1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 * This file is part of Lustre, http://www.lustre.org/
38 * Lustre is a trademark of Sun Microsystems, Inc.
40 * Internal interfaces of LOV layer.
42 * Author: Nikita Danilov <nikita.danilov@sun.com>
45 #ifndef LOV_CL_INTERNAL_H
46 #define LOV_CL_INTERNAL_H
49 # include <libcfs/libcfs.h>
51 # include <liblustre.h>
55 #include <cl_object.h>
56 #include "lov_internal.h"
58 /** \addtogroup lov lov @{ */
61 * Logical object volume layer. This layer implements data striping (raid0).
63 * At the lov layer top-entity (object, page, lock, io) is connected to one or
64 * more sub-entities: top-object, representing a file is connected to a set of
65 * sub-objects, each representing a stripe, file-level top-lock is connected
66 * to a set of per-stripe sub-locks, top-page is connected to a (single)
67 * sub-page, and a top-level IO is connected to a set of (potentially
68 * concurrent) sub-IO's.
70 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
71 * respectively, while a single sub-lock can be part of multiple top-locks.
73 * Reference counting models are different for different types of entities:
75 * - top-object keeps a reference to its sub-objects, and destroys them
76 * when it is destroyed.
78 * - top-page keeps a reference to its sub-page, and destroys it when it
81 * - sub-lock keep a reference to its top-locks. Top-lock keeps a
82 * reference (and a hold, see cl_lock_hold()) on its sub-locks when it
83 * actively using them (that is, in cl_lock_state::CLS_QUEUING,
84 * cl_lock_state::CLS_ENQUEUED, cl_lock_state::CLS_HELD states). When
85 * moving into cl_lock_state::CLS_CACHED state, top-lock releases a
86 * hold. From this moment top-lock has only a 'weak' reference to its
87 * sub-locks. This reference is protected by top-lock
88 * cl_lock::cll_guard, and will be automatically cleared by the sub-lock
89 * when the latter is destroyed. When a sub-lock is canceled, a
90 * reference to it is removed from the top-lock array, and top-lock is
91 * moved into CLS_NEW state. It is guaranteed that all sub-locks exist
92 * while their top-lock is in CLS_HELD or CLS_CACHED states.
94 * - IO's are not reference counted.
96 * To implement a connection between top and sub entities, lov layer is split
97 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
98 * implementing full set of cl-interfaces. For example, top-object has vvp and
99 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
100 * used to track child-parent relationship.
105 struct lovsub_device;
106 struct lovsub_object;
109 enum lov_device_flags {
110 LOV_DEV_INITIALIZED = 1 << 0
118 * Resources that are used in memory-cleaning path, and whose allocation
119 * cannot fail even when memory is tight. They are preallocated in sufficient
120 * quantities in lov_device::ld_emerg[], and access to them is serialized
121 * lov_device::ld_mutex.
123 struct lov_device_emerg {
125 * Page list used to submit IO when memory is in pressure.
127 struct cl_page_list emrg_page_list;
129 * sub-io's shared by all threads accessing this device when memory is
130 * too low to allocate sub-io's dynamically.
132 struct cl_io emrg_subio;
134 * Environments used by sub-io's in
135 * lov_device_emerg::emrg_subio.
137 struct lu_env *emrg_env;
139 * Refchecks for lov_device_emerg::emrg_env.
148 * XXX Locking of lov-private data is missing.
150 struct cl_device ld_cl;
151 struct lov_obd *ld_lov;
152 /** size of lov_device::ld_target[] array */
154 struct lovsub_device **ld_target;
157 /** Emergency resources used in memory-cleansing paths. */
158 struct lov_device_emerg **ld_emrg;
160 * Serializes access to lov_device::ld_emrg in low-memory
163 struct mutex ld_mutex;
169 enum lov_layout_type {
170 /** empty file without body */
180 * lov-specific file state.
182 * lov object has particular layout type, determining how top-object is built
183 * on top of sub-objects. Layout type can change dynamically. When this
184 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
185 * all state pertaining to the old layout type is destroyed, and new state is
186 * constructed. All object methods take said semaphore in the shared mode,
187 * providing serialization against transition between layout types.
189 * To avoid multiple `if' or `switch' statements, selecting behavior for the
190 * current layout type, object methods perform double-dispatch, invoking
191 * function corresponding to the current layout type.
194 struct cl_object lo_cl;
196 * Serializes object operations with transitions between layout types.
198 * This semaphore is taken in shared mode by all object methods, and
199 * is taken in exclusive mode when object type is changed.
201 * \see lov_object::lo_type
203 struct rw_semaphore lo_type_guard;
205 * Type of an object. Protected by lov_object::lo_type_guard.
