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/
32 * lustre/include/lustre_osc.h
34 * OSC layer structures and methods common for both OSC and MDC.
36 * This file contains OSC interfaces used by OSC and MDC. Most of them
37 * were just moved from lustre/osc/osc_cl_internal.h for Data-on-MDT
40 * Author: Nikita Danilov <nikita.danilov@sun.com>
41 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
48 #include <libcfs/libcfs.h>
50 #include <cl_object.h>
51 #include <lustre_crypto.h>
57 enum oap_async_flags {
58 /* ap_make_ready will not be called before page is added to an rpc */
60 ASYNC_URGENT = 0x2, /* page must be put into RPC before return */
61 /* ap_refresh_count will not be called to give the caller a chance to
62 * update or cancel the size of the io
64 ASYNC_COUNT_STABLE = 0x4,
70 /* add explicit padding to keep fields aligned despite "packed",
71 * which is needed to pack with following field in osc_page
73 #define OAP_PAD_BITS (16 - OBD_BRW_WRITE - OAP_ASYNC_BITS)
74 struct osc_async_page {
75 unsigned short oap_page_off /* :PAGE_SHIFT */;
76 unsigned int oap_cmd:OBD_BRW_WRITE;
77 enum oap_async_flags oap_async_flags:OAP_ASYNC_BITS;
78 unsigned int oap_padding1:OAP_PAD_BITS; /* unused */
79 unsigned int oap_padding2; /* unused */
81 struct list_head oap_pending_item;
82 struct list_head oap_rpc_item;
86 struct osc_object *oap_obj;
88 struct brw_page oap_brw_page;
89 } __attribute__((packed));
91 #define oap_page oap_brw_page.bp_page
92 #define oap_count oap_brw_page.bp_count
93 #define oap_brw_flags oap_brw_page.bp_flag
95 static inline struct osc_async_page *brw_page2oap(struct brw_page *pga)
97 BUILD_BUG_ON(OAP_ASYNC_MAX - 1 >= (1 << OAP_ASYNC_BITS));
98 return container_of(pga, struct osc_async_page, oap_brw_page);
102 struct cl_device osc_cl;
103 struct obd_export *osc_exp;
105 /* 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 */
112 /* configuration item(s) */
113 time64_t osc_contention_time;
119 * State maintained by osc layer for each IO context.
123 struct cl_io_slice oi_cl;
124 /** true if this io is lockless. */
125 unsigned int oi_lockless:1,
126 /** true if this io is counted as active IO */
128 /** true if this io has CAP_SYS_RESOURCE */
129 oi_cap_sys_resource:1,
130 /** true if this io issued by readahead */
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
138 struct osc_extent *oi_active;
139 /** partially truncated extent, we need to hold this extent to prevent
140 * page writeback from happening.
142 struct osc_extent *oi_trunc;
143 /** write osc_lock for this IO, used by osc_extent_find(). */
144 struct osc_lock *oi_write_osclock;
145 struct osc_lock *oi_read_osclock;
147 struct osc_async_cbargs {
150 struct completion opc_sync;
155 * State maintained by osc layer for the duration of a system call.
161 #define OTI_PVEC_SIZE 256
162 struct osc_thread_info {
163 struct ldlm_res_id oti_resname;
164 union ldlm_policy_data oti_policy;
165 struct cl_attr oti_attr;
167 struct folio_batch oti_fbatch;
168 void *oti_pvec[OTI_PVEC_SIZE];
170 * Fields used by cl_lock_discard_pages().
