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>
52 #include <lustre_crypto.h>
58 struct osc_quota_info {
59 /** linkage for quota hash table */
60 struct hlist_node oqi_hash;
65 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
66 page is added to an rpc */
67 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
68 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
69 to give the caller a chance to update
70 or cancel the size of the io */
74 struct osc_async_page {
76 unsigned short oap_cmd;
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);
104 struct cl_device od_cl;
105 struct obd_export *od_exp;
107 /* Write stats is actually protected by client_obd's lock. */
109 uint64_t os_lockless_writes; /* by bytes */
110 uint64_t os_lockless_reads; /* by bytes */
111 uint64_t os_lockless_truncates; /* by times */
114 /* configuration item(s) */
115 time64_t od_contention_time;
116 int od_lockless_truncate;
122 * State maintained by osc layer for each IO context.
126 struct cl_io_slice oi_cl;
127 /** true if this io is lockless. */
128 unsigned int oi_lockless:1,
129 /** true if this io is counted as active IO */
131 /** true if this io has CAP_SYS_RESOURCE */
132 oi_cap_sys_resource:1;
133 /** how many LRU pages are reserved for this IO */
134 unsigned long oi_lru_reserved;
136 /** active extents, we know how many bytes is going to be written,
137 * so having an active extent will prevent it from being fragmented */
138 struct osc_extent *oi_active;
139 /** partially truncated extent, we need to hold this extent to prevent
140 * page writeback from happening. */
141 struct osc_extent *oi_trunc;
142 /** write osc_lock for this IO, used by osc_extent_find(). */
143 struct osc_lock *oi_write_osclock;
145 struct osc_async_cbargs {
148 struct completion opc_sync;
153 * State maintained by osc layer for the duration of a system call.
159 #define OTI_PVEC_SIZE 256
160 struct osc_thread_info {
161 struct ldlm_res_id oti_resname;
162 union ldlm_policy_data oti_policy;
163 struct cl_attr oti_attr;
165 struct pagevec oti_pagevec;
166 void *oti_pvec[OTI_PVEC_SIZE];
168 * Fields used by cl_lock_discard_pages().
170 pgoff_t oti_next_index;
171 pgoff_t oti_fn_index; /* first non-overlapped index */
172 struct cl_sync_io oti_anchor;
173 struct cl_req_attr oti_req_attr;
174 struct lu_buf oti_ladvise_buf;
177 static inline __u64 osc_enq2ldlm_flags(__u32 enqflags)
181 CDEBUG(D_DLMTRACE, "flags: %x\n", enqflags);
183 LASSERT((enqflags & ~CEF_MASK) == 0);
185 if (enqflags & CEF_NONBLOCK)
186 result |= LDLM_FL_BLOCK_NOWAIT;
187 if (enqflags & CEF_GLIMPSE)
188 result |= LDLM_FL_HAS_INTENT;
189 if (enqflags & CEF_DISCARD_DATA)
190 result |= LDLM_FL_AST_DISCARD_DATA;
191 if (enqflags & CEF_PEEK)
192 result |= LDLM_FL_TEST_LOCK;
193 if (enqflags & CEF_LOCK_MATCH)
194 result |= LDLM_FL_MATCH_LOCK;
195 if (enqflags & CEF_LOCK_NO_EXPAND)
196 result |= LDLM_FL_NO_EXPANSION;
197 if (enqflags & CEF_SPECULATIVE)
198 result |= LDLM_FL_SPECULATIVE;
202 typedef int (*osc_enqueue_upcall_f)(void *cookie, struct lustre_handle *lockh,
205 struct osc_enqueue_args {
206 struct obd_export *oa_exp;
207 enum ldlm_type oa_type;
208 enum ldlm_mode oa_mode;
210 osc_enqueue_upcall_f oa_upcall;
212 struct ost_lvb *oa_lvb;
213 struct lustre_handle oa_lockh;
218 * Bit flags for osc_dlm_lock_at_pageoff().
222 * Just check if the desired lock exists, it won't hold reference
225 OSC_DAP_FL_TEST_LOCK = BIT(0),
227 * Return the lock even if it is being canceled.