207 enum lov_layout_type lo_type;
209 union lov_layout_state {
210 struct lov_layout_raid0 {
212 struct lov_stripe_md *lo_lsm;
214 * Array of sub-objects. Allocated when top-object is
215 * created (lov_init_raid0()).
217 * Top-object is a strict master of its sub-objects:
218 * it is created before them, and outlives its
219 * children (this later is necessary so that basic
220 * functions like cl_object_top() always
221 * work). Top-object keeps a reference on every
224 * When top-object is destroyed (lov_delete_raid0())
225 * it releases its reference to a sub-object and waits
226 * until the latter is finally destroyed.
228 struct lovsub_object **lo_sub;
230 * When this is true, lov_object::lo_attr contains
231 * valid up to date attributes for a top-level
232 * object. This field is reset to 0 when attributes of
233 * any sub-object change.
237 * Cached object attribute, built from sub-object
240 struct cl_attr lo_attr;
242 struct lov_layout_state_empty {
244 struct lov_layout_state_join {
248 * Thread that acquired lov_object::lo_type_guard in an exclusive
251 cfs_task_t *lo_owner;
255 * Flags that top-lock can set on each of its sub-locks.
258 /** Top-lock acquired a hold (cl_lock_hold()) on a sub-lock. */
263 * State lov_lock keeps for each sub-lock.
265 struct lov_lock_sub {
266 /** sub-lock itself */
267 struct lovsub_lock *sub_lock;
268 /** An array of per-sub-lock flags, taken from enum lov_sub_flags */
271 struct cl_lock_descr sub_descr;
272 struct cl_lock_descr sub_got;
276 * lov-specific lock state.
279 struct cl_lock_slice lls_cl;
280 /** Number of sub-locks in this lock */
283 * Number of existing sub-locks.
285 unsigned lls_nr_filled;
287 * Set when sub-lock was canceled, while top-lock was being
292 * An array of sub-locks
294 * There are two issues with managing sub-locks:
296 * - sub-locks are concurrently canceled, and
298 * - sub-locks are shared with other top-locks.
300 * To manage cancellation, top-lock acquires a hold on a sublock
301 * (lov_sublock_adopt()) when the latter is inserted into
302 * lov_lock::lls_sub[]. This hold is released (lov_sublock_release())
303 * when top-lock is going into CLS_CACHED state or destroyed. Hold
304 * prevents sub-lock from cancellation.
306 * Sub-lock sharing means, among other things, that top-lock that is
307 * in the process of creation (i.e., not yet inserted into lock list)
308 * is already accessible to other threads once at least one of its
309 * sub-locks is created, see lov_lock_sub_init().
311 * Sub-lock can be in one of the following states:
313 * - doesn't exist, lov_lock::lls_sub[]::sub_lock == NULL. Such
314 * sub-lock was either never created (top-lock is in CLS_NEW
315 * state), or it was created, then canceled, then destroyed
316 * (lov_lock_unlink() cleared sub-lock pointer in the top-lock).
318 * - sub-lock exists and is on
319 * hold. (lov_lock::lls_sub[]::sub_flags & LSF_HELD). This is a
320 * normal state of a sub-lock in CLS_HELD and CLS_CACHED states
323 * - sub-lock exists, but is not held by the top-lock. This
324 * happens after top-lock released a hold on sub-locks before
325 * going into cache (lov_lock_unuse()).
327 * \todo To support wide-striping, array has to be replaced with a set
328 * of queues to avoid scanning.
330 struct lov_lock_sub *lls_sub;
332 * Original description with which lock was enqueued.
334 struct cl_lock_descr lls_orig;
338 struct cl_page_slice lps_cl;
346 struct lovsub_device {
347 struct cl_device acid_cl;
348 struct lov_device *acid_super;
350 struct cl_device *acid_next;
353 struct lovsub_object {
354 struct cl_object_header lso_header;
355 struct cl_object lso_cl;
356 struct lov_object *lso_super;
361 * A link between a top-lock and a sub-lock. Separate data-structure is
362 * necessary, because top-locks and sub-locks are in M:N relationship.
364 * \todo This can be optimized for a (by far) most frequent case of a single
365 * top-lock per sub-lock.
367 struct lov_lock_link {
368 struct lov_lock *lll_super;
369 /** An index within parent lock. */
372 * A linkage into per sub-lock list of all corresponding top-locks,
373 * hanging off lovsub_lock::lss_parents.
375 struct list_head lll_list;
379 * Lock state at lovsub layer.
382 struct cl_lock_slice lss_cl;
384 * List of top-locks that have given sub-lock as their part. Protected
385 * by cl_lock::cll_guard mutex.
387 struct list_head lss_parents;
389 * Top-lock that initiated current operation on this sub-lock. This is
390 * only set during top-to-bottom lock operations like enqueue, and is
391 * used to optimize state change notification. Protected by
392 * cl_lock::cll_guard mutex.