172 pgoff_t oti_next_index;
173 pgoff_t oti_fn_index; /* first non-overlapped index */
174 pgoff_t oti_ng_index; /* negative lock caching */
175 struct cl_sync_io oti_anchor;
176 struct cl_req_attr oti_req_attr;
177 struct lu_buf oti_ladvise_buf;
180 static inline __u64 osc_enq2ldlm_flags(__u32 enqflags)
184 CDEBUG(D_DLMTRACE, "flags: %x\n", enqflags);
186 LASSERT((enqflags & ~CEF_MASK) == 0);
188 if (enqflags & CEF_NONBLOCK)
189 result |= LDLM_FL_BLOCK_NOWAIT;
190 if (enqflags & CEF_GLIMPSE)
191 result |= LDLM_FL_HAS_INTENT|LDLM_FL_CBPENDING;
192 if (enqflags & CEF_DISCARD_DATA)
193 result |= LDLM_FL_AST_DISCARD_DATA;
194 if (enqflags & CEF_PEEK)
195 result |= LDLM_FL_TEST_LOCK;
196 if (enqflags & CEF_LOCK_MATCH)
197 result |= LDLM_FL_MATCH_LOCK;
198 if (enqflags & CEF_LOCK_NO_EXPAND)
199 result |= LDLM_FL_NO_EXPANSION;
200 if (enqflags & CEF_SPECULATIVE)
201 result |= LDLM_FL_SPECULATIVE;
205 typedef int (*osc_enqueue_upcall_f)(void *cookie, struct lustre_handle *lockh,
208 struct osc_enqueue_args {
209 struct obd_export *oa_exp;
210 enum ldlm_type oa_type;
211 enum ldlm_mode oa_mode;
213 osc_enqueue_upcall_f oa_upcall;
215 struct ost_lvb *oa_lvb;
216 struct lustre_handle oa_lockh;
221 * Bit flags for osc_dlm_lock_at_pageoff().
225 * Just check if the desired lock exists, it won't hold reference
228 OSC_DAP_FL_TEST_LOCK = BIT(0),
230 * Return the lock even if it is being canceled.
232 OSC_DAP_FL_CANCELING = BIT(1),
234 * check ast data is present, requested to cancel cb
236 OSC_DAP_FL_AST = BIT(2),
238 * look at right region for the desired lock
240 OSC_DAP_FL_RIGHT = BIT(3),
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. used to protect oo_{read|write}_pages soon. */
300 * Radix tree for caching pages
302 spinlock_t oo_tree_lock;
303 struct radix_tree_root oo_tree;
304 unsigned long oo_npages;
306 /* Protect osc_lock this osc_object has */
307 struct list_head oo_ol_list;
308 spinlock_t oo_ol_spin;
310 /** number of active IOs of this object */
312 wait_queue_head_t oo_io_waitq;
314 const struct osc_object_operations *oo_obj_ops;
318 static inline void osc_build_res_name(struct osc_object *osc,
319 struct ldlm_res_id *resname)
321 return osc->oo_obj_ops->oto_build_res_name(osc, resname);
324 static inline struct ldlm_lock *osc_dlmlock_at_pgoff(const struct lu_env *env,
325 struct osc_object *obj,
327 enum osc_dap_flags flags)
329 return obj->oo_obj_ops->oto_dlmlock_at_pgoff(env, obj, index, flags);
332 static inline void osc_object_lock(struct osc_object *obj)
334 spin_lock(&obj->oo_lock);
337 static inline int osc_object_trylock(struct osc_object *obj)
339 return spin_trylock(&obj->oo_lock);
342 static inline void osc_object_unlock(struct osc_object *obj)
344 spin_unlock(&obj->oo_lock);
347 #define assert_osc_object_is_locked(obj) \
348 assert_spin_locked(&obj->oo_lock)
350 static inline void osc_object_set_contended(struct osc_object *obj)
352 obj->oo_contention_time = ktime_get();
354 obj->oo_contended = 1;
357 static inline void osc_object_clear_contended(struct osc_object *obj)
359 obj->oo_contended = 0;
363 * Lock "micro-states" for osc layer.
365 enum osc_lock_state {
374 * osc-private state of cl_lock.
376 * Interaction with DLM.
378 * Once receive upcall is invoked, osc_lock remembers a handle of DLM lock in
379 * osc_lock::ols_handle and a pointer to that lock in osc_lock::ols_dlmlock.
381 * This pointer is protected through a reference, acquired by
382 * osc_lock_upcall0(). Also, an additional reference is acquired by
383 * ldlm_lock_addref() call protecting the lock from cancellation, until
384 * osc_lock_unuse() releases it.
386 * Below is a description of how lock references are acquired and released
389 * - When new lock is created and enqueued to the server (ldlm_cli_enqueue())
390 * - ldlm_lock_create()
391 * - ldlm_lock_new(): initializes a lock with 2 references. One for
392 * the caller (released when reply from the server is received, or on
393 * error), and another for the hash table.
394 * - ldlm_lock_addref_internal(): protects the lock from cancellation.