229 OSC_DAP_FL_CANCELING = BIT(1),
231 * check ast data is present, requested to cancel cb
233 OSC_DAP_FL_AST = BIT(2),
237 * The set of operations which are different for MDC and OSC objects
239 struct osc_object_operations {
240 void (*oto_build_res_name)(struct osc_object *osc,
241 struct ldlm_res_id *resname);
242 struct ldlm_lock* (*oto_dlmlock_at_pgoff)(const struct lu_env *env,
243 struct osc_object *obj,
245 enum osc_dap_flags dap_flags);
249 struct cl_object oo_cl;
250 struct lov_oinfo *oo_oinfo;
252 * True if locking against this stripe got -EUSERS.
255 ktime_t oo_contention_time;
256 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
258 * IO context used for invariant checks in osc_lock_has_pages().
260 struct cl_io oo_debug_io;
261 /** Serialization object for osc_object::oo_debug_io. */
262 struct mutex oo_debug_mutex;
265 * used by the osc to keep track of what objects to build into rpcs.
266 * Protected by client_obd->cli_loi_list_lock.
268 struct list_head oo_ready_item;
269 struct list_head oo_hp_ready_item;
270 struct list_head oo_write_item;
271 struct list_head oo_read_item;
274 * extent is a red black tree to manage (async) dirty pages.
276 struct rb_root oo_root;
278 * Manage write(dirty) extents.
280 struct list_head oo_hp_exts; /* list of hp extents */
281 struct list_head oo_urgent_exts; /* list of writeback extents */
282 struct list_head oo_full_exts;
284 struct list_head oo_reading_exts;
286 atomic_t oo_nr_reads;
287 atomic_t oo_nr_writes;
289 /** Protect extent tree. Will be used to protect
290 * oo_{read|write}_pages soon. */
294 * Radix tree for caching pages
296 spinlock_t oo_tree_lock;
297 struct radix_tree_root oo_tree;
298 unsigned long oo_npages;
300 /* Protect osc_lock this osc_object has */
301 struct list_head oo_ol_list;
302 spinlock_t oo_ol_spin;
304 /** number of active IOs of this object */
306 wait_queue_head_t oo_io_waitq;
308 const struct osc_object_operations *oo_obj_ops;
312 static inline void osc_build_res_name(struct osc_object *osc,
313 struct ldlm_res_id *resname)
315 return osc->oo_obj_ops->oto_build_res_name(osc, resname);
318 static inline struct ldlm_lock *osc_dlmlock_at_pgoff(const struct lu_env *env,
319 struct osc_object *obj,
321 enum osc_dap_flags flags)
323 return obj->oo_obj_ops->oto_dlmlock_at_pgoff(env, obj, index, flags);
326 static inline void osc_object_lock(struct osc_object *obj)
328 spin_lock(&obj->oo_lock);
331 static inline int osc_object_trylock(struct osc_object *obj)
333 return spin_trylock(&obj->oo_lock);
336 static inline void osc_object_unlock(struct osc_object *obj)
338 spin_unlock(&obj->oo_lock);
341 #define assert_osc_object_is_locked(obj) \
342 assert_spin_locked(&obj->oo_lock)
344 static inline void osc_object_set_contended(struct osc_object *obj)
346 obj->oo_contention_time = ktime_get();
348 obj->oo_contended = 1;
351 static inline void osc_object_clear_contended(struct osc_object *obj)
353 obj->oo_contended = 0;
357 * Lock "micro-states" for osc layer.
359 enum osc_lock_state {
368 * osc-private state of cl_lock.
370 * Interaction with DLM.
372 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
373 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_dlmlock.
375 * This pointer is protected through a reference, acquired by
376 * osc_lock_upcall0(). Also, an additional reference is acquired by
377 * ldlm_lock_addref() call protecting the lock from cancellation, until
378 * osc_lock_unuse() releases it.
380 * Below is a description of how lock references are acquired and released
383 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
384 * - ldlm_lock_create()
385 * - ldlm_lock_new(): initializes a lock with 2 references. One for
386 * the caller (released when reply from the server is received, or on
387 * error), and another for the hash table.
