upcall0(). Also, an additional reference is acquired by
395 * ldlm_lock_addref() call protecting the lock from cancellation, until
396 * osc_lock_unuse() releases it.
398 * Below is a description of how lock references are acquired and released
401 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
402 * - ldlm_lock_create()
403 * - ldlm_lock_new(): initializes a lock with 2 references. One for
404 * the caller (released when reply from the server is received, or on
405 * error), and another for the hash table.
406 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
408 * - When reply is received from the server (osc_enqueue_interpret())
409 * - ldlm_cli_enqueue_fini()
410 * - LDLM_LOCK_PUT(): releases caller reference acquired by
413 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
414 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
416 * - When lock is being cancelled (ldlm_lock_cancel())
417 * - ldlm_lock_destroy()
418 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
421 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
422 * either when lock is cancelled (osc_lock_blocking()), or when locks is
423 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
424 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
428 struct cl_lock_slice ols_cl;
429 /** Internal lock to protect states, etc. */
431 /** Owner sleeps on this channel for state change */
432 struct cl_sync_io *ols_owner;
433 /** waiting list for this lock to be cancelled */
434 struct list_head ols_waiting_list;
435 /** wait entry of ols_waiting_list */
436 struct list_head ols_wait_entry;
437 /** list entry for osc_object::oo_ol_list */
438 struct list_head ols_nextlock_oscobj;
440 /** underlying DLM lock */
441 struct ldlm_lock *ols_dlmlock;
442 /** DLM flags with which osc_lock::ols_lock was enqueued */
444 /** osc_lock::ols_lock handle */
445 struct lustre_handle ols_handle;
446 struct ldlm_enqueue_info ols_einfo;
447 enum osc_lock_state ols_state;
448 /** lock value block */
449 struct ost_lvb ols_lvb;
450 /** Lockless operations to be used by lockless lock */
451 const struct cl_lock_operations *ols_lockless_ops;
453 * true, if ldlm_lock_addref() was called against
454 * osc_lock::ols_lock. This is used for sanity checking.
456 * \see osc_lock::ols_has_ref
458 unsigned ols_hold :1,
460 * this is much like osc_lock::ols_hold, except that this bit is
461 * cleared _after_ reference in released in osc_lock_unuse(). This
462 * fine distinction is needed because:
464 * - if ldlm lock still has a reference, osc_ast_data_get() needs
465 * to return associated cl_lock (so that a flag is needed that is
466 * cleared after ldlm_lock_decref() returned), and
468 * - ldlm_lock_decref() can invoke blocking ast (for a
469 * LDLM_FL_CBPENDING lock), and osc_lock functions like
470 * osc_lock_cancel() called from there need to know whether to
471 * release lock reference (so that a flag is needed that is
472 * cleared before ldlm_lock_decref() is called).
476 * inherit the lockless attribute from top level cl_io.
477 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
481 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
482 * the EVAVAIL error as torerable, this will make upper logic happy
483 * to wait all glimpse locks to each OSTs to be completed.
484 * Glimpse lock converts to normal lock if the server lock is granted.
485 * Glimpse lock should be destroyed immediately after use.
489 * For async glimpse lock.
493 * for speculative locks - asynchronous glimpse locks and ladvise
494 * lockahead manual lock requests
496 * Used to tell osc layer to not wait for the ldlm reply from the
497 * server, so the osc lock will be short lived - It only exists to
498 * create the ldlm request and is not updated on request completion.
503 static inline int osc_lock_is_lockless(const struct osc_lock *ols)
505 return (ols->ols_cl.cls_ops == ols->ols_lockless_ops);
509 * Page state private for osc layer.
512 struct cl_page_slice ops_cl;
514 * Page queues used by osc to detect when RPC can be formed.
516 struct osc_async_page ops_oap;
518 * An offset within page from which next transfer starts. This is used
519 * by cl_page_clip() to submit partial page transfers.
523 * An offset within page at which next transfer ends.
525 * \see osc_page::ops_from.
529 * Boolean, true iff page is under transfer. Used for sanity checking.
531 unsigned ops_transfer_pinned:1,
537 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
541 * If the page is in osc_object::oo_tree.
545 * lru page list. See osc_lru_{del|use}() in osc_page.c for usage.
547 struct list_head ops_lru;
549 * Submit time - the time when the page is starting RPC. For debugging.