396 * - When reply is received from the server (osc_enqueue_interpret())
397 * - ldlm_cli_enqueue_fini()
398 * - LDLM_LOCK_PUT(): releases caller reference acquired by
401 * ldlm_lock_decref(): error case: matches ldlm_cli_enqueue().
402 * - ldlm_lock_decref(): for async locks, matches ldlm_cli_enqueue().
404 * - When lock is being cancelled (ldlm_lock_cancel())
405 * - ldlm_lock_destroy()
406 * - LDLM_LOCK_PUT(): releases hash-table reference acquired by
409 * osc_lock is detached from ldlm_lock by osc_lock_detach() that is called
410 * either when lock is cancelled (osc_lock_blocking()), or when locks is
411 * deleted without cancellation (e.g., from cl_locks_prune()). In the latter
412 * case ldlm lock remains in memory, and can be re-attached to osc_lock in the
416 struct cl_lock_slice ols_cl;
417 /** Internal lock to protect states, etc. */
419 /** Owner sleeps on this channel for state change */
420 struct cl_sync_io *ols_owner;
421 /** waiting list for this lock to be cancelled */
422 struct list_head ols_waiting_list;
423 /** wait entry of ols_waiting_list */
424 struct list_head ols_wait_entry;
425 /** list entry for osc_object::oo_ol_list */
426 struct list_head ols_nextlock_oscobj;
428 /** underlying DLM lock */
429 struct ldlm_lock *ols_dlmlock;
430 /** DLM flags with which osc_lock::ols_lock was enqueued */
432 /** osc_lock::ols_lock handle */
433 struct lustre_handle ols_handle;
434 struct ldlm_enqueue_info ols_einfo;
435 enum osc_lock_state ols_state;
436 /** lock value block */
437 struct ost_lvb ols_lvb;
438 /** Lockless operations to be used by lockless lock */
439 const struct cl_lock_operations *ols_lockless_ops;
441 * true, if ldlm_lock_addref() was called against
442 * osc_lock::ols_lock. This is used for sanity checking.
444 * \see osc_lock::ols_has_ref
446 unsigned ols_hold :1,
448 * this is much like osc_lock::ols_hold, except that this bit is
449 * cleared _after_ reference in released in osc_lock_unuse(). This
450 * fine distinction is needed because:
452 * - if ldlm lock still has a reference, osc_ast_data_get() needs
453 * to return associated cl_lock (so that a flag is needed that is
454 * cleared after ldlm_lock_decref() returned), and
456 * - ldlm_lock_decref() can invoke blocking ast (for a
457 * LDLM_FL_CBPENDING lock), and osc_lock functions like
458 * osc_lock_cancel() called from there need to know whether to
459 * release lock reference (so that a flag is needed that is
460 * cleared before ldlm_lock_decref() is called).
464 * inherit the lockless attribute from top level cl_io.
465 * If true, osc_lock_enqueue is able to tolerate the -EUSERS error.
469 * if set, the osc_lock is a glimpse lock. For glimpse locks, we treat
470 * the EVAVAIL error as torerable, this will make upper logic happy
471 * to wait all glimpse locks to each OSTs to be completed.
472 * Glimpse lock converts to normal lock if the server lock is granted.
473 * Glimpse lock should be destroyed immediately after use.
477 * For async glimpse lock.
481 * for speculative locks - asynchronous glimpse locks and ladvise
482 * lockahead manual lock requests
484 * Used to tell osc layer to not wait for the ldlm reply from the
485 * server, so the osc lock will be short lived - It only exists to
486 * create the ldlm request and is not updated on request completion.
491 static inline int osc_lock_is_lockless(const struct osc_lock *ols)
493 return (ols->ols_cl.cls_ops == ols->ols_lockless_ops);
497 * Page state private for osc layer.
500 struct cl_page_slice ops_cl;
502 * Page queues used by osc to detect when RPC can be formed.
504 struct osc_async_page ops_oap;
506 * An offset within page from which next transfer starts. This is used
507 * by cl_page_clip() to submit partial page transfers.
509 unsigned int ops_from:PAGE_SHIFT,
511 * An offset within page at which next transfer ends(inclusive).
513 * \see osc_page::ops_from.
517 * Boolean, true iff page is under transfer. Used for sanity checking.
519 ops_transfer_pinned:1,
525 * Set if the page must be transferred with OBD_BRW_SRVLOCK.