388 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
390 * - When reply is received from the server (osc_enqueue_interpret())
391 * - ldlm_cli_enqueue_fini()
392 * - LDLM_LOCK_PUT(): releases caller reference acquired by
395 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
396 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
398 * - When lock is being cancelled (ldlm_lock_cancel())
399 * - ldlm_lock_destroy()
400 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
403 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
404 * either when lock is cancelled (osc_lock_blocking()), or when locks is
405 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
406 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
410 struct cl_lock_slice ols_cl;
411 /** Internal lock to protect states, etc. */
413 /** Owner sleeps on this channel for state change */
414 struct cl_sync_io *ols_owner;
415 /** waiting list for this lock to be cancelled */
416 struct list_head ols_waiting_list;
417 /** wait entry of ols_waiting_list */
418 struct list_head ols_wait_entry;
419 /** list entry for osc_object::oo_ol_list */
420 struct list_head ols_nextlock_oscobj;
422 /** underlying DLM lock */
423 struct ldlm_lock *ols_dlmlock;
424 /** DLM flags with which osc_lock::ols_lock was enqueued */
426 /** osc_lock::ols_lock handle */
427 struct lustre_handle ols_handle;
428 struct ldlm_enqueue_info ols_einfo;
429 enum osc_lock_state ols_state;
430 /** lock value block */
431 struct ost_lvb ols_lvb;
432 /** Lockless operations to be used by lockless lock */
433 const struct cl_lock_operations *ols_lockless_ops;
435 * true, if ldlm_lock_addref() was called against
436 * osc_lock::ols_lock. This is used for sanity checking.
438 * \see osc_lock::ols_has_ref
440 unsigned ols_hold :1,
442 * this is much like osc_lock::ols_hold, except that this bit is
443 * cleared _after_ reference in released in osc_lock_unuse(). This
444 * fine distinction is needed because:
446 * - if ldlm lock still has a reference, osc_ast_data_get() needs
447 * to return associated cl_lock (so that a flag is needed that is
448 * cleared after ldlm_lock_decref() returned), and
450 * - ldlm_lock_decref() can invoke blocking ast (for a
451 * LDLM_FL_CBPENDING lock), and osc_lock functions like
452 * osc_lock_cancel() called from there need to know whether to
453 * release lock reference (so that a flag is needed that is
454 * cleared before ldlm_lock_decref() is called).
458 * inherit the lockless attribute from top level cl_io.
459 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
463 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
464 * the EVAVAIL error as torerable, this will make upper logic happy
465 * to wait all glimpse locks to each OSTs to be completed.
466 * Glimpse lock converts to normal lock if the server lock is granted.
467 * Glimpse lock should be destroyed immediately after use.
471 * For async glimpse lock.
475 * for speculative locks - asynchronous glimpse locks and ladvise
476 * lockahead manual lock requests
478 * Used to tell osc layer to not wait for the ldlm reply from the
479 * server, so the osc lock will be short lived - It only exists to
480 * create the ldlm request and is not updated on request completion.
485 static inline int osc_lock_is_lockless(const struct osc_lock *ols)
487 return (ols->ols_cl.cls_ops == ols->ols_lockless_ops);
491 * Page state private for osc layer.
494 struct cl_page_slice ops_cl;
496 * Page queues used by osc to detect when RPC can be formed.
498 struct osc_async_page ops_oap;
500 * An offset within page from which next transfer starts. This is used
501 * by cl_page_clip() to submit partial page transfers.
503 unsigned int ops_from:PAGE_SHIFT,
505 * An offset within page at which next transfer ends(inclusive).
507 * \see osc_page::ops_from.
511 * Boolean, true iff page is under transfer. Used for sanity checking.
513 ops_transfer_pinned:1,
519 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
523 * If the page is in osc_object::oo_tree.
527 * lru page list. See osc_lru_{del|use}() in osc_page.c for usage.
529 struct list_head ops_lru;
531 * Submit time - the time when the page is starting RPC. For debugging.