551 ktime_t ops_submit_time;
554 struct osc_brw_async_args {
556 int aa_requested_nob;
560 struct brw_page **aa_ppga;
561 struct client_obd *aa_cli;
562 struct list_head aa_oaps;
563 struct list_head aa_exts;
566 extern struct kmem_cache *osc_lock_kmem;
567 extern struct kmem_cache *osc_object_kmem;
568 extern struct kmem_cache *osc_thread_kmem;
569 extern struct kmem_cache *osc_session_kmem;
570 extern struct kmem_cache *osc_extent_kmem;
571 extern struct kmem_cache *osc_quota_kmem;
572 extern struct kmem_cache *osc_obdo_kmem;
574 extern struct lu_context_key osc_key;
575 extern struct lu_context_key osc_session_key;
577 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
580 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
581 struct cl_page *page, pgoff_t ind);
582 void osc_index2policy(union ldlm_policy_data *policy, const struct cl_object *obj,
583 pgoff_t start, pgoff_t end);
584 void osc_lru_add_batch(struct client_obd *cli, struct list_head *list);
585 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
586 enum cl_req_type crt, int brw_flags);
587 int lru_queue_work(const struct lu_env *env, void *data);
588 long osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
589 long target, bool force);
592 int osc_cancel_async_page(const struct lu_env *env, struct osc_page *ops);
593 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
595 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
596 struct page *page, loff_t offset);
597 int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
598 struct osc_page *ops);
599 int osc_page_cache_add(const struct lu_env *env,
600 const struct cl_page_slice *slice, struct cl_io *io);
601 int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj,
602 struct osc_page *ops);
603 int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
604 struct osc_page *ops);
605 int osc_queue_sync_pages(const struct lu_env *env, const struct cl_io *io,
606 struct osc_object *obj, struct list_head *list,
608 int osc_cache_truncate_start(const struct lu_env *env, struct osc_object *obj,
609 __u64 size, struct osc_extent **extp);
610 void osc_cache_truncate_end(const struct lu_env *env, struct osc_extent *ext);
611 int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
612 pgoff_t start, pgoff_t end, int hp, int discard);
613 int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
614 pgoff_t start, pgoff_t end);
615 int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
616 struct osc_object *osc, int async);
617 void osc_wake_cache_waiters(struct client_obd *cli);
619 static inline int osc_io_unplug_async(const struct lu_env *env,
620 struct client_obd *cli,
621 struct osc_object *osc)
623 return osc_io_unplug0(env, cli, osc, 1);
626 static inline void osc_io_unplug(const struct lu_env *env,
627 struct client_obd *cli,
628 struct osc_object *osc)
630 (void)osc_io_unplug0(env, cli, osc, 0);
633 typedef int (*osc_page_gang_cbt)(const struct lu_env *, struct cl_io *,
634 struct osc_page *, void *);
635 int osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io,
636 struct osc_object *osc, pgoff_t start, pgoff_t end,
637 osc_page_gang_cbt cb, void *cbdata);
638 int osc_discard_cb(const struct lu_env *env, struct cl_io *io,
639 struct osc_page *ops, void *cbdata);
642 int osc_device_init(const struct lu_env *env, struct lu_device *d,
643 const char *name, struct lu_device *next);
644 struct lu_device *osc_device_fini(const struct lu_env *env,
645 struct lu_device *d);
646 struct lu_device *osc_device_free(const struct lu_env *env,
647 struct lu_device *d);
650 int osc_object_init(const struct lu_env *env, struct lu_object *obj,
651 const struct lu_object_conf *conf);
652 void osc_object_free(const struct lu_env *env, struct lu_object *obj);
653 int osc_lvb_print(const struct lu_env *env, void *cookie,
654 lu_printer_t p, const struct ost_lvb *lvb);
655 int osc_object_print(const struct lu_env *env, void *cookie,
656 lu_printer_t p, const struct lu_object *obj);
657 int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
658 struct cl_attr *attr);
659 int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
660 const struct cl_attr *attr, unsigned valid);
661 int osc_object_glimpse(const struct lu_env *env, const struct cl_object *obj,
662 struct ost_lvb *lvb);
663 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc);
664 int osc_object_is_contended(struct osc_object *obj);
665 int osc_object_find_cbdata(const struct lu_env *env, struct cl_object *obj,
666 ldlm_iterator_t iter, void *data);
667 int osc_object_prune(const struct lu_env *env, struct cl_object *obj);
670 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd);
671 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg);
672 int osc_precleanup_common(struct obd_device *obd);
673 int osc_cleanup_common(struct obd_device *obd);
674 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
675 u32 keylen, void *key, u32 vallen, void *val,
676 struct ptlrpc_request_set *set);
677 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
678 struct hlist_node *hnode, void *arg);
679 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
680 struct obd_device *obd, struct obd_uuid *cluuid,
681 struct obd_connect_data *data, void *localdata);
682 int osc_disconnect(struct obd_export *exp);
683 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
684 obd_enqueue_update_f upcall, void *cookie);
687 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
688 enum cl_req_type crt, struct cl_2queue *queue);
689 int osc_io_commit_async(const struct lu_env *env,
690 const struct cl_io_slice *ios,