529 * If the page is in osc_object::oo_tree.
533 * lru page list. See osc_lru_{del|use}() in osc_page.c for usage.
535 struct list_head ops_lru;
538 struct osc_brw_async_args {
540 int aa_requested_nob;
544 struct brw_page **aa_ppga;
545 struct client_obd *aa_cli;
546 struct list_head aa_oaps;
547 struct list_head aa_exts;
548 struct ptlrpc_request *aa_request;
551 extern struct kmem_cache *osc_lock_kmem;
552 extern struct kmem_cache *osc_object_kmem;
553 extern struct kmem_cache *osc_thread_kmem;
554 extern struct kmem_cache *osc_session_kmem;
555 extern struct kmem_cache *osc_extent_kmem;
556 extern struct kmem_cache *osc_obdo_kmem;
558 extern struct lu_context_key osc_key;
559 extern struct lu_context_key osc_session_key;
561 #define OSC_FLAGS (ASYNC_URGENT|ASYNC_READY)
564 int osc_page_init(const struct lu_env *env, struct cl_object *obj,
565 struct cl_page *page, pgoff_t ind);
566 void osc_index2policy(union ldlm_policy_data *policy,
567 const struct cl_object *obj, pgoff_t start, pgoff_t end);
568 void osc_lru_add_batch(struct client_obd *cli, struct list_head *list);
569 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
570 enum cl_req_type crt, int brw_flags);
571 int lru_queue_work(const struct lu_env *env, void *data);
572 long osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
573 long target, bool force);
576 int osc_set_async_flags(struct osc_object *obj, struct osc_page *opg,
578 int osc_prep_async_page(struct osc_object *osc, struct osc_page *ops,
579 struct cl_page *page, loff_t offset);
580 int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
581 struct osc_object *osc, struct osc_page *ops,
583 int osc_page_cache_add(const struct lu_env *env, struct osc_object *osc,
584 struct osc_page *opg, struct cl_io *io,
586 int osc_teardown_async_page(const struct lu_env *env, struct osc_object *obj,
587 struct osc_page *ops);
588 int osc_flush_async_page(const struct lu_env *env, struct cl_io *io,
589 struct osc_page *ops);
590 int osc_queue_sync_pages(const struct lu_env *env, struct cl_io *io,
591 struct osc_object *obj, struct list_head *list,
593 int osc_cache_truncate_start(const struct lu_env *env, struct osc_object *obj,
594 __u64 size, struct osc_extent **extp);
595 void osc_cache_truncate_end(const struct lu_env *env, struct osc_extent *ext);
596 int osc_cache_writeback_range(const struct lu_env *env, struct osc_object *obj,
597 pgoff_t start, pgoff_t end, int hp, int discard);
598 int osc_cache_wait_range(const struct lu_env *env, struct osc_object *obj,
599 pgoff_t start, pgoff_t end);
600 int osc_io_unplug0(const struct lu_env *env, struct client_obd *cli,
601 struct osc_object *osc, int async);
602 static inline void osc_wake_cache_waiters(struct client_obd *cli)
604 wake_up(&cli->cl_cache_waiters);
607 static inline int osc_io_unplug_async(const struct lu_env *env,
608 struct client_obd *cli,
609 struct osc_object *osc)
611 return osc_io_unplug0(env, cli, osc, 1);
614 static inline void osc_io_unplug(const struct lu_env *env,
615 struct client_obd *cli,
616 struct osc_object *osc)
618 (void)osc_io_unplug0(env, cli, osc, 0);
621 typedef bool (*osc_page_gang_cbt)(const struct lu_env *, struct cl_io *,
622 void**, int, void *);
623 bool osc_page_gang_lookup(const struct lu_env *env, struct cl_io *io,
624 struct osc_object *osc, pgoff_t start, pgoff_t end,
625 osc_page_gang_cbt cb, void *cbdata);
626 bool osc_discard_cb(const struct lu_env *env, struct cl_io *io,
627 void **pvec, int count, void *cbdata);
630 int osc_device_init(const struct lu_env *env, struct lu_device *d,
631 const char *name, struct lu_device *next);
632 struct lu_device *osc_device_fini(const struct lu_env *env,
633 struct lu_device *d);
634 struct lu_device *osc_device_free(const struct lu_env *env,
635 struct lu_device *d);
638 int osc_object_init(const struct lu_env *env, struct lu_object *obj,
639 const struct lu_object_conf *conf);
640 void osc_object_free(const struct lu_env *env, struct lu_object *obj);
641 int osc_lvb_print(const struct lu_env *env, void *cookie,
642 lu_printer_t p, const struct ost_lvb *lvb);
643 int osc_object_print(const struct lu_env *env, void *cookie,