533 ktime_t ops_submit_time;
536 struct osc_brw_async_args {
538 int aa_requested_nob;
542 struct brw_page **aa_ppga;
543 struct client_obd *aa_cli;
544 struct list_head aa_oaps;
545 struct list_head aa_exts;
548 extern struct kmem_cache *osc_lock_kmem;
549 extern struct kmem_cache *osc_object_kmem;
550 extern struct kmem_cache *osc_thread_kmem;
551 extern struct kmem_cache *osc_session_kmem;
552 extern struct kmem_cache *osc_extent_kmem;
553 extern struct kmem_cache *osc_quota_kmem;
554 extern struct kmem_cache *osc_obdo_kmem;
556 extern struct lu_context_key osc_key;
557 extern struct lu_context_key osc_session_key;
559 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
562 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
563 struct cl_page *page, pgoff_t ind);
564 void osc_index2policy(union ldlm_policy_data *policy, const struct cl_object *obj,
565 pgoff_t start, pgoff_t end);
566 void osc_lru_add_batch(struct client_obd *cli, struct list_head *list);
567 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
568 enum cl_req_type crt, int brw_flags);
569 int lru_queue_work(const struct lu_env *env, void *data);
570 long osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
571 long target, bool force);
574 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
576 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
577 struct page *page, loff_t offset);
578 int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
579 struct osc_page *ops, cl_commit_cbt cb);
580 int osc_page_cache_add(const struct lu_env *env, struct osc_page *opg,
581 struct cl_io *io, cl_commit_cbt cb);
582 int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj,
583 struct osc_page *ops);
584 int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
585 struct osc_page *ops);
586 int osc_queue_sync_pages(const struct lu_env *env, const struct cl_io *io,
587 struct osc_object *obj, struct list_head *list,
589 int osc_cache_truncate_start(const struct lu_env *env, struct osc_object *obj,
590 __u64 size, struct osc_extent **extp);
591 void osc_cache_truncate_end(const struct lu_env *env, struct osc_extent *ext);
592 int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
593 pgoff_t start, pgoff_t end, int hp, int discard);
594 int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
595 pgoff_t start, pgoff_t end);
596 int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
597 struct osc_object *osc, int async);
598 static inline void osc_wake_cache_waiters(struct client_obd *cli)
600 wake_up(&cli->cl_cache_waiters);
603 static inline int osc_io_unplug_async(const struct lu_env *env,
604 struct client_obd *cli,
605 struct osc_object *osc)
607 return osc_io_unplug0(env, cli, osc, 1);
610 static inline void osc_io_unplug(const struct lu_env *env,
611 struct client_obd *cli,
612 struct osc_object *osc)
614 (void)osc_io_unplug0(env, cli, osc, 0);
617 typedef bool (*osc_page_gang_cbt)(const struct lu_env *, struct cl_io *,
618 struct osc_page *, void *);
619 bool osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io,
620 struct osc_object *osc, pgoff_t start, pgoff_t end,
621 osc_page_gang_cbt cb, void *cbdata);
622 bool osc_discard_cb(const struct lu_env *env, struct cl_io *io,
623 struct osc_page *ops, void *cbdata);
626 int osc_device_init(const struct lu_env *env, struct lu_device *d,
627 const char *name, struct lu_device *next);
628 struct lu_device *osc_device_fini(const struct lu_env *env,
629 struct lu_device *d);
630 struct lu_device *osc_device_free(const struct lu_env *env,
631 struct lu_device *d);
634 int osc_object_init(const struct lu_env *env, struct lu_object *obj,
635 const struct lu_object_conf *conf);
636 void osc_object_free(const struct lu_env *env, struct lu_object *obj);
637 int osc_lvb_print(const struct lu_env *env, void *cookie,
638 lu_printer_t p, const struct ost_lvb *lvb);
639 int osc_object_print(const struct lu_env *env, void *cookie,
640 lu_printer_t p, const struct lu_object *obj);
641 int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
642 struct cl_attr *attr);
643 int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
644 const struct cl_attr *attr, unsigned valid);
645 int osc_object_glimpse(const struct lu_env *env, const struct cl_object *obj,
646 struct ost_lvb *lvb);
647 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc);
648 int osc_object_is_contended(struct osc_object *obj);
649 int osc_object_find_cbdata(const struct lu_env *env, struct cl_object *obj,
650 ldlm_iterator_t iter, void *data);
651 int osc_object_prune(const struct lu_env *env, struct cl_object *obj);
654 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd);
655 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg);
656 int osc_precleanup_common(struct obd_device *obd);
657 int osc_cleanup_common(struct obd_device *obd);
658 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
659 u32 keylen, void *key, u32 vallen, void *val,
660 struct ptlrpc_request_set *set);
661 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
662 struct hlist_node *hnode, void *arg);
663 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
664 struct obd_device *obd, struct obd_uuid *cluuid,
665 struct obd_connect_data *data, void *localdata);
666 int osc_disconnect(struct obd_export *exp);
667 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
668 obd_enqueue_update_f upcall, void *cookie);
671 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
672 enum cl_req_type crt, struct cl_2queue *queue);
673 int osc_io_commit_async(const struct lu_env *env,
674 const struct cl_io_slice *ios,
675 struct cl_page_list *qin, int from, int to,