394 * \see lovsub_lock_state_one().
396 struct cl_lock *lss_active;
400 * Describe the environment settings for sublocks.
402 struct lov_sublock_env {
403 const struct lu_env *lse_env;
404 struct cl_io *lse_io;
405 struct lov_io_sub *lse_sub;
409 struct cl_page_slice lsb_cl;
413 struct lov_thread_info {
414 struct cl_object_conf lti_stripe_conf;
415 struct lu_fid lti_fid;
416 struct cl_lock_descr lti_ldescr;
417 struct ost_lvb lti_lvb;
418 struct cl_2queue lti_cl2q;
419 union lov_layout_state lti_state;
420 struct cl_lock_closure lti_closure;
421 cfs_waitlink_t lti_waiter;
425 * State that lov_io maintains for every sub-io.
430 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
431 * independently, with lov acting as a scheduler to maximize overall
434 struct cl_io *sub_io;
436 * Linkage into a list (hanging off lov_io::lis_active) of all
437 * sub-io's active for the current IO iteration.
439 struct list_head sub_linkage;
441 * true, iff cl_io_init() was successfully executed against
442 * lov_io_sub::sub_io.
444 int sub_io_initialized;
446 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
447 * allocated, but borrowed from a per-device emergency pool.
451 * environment, in which sub-io executes.
453 struct lu_env *sub_env;
455 * environment's refcheck.
466 * IO state private for LOV.
470 struct cl_io_slice lis_cl;
472 * Pointer to the object slice. This is a duplicate of
473 * lov_io::lis_cl::cis_object.
475 struct lov_object *lis_object;
477 * Original end-of-io position for this IO, set by the upper layer as
478 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
479 * changes pos and count to fit IO into a single stripe and uses saved
480 * value to determine when IO iterations have to stop.
482 * This is used only for CIT_READ and CIT_WRITE io's.
484 loff_t lis_io_endpos;
487 * starting position within a file, for the current io loop iteration
488 * (stripe), used by ci_io_loop().
492 * end position with in a file, for the current stripe io. This is
493 * exclusive (i.e., next offset after last byte affected by io).
498 int lis_stripe_count;
499 int lis_active_subios;
502 * the index of ls_single_subio in ls_subios array
504 int lis_single_subio_index;
505 struct cl_io lis_single_subio;
508 * size of ls_subios array, actually the highest stripe #
511 struct lov_io_sub *lis_subs;
513 * List of active sub-io's.
515 struct list_head lis_active;
520 struct lov_sublock_env ls_subenv;
524 * State of transfer for lov.
527 struct cl_req_slice lr_cl;
531 * State of transfer for lovsub.
534 struct cl_req_slice lsrq_cl;
537 extern struct lu_device_type lov_device_type;
538 extern struct lu_device_type lovsub_device_type;
540 extern struct lu_context_key lov_key;
541 extern struct lu_context_key lov_session_key;
543 extern cfs_mem_cache_t *lov_page_kmem;
544 extern cfs_mem_cache_t *lov_lock_kmem;
545 extern cfs_mem_cache_t *lov_object_kmem;
546 extern cfs_mem_cache_t *lov_thread_kmem;
547 extern cfs_mem_cache_t *lov_session_kmem;
548 extern cfs_mem_cache_t *lov_req_kmem;
550 extern cfs_mem_cache_t *lovsub_page_kmem;
551 extern cfs_mem_cache_t *lovsub_lock_kmem;
552 extern cfs_mem_cache_t *lovsub_object_kmem;
553 extern cfs_mem_cache_t *lovsub_req_kmem;
555 extern cfs_mem_cache_t *lov_lock_link_kmem;
557 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
558 const struct lu_object_conf *conf);
559 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
560 const struct lu_object_conf *conf);
561 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
562 struct cl_lock *lock, const struct cl_io *io);
563 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
565 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
566 struct cl_lock *lock, const struct cl_io *io);
568 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
569 struct cl_lock *lock, const struct cl_io *io);
570 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
572 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
574 void lov_lock_unlink (const struct lu_env *env, struct lov_lock_link *link,
575 struct lovsub_lock *sub);
577 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
579 void lov_sub_put (struct lov_io_sub *sub);
580 int lov_sublock_modify (const struct lu_env *env, struct lov_lock *lov,
581 struct lovsub_lock *sublock,
582 const struct cl_lock_descr *d, int idx);
585 struct cl_page *lov_page_init (const struct lu_env *env, struct cl_object *ob,
586 struct cl_page *page, cfs_page_t *vmpage);
587 struct cl_page *lovsub_page_init(const struct lu_env *env, struct cl_object *ob,
588 struct cl_page *page, cfs_page_t *vmpage);
590 struct cl_page *lov_page_init_empty(const struct lu_env *env,
591 struct cl_object *obj,
592 struct cl_page *page, cfs_page_t *vmpage);
593 struct cl_page *lov_page_init_raid0(const struct lu_env *env,
594 struct cl_object *obj,
595 struct cl_page *page, cfs_page_t *vmpage);
596 struct lu_object *lov_object_alloc (const struct lu_env *env,
597 const struct lu_object_header *hdr,