691 struct cl_page_list *qin, int from, int to,
693 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios);
694 void osc_io_iter_fini(const struct lu_env *env,
695 const struct cl_io_slice *ios);
696 int osc_io_write_iter_init(const struct lu_env *env,
697 const struct cl_io_slice *ios);
698 void osc_io_write_iter_fini(const struct lu_env *env,
699 const struct cl_io_slice *ios);
700 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios);
701 void osc_io_setattr_end(const struct lu_env *env,
702 const struct cl_io_slice *slice);
703 int osc_io_read_start(const struct lu_env *env,
704 const struct cl_io_slice *slice);
705 int osc_io_write_start(const struct lu_env *env,
706 const struct cl_io_slice *slice);
707 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice);
708 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
709 struct cl_fsync_io *fio);
710 void osc_io_fsync_end(const struct lu_env *env,
711 const struct cl_io_slice *slice);
712 void osc_read_ahead_release(const struct lu_env *env, void *cbdata);
715 void osc_lock_to_lockless(const struct lu_env *env, struct osc_lock *ols,
717 void osc_lock_wake_waiters(const struct lu_env *env, struct osc_object *osc,
718 struct osc_lock *oscl);
719 int osc_lock_enqueue_wait(const struct lu_env *env, struct osc_object *obj,
720 struct osc_lock *oscl);
721 void osc_lock_set_writer(const struct lu_env *env, const struct cl_io *io,
722 struct cl_object *obj, struct osc_lock *oscl);
723 int osc_lock_print(const struct lu_env *env, void *cookie,
724 lu_printer_t p, const struct cl_lock_slice *slice);
725 void osc_lock_cancel(const struct lu_env *env,
726 const struct cl_lock_slice *slice);
727 void osc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice);
728 int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data);
729 unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock);
731 /*****************************************************************************
733 * Accessors and type conversions.
736 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
738 struct osc_thread_info *info;
740 info = lu_context_key_get(&env->le_ctx, &osc_key);
741 LASSERT(info != NULL);
745 static inline struct osc_session *osc_env_session(const struct lu_env *env)
747 struct osc_session *ses;
749 ses = lu_context_key_get(env->le_ses, &osc_session_key);
750 LASSERT(ses != NULL);
754 static inline struct osc_io *osc_env_io(const struct lu_env *env)
756 return &osc_env_session(env)->os_io;
759 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
761 return container_of0(d, struct osc_device, od_cl.cd_lu_dev);
764 static inline struct obd_export *osc_export(const struct osc_object *obj)
766 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
769 static inline struct client_obd *osc_cli(const struct osc_object *obj)
771 return &osc_export(obj)->exp_obd->u.cli;
774 static inline struct osc_object *cl2osc(const struct cl_object *obj)
776 return container_of0(obj, struct osc_object, oo_cl);
779 static inline struct cl_object *osc2cl(const struct osc_object *obj)
781 return (struct cl_object *)&obj->oo_cl;
784 static inline struct osc_device *obd2osc_dev(const struct obd_device *d)
786 return container_of0(d->obd_lu_dev, struct osc_device,
790 static inline struct lu_device *osc2lu_dev(struct osc_device *osc)
792 return &osc->od_cl.cd_lu_dev;
795 static inline struct lu_object *osc2lu(struct osc_object *osc)
797 return &osc->oo_cl.co_lu;
800 static inline struct osc_object *lu2osc(const struct lu_object *obj)
802 return container_of0(obj, struct osc_object, oo_cl.co_lu);
805 static inline struct osc_io *cl2osc_io(const struct lu_env *env,
806 const struct cl_io_slice *slice)
808 struct osc_io *oio = container_of0(slice, struct osc_io, oi_cl);
810 LINVRNT(oio == osc_env_io(env));
814 static inline enum ldlm_mode osc_cl_lock2ldlm(enum cl_lock_mode mode)
816 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
817 if (mode == CLM_READ)
819 if (mode == CLM_WRITE)
824 static inline enum cl_lock_mode osc_ldlm2cl_lock(enum ldlm_mode mode)
826 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
834 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
836 return container_of0(slice, struct osc_page, ops_cl);
839 static inline struct osc_page *oap2osc(struct osc_async_page *oap)
841 return container_of0(oap, struct osc_page, ops_oap);
844 static inline pgoff_t osc_index(struct osc_page *opg)
846 return opg->ops_cl.cpl_index;
849 static inline struct cl_page *oap2cl_page(struct osc_async_page *oap)
851 return oap2osc(oap)->ops_cl.cpl_page;
854 static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
856 return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
859 static inline struct osc_page *
860 osc_cl_page_osc(struct cl_page *page, struct osc_object *osc)
862 const struct cl_page_slice *slice;
864 LASSERT(osc != NULL);
865 slice = cl_object_page_slice(&osc->oo_cl, page);
866 return cl2osc_page(slice);
869 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
871 return container_of0(slice, struct osc_lock, ols_cl);
874 static inline int osc_io_srvlock(struct osc_io *oio)
876 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
879 enum osc_extent_state {
880 OES_INV = 0, /** extent is just initialized or destroyed */
881 OES_ACTIVE = 1, /** process is using this extent */
882 OES_CACHE = 2, /** extent is ready for IO */
883 OES_LOCKING = 3, /** locking page to prepare IO */
884 OES_LOCK_DONE = 4, /** locking finished, ready to send */
885 OES_RPC = 5, /** in RPC */
886 OES_TRUNC = 6, /** being truncated */
891 * osc_extent data to manage dirty pages.