677 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios);
678 void osc_io_iter_fini(const struct lu_env *env,
679 const struct cl_io_slice *ios);
680 int osc_io_rw_iter_init(const struct lu_env *env,
681 const struct cl_io_slice *ios);
682 void osc_io_rw_iter_fini(const struct lu_env *env,
683 const struct cl_io_slice *ios);
684 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios);
685 void osc_io_setattr_end(const struct lu_env *env,
686 const struct cl_io_slice *slice);
687 int osc_io_read_start(const struct lu_env *env,
688 const struct cl_io_slice *slice);
689 int osc_io_write_start(const struct lu_env *env,
690 const struct cl_io_slice *slice);
691 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice);
692 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
693 struct cl_fsync_io *fio);
694 void osc_io_fsync_end(const struct lu_env *env,
695 const struct cl_io_slice *slice);
696 void osc_read_ahead_release(const struct lu_env *env, void *cbdata);
699 void osc_lock_to_lockless(const struct lu_env *env, struct osc_lock *ols,
701 void osc_lock_wake_waiters(const struct lu_env *env, struct osc_object *osc,
702 struct osc_lock *oscl);
703 int osc_lock_enqueue_wait(const struct lu_env *env, struct osc_object *obj,
704 struct osc_lock *oscl);
705 void osc_lock_set_writer(const struct lu_env *env, const struct cl_io *io,
706 struct cl_object *obj, struct osc_lock *oscl);
707 int osc_lock_print(const struct lu_env *env, void *cookie,
708 lu_printer_t p, const struct cl_lock_slice *slice);
709 void osc_lock_cancel(const struct lu_env *env,
710 const struct cl_lock_slice *slice);
711 void osc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice);
712 int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data);
713 unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock);
715 /*****************************************************************************
717 * Accessors and type conversions.
720 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
722 struct osc_thread_info *info;
724 info = lu_context_key_get(&env->le_ctx, &osc_key);
725 LASSERT(info != NULL);
729 static inline struct osc_session *osc_env_session(const struct lu_env *env)
731 struct osc_session *ses;
733 ses = lu_context_key_get(env->le_ses, &osc_session_key);
734 LASSERT(ses != NULL);
738 static inline struct osc_io *osc_env_io(const struct lu_env *env)
740 return &osc_env_session(env)->os_io;
743 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
745 return container_of_safe(d, struct osc_device, od_cl.cd_lu_dev);
748 static inline struct obd_export *osc_export(const struct osc_object *obj)
750 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->od_exp;
753 static inline struct client_obd *osc_cli(const struct osc_object *obj)
755 return &osc_export(obj)->exp_obd->u.cli;
758 static inline struct osc_object *cl2osc(const struct cl_object *obj)
760 return container_of_safe(obj, struct osc_object, oo_cl);
763 static inline struct cl_object *osc2cl(const struct osc_object *obj)
765 return (struct cl_object *)&obj->oo_cl;
768 static inline struct osc_device *obd2osc_dev(const struct obd_device *obd)
770 return container_of_safe(obd->obd_lu_dev, struct osc_device,
774 static inline struct lu_device *osc2lu_dev(struct osc_device *osc)
776 return &osc->od_cl.cd_lu_dev;
779 static inline struct lu_object *osc2lu(struct osc_object *osc)
781 return &osc->oo_cl.co_lu;
784 static inline struct osc_object *lu2osc(const struct lu_object *obj)
786 return container_of_safe(obj, struct osc_object, oo_cl.co_lu);
789 static inline struct osc_io *cl2osc_io(const struct lu_env *env,
790 const struct cl_io_slice *slice)
792 struct osc_io *oio = container_of(slice, struct osc_io, oi_cl);
794 LINVRNT(oio == osc_env_io(env));
798 static inline enum ldlm_mode osc_cl_lock2ldlm(enum cl_lock_mode mode)
800 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
801 if (mode == CLM_READ)
803 if (mode == CLM_WRITE)
808 static inline enum cl_lock_mode osc_ldlm2cl_lock(enum ldlm_mode mode)
810 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
818 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
820 return container_of_safe(slice, struct osc_page, ops_cl);
823 static inline struct osc_page *oap2osc(struct osc_async_page *oap)
825 return container_of_safe(oap, struct osc_page, ops_oap);
828 static inline pgoff_t osc_index(struct osc_page *opg)
830 return opg->ops_cl.cpl_page->cp_osc_index;
833 static inline struct cl_page *oap2cl_page(struct osc_async_page *oap)
835 return oap2osc(oap)->ops_cl.cpl_page;
838 static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
840 return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
843 static inline struct osc_page *
844 osc_cl_page_osc(struct cl_page *page, struct osc_object *osc)
846 const struct cl_page_slice *slice;
848 LASSERT(osc != NULL);
849 slice = cl_object_page_slice(&osc->oo_cl, page);
850 return cl2osc_page(slice);
853 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
855 return container_of_safe(slice, struct osc_lock, ols_cl);
858 static inline int osc_io_srvlock(struct osc_io *oio)
860 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
863 enum osc_extent_state {
864 OES_INV = 0, /** extent is just initialized or destroyed */
865 OES_ACTIVE = 1, /** process is using this extent */
866 OES_CACHE = 2, /** extent is ready for IO */
867 OES_LOCKING = 3, /** locking page to prepare IO */
868 OES_LOCK_DONE = 4, /** locking finished, ready to send */
869 OES_RPC = 5, /** in RPC */
870 OES_TRUNC = 6, /** being truncated */
875 * osc_extent data to manage dirty pages.