644 lu_printer_t p, const struct lu_object *obj);
645 int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
646 struct cl_attr *attr);
647 int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
648 const struct cl_attr *attr, unsigned int valid);
649 int osc_object_glimpse(const struct lu_env *env, const struct cl_object *obj,
650 struct ost_lvb *lvb);
651 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc);
652 int osc_object_find_cbdata(const struct lu_env *env, struct cl_object *obj,
653 ldlm_iterator_t iter, void *data);
654 int osc_object_prune(const struct lu_env *env, struct cl_object *obj);
657 void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd);
658 int osc_setup_common(struct obd_device *obd, struct lustre_cfg *lcfg);
659 int osc_precleanup_common(struct obd_device *obd);
660 int osc_cleanup_common(struct obd_device *obd);
661 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
662 u32 keylen, void *key, u32 vallen, void *val,
663 struct ptlrpc_request_set *set);
664 int osc_ldlm_resource_invalidate(struct cfs_hash *hs, struct cfs_hash_bd *bd,
665 struct hlist_node *hnode, void *arg);
666 int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
667 struct obd_device *obd, struct obd_uuid *cluuid,
668 struct obd_connect_data *data, void *localdata);
669 int osc_disconnect(struct obd_export *exp);
670 int osc_punch_send(struct obd_export *exp, struct obdo *oa,
671 obd_enqueue_update_f upcall, void *cookie);
672 int osc_fallocate_base(struct obd_export *exp, struct obdo *oa,
673 obd_enqueue_update_f upcall, void *cookie, int mode);
674 void osc_update_next_shrink(struct client_obd *cli);
675 void osc_schedule_grant_work(void);
678 int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
679 enum cl_req_type crt, struct cl_2queue *queue);
680 int osc_io_commit_async(const struct lu_env *env,
681 const struct cl_io_slice *ios,
682 struct cl_page_list *qin, int from, int to,
684 void osc_io_extent_release(const struct lu_env *env,
685 const struct cl_io_slice *ios);
686 int osc_io_iter_init(const struct lu_env *env, const struct cl_io_slice *ios);
687 void osc_io_iter_fini(const struct lu_env *env,
688 const struct cl_io_slice *ios);
689 void osc_io_rw_iter_fini(const struct lu_env *env,
690 const struct cl_io_slice *ios);
691 int osc_io_fault_start(const struct lu_env *env, const struct cl_io_slice *ios);
692 void osc_io_setattr_end(const struct lu_env *env,
693 const struct cl_io_slice *slice);
694 int osc_io_read_start(const struct lu_env *env,
695 const struct cl_io_slice *slice);
696 int osc_io_write_start(const struct lu_env *env,
697 const struct cl_io_slice *slice);
698 void osc_io_end(const struct lu_env *env, const struct cl_io_slice *slice);
699 int osc_fsync_ost(const struct lu_env *env, struct osc_object *obj,
700 struct cl_fsync_io *fio);
701 void osc_io_fsync_end(const struct lu_env *env,
702 const struct cl_io_slice *slice);
703 void osc_read_ahead_release(const struct lu_env *env, struct cl_read_ahead *ra);
704 int osc_io_lseek_start(const struct lu_env *env,
705 const struct cl_io_slice *slice);
706 void osc_io_lseek_end(const struct lu_env *env,
707 const struct cl_io_slice *slice);
708 int osc_io_lru_reserve(const struct lu_env *env, const struct cl_io_slice *ios,
709 loff_t pos, size_t count);
710 int osc_punch_start(const struct lu_env *env, struct cl_io *io,
711 struct cl_object *obj);
714 void osc_lock_to_lockless(const struct lu_env *env, struct osc_lock *ols,
716 void osc_lock_wake_waiters(const struct lu_env *env, struct osc_object *osc,
717 struct osc_lock *oscl);
718 int osc_lock_enqueue_wait(const struct lu_env *env, struct osc_object *obj,
719 struct osc_lock *oscl);
720 void osc_lock_set_writer(const struct lu_env *env, const struct cl_io *io,
721 struct cl_object *obj, struct osc_lock *oscl);
722 void osc_lock_set_reader(const struct lu_env *env, const struct cl_io *io,
723 struct cl_object *obj, struct osc_lock *oscl);
724 int osc_lock_print(const struct lu_env *env, void *cookie,
725 lu_printer_t p, const struct cl_lock_slice *slice);
726 void osc_lock_cancel(const struct lu_env *env,
727 const struct cl_lock_slice *slice);