892 * osc_extent has the following attributes:
893 * 1. all pages in the same must be in one RPC in write back;
894 * 2. # of pages must be less than max_pages_per_rpc - implied by 1;
895 * 3. must be covered by only 1 osc_lock;
896 * 4. exclusive. It's impossible to have overlapped osc_extent.
898 * The lifetime of an extent is from when the 1st page is dirtied to when
899 * all pages inside it are written out.
903 * page lock -> client_obd_list_lock -> object lock(osc_object::oo_lock)
906 /** red-black tree node */
907 struct rb_node oe_node;
908 /** osc_object of this extent */
909 struct osc_object *oe_obj;
910 /** refcount, removed from red-black tree if reaches zero. */
912 /** busy if non-zero */
914 /** link list of osc_object's oo_{hp|urgent|locking}_exts. */
915 struct list_head oe_link;
916 /** state of this extent */
917 enum osc_extent_state oe_state;
918 /** flags for this extent. */
919 unsigned int oe_intree:1,
920 /** 0 is write, 1 is read */
922 /** sync extent, queued by osc_queue_sync_pages() */
924 /** set if this extent has partial, sync pages.
925 * Extents with partial page(s) can't merge with others in RPC */
929 /** an ACTIVE extent is going to be truncated, so when this extent
930 * is released, it will turn into TRUNC state instead of CACHE. */
932 /** this extent should be written asap and someone may wait for the
933 * write to finish. This bit is usually set along with urgent if
934 * the extent was CACHE state.
935 * fsync_wait extent can't be merged because new extent region may
936 * exceed fsync range. */
938 /** covering lock is being canceled */
940 /** this extent should be written back asap. set if one of pages is
941 * called by page WB daemon, or sync write or reading requests. */
943 /** Non-delay RPC should be used for this extent. */
945 /** how many grants allocated for this extent.
946 * Grant allocated for this extent. There is no grant allocated
947 * for reading extents and sync write extents. */
948 unsigned int oe_grants;
949 /** # of dirty pages in this extent */
950 unsigned int oe_nr_pages;
951 /** list of pending oap pages. Pages in this list are NOT sorted. */
952 struct list_head oe_pages;
953 /** Since an extent has to be written out in atomic, this is used to
954 * remember the next page need to be locked to write this extent out.
955 * Not used right now.
957 struct osc_page *oe_next_page;
958 /** start and end index of this extent, include start and end
959 * themselves. Page offset here is the page index of osc_pages.
960 * oe_start is used as keyword for red-black tree. */
963 /** maximum ending index of this extent, this is limited by
964 * max_pages_per_rpc, lock extent and chunk size. */
966 /** waitqueue - for those who want to be notified if this extent's
967 * state has changed. */
968 wait_queue_head_t oe_waitq;
969 /** lock covering this extent */
970 struct ldlm_lock *oe_dlmlock;
971 /** terminator of this extent. Must be true if this extent is in IO. */
972 struct task_struct *oe_owner;
973 /** return value of writeback. If somebody is waiting for this extent,
974 * this value can be known by outside world. */
976 /** max pages per rpc when this extent was created */
977 unsigned int oe_mppr;
978 /** FLR: layout version when this osc_extent is publised */
979 __u32 oe_layout_version;
984 #endif /* LUSTRE_OSC_H */