876 * osc_extent has the following attributes:
877 * 1. all pages in the same must be in one RPC in write back;
878 * 2. # of pages must be less than max_pages_per_rpc - implied by 1;
879 * 3. must be covered by only 1 osc_lock;
880 * 4. exclusive. It's impossible to have overlapped osc_extent.
882 * The lifetime of an extent is from when the 1st page is dirtied to when
883 * all pages inside it are written out.
887 * page lock -> client_obd_list_lock -> object lock(osc_object::oo_lock)
890 /** red-black tree node */
891 struct rb_node oe_node;
892 /** osc_object of this extent */
893 struct osc_object *oe_obj;
894 /** refcount, removed from red-black tree if reaches zero. */
896 /** busy if non-zero */
898 /** link list of osc_object's oo_{hp|urgent|locking}_exts. */
899 struct list_head oe_link;
900 /** state of this extent */
901 enum osc_extent_state oe_state;
902 /** flags for this extent. */
903 /** 0 is write, 1 is read */
904 unsigned int oe_rw:1,
905 /** sync extent, queued by osc_queue_sync_pages() */
907 /** set if this extent has partial, sync pages.
908 * Extents with partial page(s) can't merge with others in RPC */
912 /** an ACTIVE extent is going to be truncated, so when this extent
913 * is released, it will turn into TRUNC state instead of CACHE. */
915 /** this extent should be written asap and someone may wait for the
916 * write to finish. This bit is usually set along with urgent if
917 * the extent was CACHE state.
918 * fsync_wait extent can't be merged because new extent region may
919 * exceed fsync range. */
921 /** covering lock is being canceled */
923 /** this extent should be written back asap. set if one of pages is
924 * called by page WB daemon, or sync write or reading requests. */
926 /** Non-delay RPC should be used for this extent. */
928 /** direct IO pages */
930 /** this extent consists of RDMA only pages */
932 /** how many grants allocated for this extent.
933 * Grant allocated for this extent. There is no grant allocated
934 * for reading extents and sync write extents. */
935 unsigned int oe_grants;
936 /** # of dirty pages in this extent */
937 unsigned int oe_nr_pages;
938 /** list of pending oap pages. Pages in this list are NOT sorted. */
939 struct list_head oe_pages;
940 /** start and end index of this extent, include start and end
941 * themselves. Page offset here is the page index of osc_pages.
942 * oe_start is used as keyword for red-black tree. */
945 /** maximum ending index of this extent, this is limited by
946 * max_pages_per_rpc, lock extent and chunk size. */
948 /** waitqueue - for those who want to be notified if this extent's
949 * state has changed. */
950 wait_queue_head_t oe_waitq;
951 /** lock covering this extent */
952 struct ldlm_lock *oe_dlmlock;
953 /** terminator of this extent. Must be true if this extent is in IO. */
954 struct task_struct *oe_owner;
955 /** return value of writeback. If somebody is waiting for this extent,
956 * this value can be known by outside world. */
958 /** max pages per rpc when this extent was created */
959 unsigned int oe_mppr;
960 /** FLR: layout version when this osc_extent is publised */
961 __u32 oe_layout_version;
966 #endif /* LUSTRE_OSC_H */