728 void osc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice);
729 int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data);
730 unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock);
732 /* Accessors and type conversions. */
733 static inline struct osc_thread_info *osc_env_info(const struct lu_env *env)
735 struct osc_thread_info *info;
737 info = lu_context_key_get(&env->le_ctx, &osc_key);
738 LASSERT(info != NULL);
742 static inline struct osc_session *osc_env_session(const struct lu_env *env)
744 struct osc_session *ses;
746 ses = lu_context_key_get(env->le_ses, &osc_session_key);
747 LASSERT(ses != NULL);
751 static inline struct osc_io *osc_env_io(const struct lu_env *env)
753 return &osc_env_session(env)->os_io;
756 static inline struct osc_device *lu2osc_dev(const struct lu_device *d)
758 return container_of_safe(d, struct osc_device, osc_cl.cd_lu_dev);
761 static inline struct obd_export *osc_export(const struct osc_object *obj)
763 return lu2osc_dev(obj->oo_cl.co_lu.lo_dev)->osc_exp;
766 static inline struct client_obd *osc_cli(const struct osc_object *obj)
768 return &osc_export(obj)->exp_obd->u.cli;
771 static inline struct osc_object *cl2osc(const struct cl_object *obj)
773 return container_of_safe(obj, struct osc_object, oo_cl);
776 static inline struct cl_object *osc2cl(const struct osc_object *obj)
778 return (struct cl_object *)&obj->oo_cl;
781 static inline struct osc_device *obd2osc_dev(const struct obd_device *obd)
783 return container_of_safe(obd->obd_lu_dev, struct osc_device,
787 static inline struct lu_device *osc2lu_dev(struct osc_device *osc)
789 return &osc->osc_cl.cd_lu_dev;
792 static inline struct lu_object *osc2lu(struct osc_object *osc)
794 return &osc->oo_cl.co_lu;
797 static inline struct osc_object *lu2osc(const struct lu_object *obj)
799 return container_of_safe(obj, struct osc_object, oo_cl.co_lu);
802 static inline struct osc_io *cl2osc_io(const struct lu_env *env,
803 const struct cl_io_slice *slice)
805 struct osc_io *oio = container_of(slice, struct osc_io, oi_cl);
807 LINVRNT(oio == osc_env_io(env));
811 static inline enum ldlm_mode osc_cl_lock2ldlm(enum cl_lock_mode mode)
813 LASSERT(mode == CLM_READ || mode == CLM_WRITE || mode == CLM_GROUP);
814 if (mode == CLM_READ)
816 if (mode == CLM_WRITE)
821 static inline enum cl_lock_mode osc_ldlm2cl_lock(enum ldlm_mode mode)
823 LASSERT(mode == LCK_PR || mode == LCK_PW || mode == LCK_GROUP);
831 static inline struct osc_page *cl2osc_page(const struct cl_page_slice *slice)
833 return container_of_safe(slice, struct osc_page, ops_cl);
836 static inline struct osc_page *oap2osc(struct osc_async_page *oap)
838 return container_of_safe(oap, struct osc_page, ops_oap);
841 static inline pgoff_t osc_index(struct osc_page *opg)
843 return opg->ops_oap.oap_obj_off >> PAGE_SHIFT;
846 static inline struct osc_object *osc_page_object(struct osc_page *ops)
848 return ops->ops_oap.oap_obj;
851 static inline struct cl_page *oap2cl_page(struct osc_async_page *oap)
853 return oap2osc(oap)->ops_cl.cpl_page;
856 static inline struct osc_page *oap2osc_page(struct osc_async_page *oap)
858 return (struct osc_page *)container_of(oap, struct osc_page, ops_oap);
861 static inline struct osc_page *
862 osc_cl_page_osc(struct cl_page *page, struct osc_object *osc)
864 const struct cl_page_slice *slice;
866 LASSERT(osc != NULL);
867 slice = cl_object_page_slice(&osc->oo_cl, page);
868 return cl2osc_page(slice);
871 static inline struct osc_lock *cl2osc_lock(const struct cl_lock_slice *slice)
873 return container_of_safe(slice, struct osc_lock, ols_cl);
876 static inline int osc_io_srvlock(struct osc_io *oio)
878 return (oio->oi_lockless && !oio->oi_cl.cis_io->ci_no_srvlock);
881 enum osc_extent_state {
882 OES_INV = 0, /** extent is just initialized or destroyed */
883 OES_ACTIVE = 1, /** process is using this extent */
884 OES_CACHE = 2, /** extent is ready for IO */
885 OES_LOCKING = 3, /** locking page to prepare IO */
886 OES_LOCK_DONE = 4, /** locking finished, ready to send */
887 OES_RPC = 5, /** in RPC */
888 OES_TRUNC = 6, /** being truncated */
893 * osc_extent data to manage dirty pages.
894 * osc_extent has the following attributes:
895 * 1. all pages in the same must be in one RPC in write back;
896 * 2. # of pages must be less than max_pages_per_rpc - implied by 1;
897 * 3. must be covered by only 1 osc_lock;
898 * 4. exclusive. It's impossible to have overlapped osc_extent.
900 * The lifetime of an extent is from when the 1st page is dirtied to when
901 * all pages inside it are written out.
905 * page lock -> client_obd_list_lock -> object lock(osc_object::oo_lock)
908 /** red-black tree node */
909 struct rb_node oe_node;
910 /** osc_object of this extent */
911 struct osc_object *oe_obj;
912 /** refcount, removed from red-black tree if reaches zero. */
914 /** busy if non-zero */
916 /** link list of osc_object's oo_{hp|urgent|locking}_exts. */
917 struct list_head oe_link;
918 /** state of this extent */
919 enum osc_extent_state oe_state;
920 /** flags for this extent. */
921 /** 0 is write, 1 is read */
922 unsigned int oe_rw:1,
923 /** sync extent, queued by osc_queue_sync_pages() */
925 /** set if this extent has partial, sync pages.
926 * Extents with partial page(s) can't merge with others in RPC
931 /** an ACTIVE extent is going to be truncated, so when this extent
932 * is released, it will turn into TRUNC state instead of CACHE.
935 /** this extent should be written asap and someone may wait for the
936 * write to finish. This bit is usually set along with urgent if
937 * the extent was CACHE state.
938 * fsync_wait extent can't be merged because new extent region may
939 * exceed fsync range.
942 /** covering lock is being canceled */
944 /** this extent should be written back asap. set if one of pages is
945 * called by page WB daemon, or sync write or reading requests.
948 /** Non-delay RPC should be used for this extent. */
950 /** direct IO pages */
952 /** this extent consists of pages that are not directly accessible
956 /** how many grants allocated for this extent.
957 * Grant allocated for this extent. There is no grant allocated
958 * for reading extents and sync write extents.
960 unsigned int oe_grants;
961 /** # of dirty pages in this extent */
962 unsigned int oe_nr_pages;
963 /** list of pending oap pages. Pages in this list are NOT sorted. */
964 struct list_head oe_pages;
965 /** start and end index of this extent, include start and end
966 * themselves. Page offset here is the page index of osc_pages.
967 * oe_start is used as keyword for red-black tree.
971 /** maximum ending index of this extent, this is limited by
972 * max_pages_per_rpc, lock extent and chunk size.
975 /** waitqueue - for those who want to be notified if this extent's
978 wait_queue_head_t oe_waitq;
979 /** lock covering this extent */
980 struct ldlm_lock *oe_dlmlock;
981 /** terminator of this extent. Must be true if this extent is in IO. */
982 struct task_struct *oe_owner;
983 /** return value of writeback. If somebody is waiting for this extent,
984 * this value can be known by outside world.
987 /** max pages per rpc when this extent was created */
988 unsigned int oe_mppr;
989 /** FLR: layout version when this osc_extent is publised */
990 __u32 oe_layout_version;
995 #endif /* LUSTRE_OSC_H */