598 struct lu_device *dev);
599 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
600 const struct lu_object_header *hdr,
601 struct lu_device *dev);
603 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
604 struct lov_lock *lck,
605 struct lovsub_lock *sub);
606 struct lov_io_sub *lov_page_subio (const struct lu_env *env,
608 const struct cl_page_slice *slice);
611 #define lov_foreach_target(lov, var) \
612 for (var = 0; var < lov_targets_nr(lov); ++var)
614 /*****************************************************************************
622 static inline struct lov_session *lov_env_session(const struct lu_env *env)
624 struct lov_session *ses;
626 ses = lu_context_key_get(env->le_ses, &lov_session_key);
627 LASSERT(ses != NULL);
631 static inline struct lov_io *lov_env_io(const struct lu_env *env)
633 return &lov_env_session(env)->ls_io;
636 static inline int lov_is_object(const struct lu_object *obj)
638 return obj->lo_dev->ld_type == &lov_device_type;
641 static inline int lovsub_is_object(const struct lu_object *obj)
643 return obj->lo_dev->ld_type == &lovsub_device_type;
646 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
648 return &lov->ld_cl.cd_lu_dev;
651 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
653 LINVRNT(d->ld_type == &lov_device_type);
654 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
657 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
659 return &lovsub->acid_cl;
662 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
664 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
667 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
669 LINVRNT(d->ld_type == &lovsub_device_type);
670 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
673 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
675 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
676 return container_of0(d, struct lovsub_device, acid_cl);
679 static inline struct lu_object *lov2lu(struct lov_object *lov)
681 return &lov->lo_cl.co_lu;
684 static inline struct cl_object *lov2cl(struct lov_object *lov)
689 static inline struct lov_object *lu2lov(const struct lu_object *obj)
691 LINVRNT(lov_is_object(obj));
692 return container_of0(obj, struct lov_object, lo_cl.co_lu);
695 static inline struct lov_object *cl2lov(const struct cl_object *obj)
697 LINVRNT(lov_is_object(&obj->co_lu));
698 return container_of0(obj, struct lov_object, lo_cl);
701 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
703 return &los->lso_cl.co_lu;
706 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
711 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
713 LINVRNT(lovsub_is_object(&obj->co_lu));
714 return container_of0(obj, struct lovsub_object, lso_cl);
717 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
719 LINVRNT(lovsub_is_object(obj));
720 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
723 static inline struct lovsub_lock *
724 cl2lovsub_lock(const struct cl_lock_slice *slice)
726 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
727 return container_of(slice, struct lovsub_lock, lss_cl);
730 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
732 const struct cl_lock_slice *slice;
734 slice = cl_lock_at(lock, &lovsub_device_type);
735 LASSERT(slice != NULL);
736 return cl2lovsub_lock(slice);
739 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
741 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
742 return container_of(slice, struct lov_lock, lls_cl);
745 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
747 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
748 return container_of0(slice, struct lov_page, lps_cl);
751 static inline struct lov_req *cl2lov_req(const struct cl_req_slice *slice)
753 return container_of0(slice, struct lov_req, lr_cl);
756 static inline struct lovsub_page *
757 cl2lovsub_page(const struct cl_page_slice *slice)
759 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
760 return container_of0(slice, struct lovsub_page, lsb_cl);
763 static inline struct lovsub_req *cl2lovsub_req(const struct cl_req_slice *slice)
765 return container_of0(slice, struct lovsub_req, lsrq_cl);
768 static inline struct cl_page *lov_sub_page(const struct cl_page_slice *slice)
770 return slice->cpl_page->cp_child;
773 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
774 const struct cl_io_slice *ios)
778 lio = container_of(ios, struct lov_io, lis_cl);
779 LASSERT(lio == lov_env_io(env));
783 static inline int lov_targets_nr(const struct lov_device *lov)
785 return lov->ld_lov->desc.ld_tgt_count;
788 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
790 struct lov_thread_info *info;
792 info = lu_context_key_get(&env->le_ctx, &lov_key);
793 LASSERT(info != NULL);
797 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
799 struct lov_layout_raid0 *raid0;
801 LASSERT(lov->lo_type == LLT_RAID0);
802 raid0 = &lov->u.raid0;
803 LASSERT(raid0->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC);