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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Intel Corporation.
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>
43 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
46 #ifndef LOV_CL_INTERNAL_H
47 #define LOV_CL_INTERNAL_H
49 #include "../../include/linux/libcfs/libcfs.h"
51 #include "../include/obd.h"
52 #include "../include/cl_object.h"
53 #include "lov_internal.h"
56 * Logical object volume layer. This layer implements data striping (raid0).
58 * At the lov layer top-entity (object, page, lock, io) is connected to one or
59 * more sub-entities: top-object, representing a file is connected to a set of
60 * sub-objects, each representing a stripe, file-level top-lock is connected
61 * to a set of per-stripe sub-locks, top-page is connected to a (single)
62 * sub-page, and a top-level IO is connected to a set of (potentially
63 * concurrent) sub-IO's.
65 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
66 * respectively, while a single sub-lock can be part of multiple top-locks.
68 * Reference counting models are different for different types of entities:
70 * - top-object keeps a reference to its sub-objects, and destroys them
71 * when it is destroyed.
73 * - top-page keeps a reference to its sub-page, and destroys it when it
76 * - sub-lock keep a reference to its top-locks. Top-lock keeps a
77 * reference (and a hold, see cl_lock_hold()) on its sub-locks when it
78 * actively using them (that is, in cl_lock_state::CLS_QUEUING,
79 * cl_lock_state::CLS_ENQUEUED, cl_lock_state::CLS_HELD states). When
80 * moving into cl_lock_state::CLS_CACHED state, top-lock releases a
81 * hold. From this moment top-lock has only a 'weak' reference to its
82 * sub-locks. This reference is protected by top-lock
83 * cl_lock::cll_guard, and will be automatically cleared by the sub-lock
84 * when the latter is destroyed. When a sub-lock is canceled, a
85 * reference to it is removed from the top-lock array, and top-lock is
86 * moved into CLS_NEW state. It is guaranteed that all sub-locks exist
87 * while their top-lock is in CLS_HELD or CLS_CACHED states.
89 * - IO's are not reference counted.
91 * To implement a connection between top and sub entities, lov layer is split
92 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
93 * implementing full set of cl-interfaces. For example, top-object has vvp and
94 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
95 * used to track child-parent relationship.
100 struct lovsub_device;
101 struct lovsub_object;
104 enum lov_device_flags {
105 LOV_DEV_INITIALIZED = 1 << 0
113 * Resources that are used in memory-cleaning path, and whose allocation
114 * cannot fail even when memory is tight. They are preallocated in sufficient
115 * quantities in lov_device::ld_emerg[], and access to them is serialized
116 * lov_device::ld_mutex.
118 struct lov_device_emerg {
120 * Page list used to submit IO when memory is in pressure.
122 struct cl_page_list emrg_page_list;
124 * sub-io's shared by all threads accessing this device when memory is
125 * too low to allocate sub-io's dynamically.
127 struct cl_io emrg_subio;
129 * Environments used by sub-io's in
130 * lov_device_emerg::emrg_subio.
132 struct lu_env *emrg_env;
134 * Refchecks for lov_device_emerg::emrg_env.
143 * XXX Locking of lov-private data is missing.
145 struct cl_device ld_cl;
146 struct lov_obd *ld_lov;
147 /** size of lov_device::ld_target[] array */
149 struct lovsub_device **ld_target;
152 /** Emergency resources used in memory-cleansing paths. */
153 struct lov_device_emerg **ld_emrg;
155 * Serializes access to lov_device::ld_emrg in low-memory
158 struct mutex ld_mutex;
164 enum lov_layout_type {
165 LLT_EMPTY, /** empty file without body (mknod + truncate) */
166 LLT_RAID0, /** striped file */
167 LLT_RELEASED, /** file with no objects (data in HSM) */
171 static inline char *llt2str(enum lov_layout_type llt)
188 * lov-specific file state.
190 * lov object has particular layout type, determining how top-object is built
191 * on top of sub-objects. Layout type can change dynamically. When this
192 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
193 * all state pertaining to the old layout type is destroyed, and new state is
194 * constructed. All object methods take said semaphore in the shared mode,
195 * providing serialization against transition between layout types.
197 * To avoid multiple `if' or `switch' statements, selecting behavior for the
198 * current layout type, object methods perform double-dispatch, invoking
199 * function corresponding to the current layout type.
202 struct cl_object lo_cl;
204 * Serializes object operations with transitions between layout types.
206 * This semaphore is taken in shared mode by all object methods, and
207 * is taken in exclusive mode when object type is changed.
209 * \see lov_object::lo_type
211 struct rw_semaphore lo_type_guard;
213 * Type of an object. Protected by lov_object::lo_type_guard.
215 enum lov_layout_type lo_type;
217 * True if layout is invalid. This bit is cleared when layout lock
220 bool lo_layout_invalid;
222 * How many IOs are on going on this object. Layout can be changed
223 * only if there is no active IO.
225 atomic_t lo_active_ios;
227 * Waitq - wait for no one else is using lo_lsm
229 wait_queue_head_t lo_waitq;
231 * Layout metadata. NULL if empty layout.
233 struct lov_stripe_md *lo_lsm;
235 union lov_layout_state {
236 struct lov_layout_raid0 {
239 * When this is true, lov_object::lo_attr contains
240 * valid up to date attributes for a top-level
241 * object. This field is reset to 0 when attributes of
242 * any sub-object change.
246 * Array of sub-objects. Allocated when top-object is
247 * created (lov_init_raid0()).
249 * Top-object is a strict master of its sub-objects:
250 * it is created before them, and outlives its
251 * children (this later is necessary so that basic
252 * functions like cl_object_top() always
253 * work). Top-object keeps a reference on every
256 * When top-object is destroyed (lov_delete_raid0())
257 * it releases its reference to a sub-object and waits
258 * until the latter is finally destroyed.
260 struct lovsub_object **lo_sub;
264 spinlock_t lo_sub_lock;
266 * Cached object attribute, built from sub-object
269 struct cl_attr lo_attr;
271 struct lov_layout_state_empty {
273 struct lov_layout_state_released {
277 * Thread that acquired lov_object::lo_type_guard in an exclusive
280 struct task_struct *lo_owner;
284 * Flags that top-lock can set on each of its sub-locks.
287 /** Top-lock acquired a hold (cl_lock_hold()) on a sub-lock. */
292 * State lov_lock keeps for each sub-lock.
294 struct lov_lock_sub {
295 /** sub-lock itself */
296 struct lovsub_lock *sub_lock;
297 /** An array of per-sub-lock flags, taken from enum lov_sub_flags */
300 struct cl_lock_descr sub_descr;
301 struct cl_lock_descr sub_got;
305 * lov-specific lock state.
308 struct cl_lock_slice lls_cl;
309 /** Number of sub-locks in this lock */
312 * Number of existing sub-locks.
314 unsigned lls_nr_filled;
316 * Set when sub-lock was canceled, while top-lock was being
319 unsigned int lls_cancel_race:1;
321 * An array of sub-locks
323 * There are two issues with managing sub-locks:
325 * - sub-locks are concurrently canceled, and
327 * - sub-locks are shared with other top-locks.
329 * To manage cancellation, top-lock acquires a hold on a sublock
330 * (lov_sublock_adopt()) when the latter is inserted into
331 * lov_lock::lls_sub[]. This hold is released (lov_sublock_release())
332 * when top-lock is going into CLS_CACHED state or destroyed. Hold
333 * prevents sub-lock from cancellation.
335 * Sub-lock sharing means, among other things, that top-lock that is
336 * in the process of creation (i.e., not yet inserted into lock list)
337 * is already accessible to other threads once at least one of its
338 * sub-locks is created, see lov_lock_sub_init().
340 * Sub-lock can be in one of the following states:
342 * - doesn't exist, lov_lock::lls_sub[]::sub_lock == NULL. Such
343 * sub-lock was either never created (top-lock is in CLS_NEW
344 * state), or it was created, then canceled, then destroyed
345 * (lov_lock_unlink() cleared sub-lock pointer in the top-lock).
347 * - sub-lock exists and is on
348 * hold. (lov_lock::lls_sub[]::sub_flags & LSF_HELD). This is a
349 * normal state of a sub-lock in CLS_HELD and CLS_CACHED states
352 * - sub-lock exists, but is not held by the top-lock. This
353 * happens after top-lock released a hold on sub-locks before
354 * going into cache (lov_lock_unuse()).
356 * \todo To support wide-striping, array has to be replaced with a set
357 * of queues to avoid scanning.
359 struct lov_lock_sub *lls_sub;
361 * Original description with which lock was enqueued.
363 struct cl_lock_descr lls_orig;
367 struct cl_page_slice lps_cl;
375 struct lovsub_device {
376 struct cl_device acid_cl;
377 struct lov_device *acid_super;
379 struct cl_device *acid_next;
382 struct lovsub_object {
383 struct cl_object_header lso_header;
384 struct cl_object lso_cl;
385 struct lov_object *lso_super;
390 * A link between a top-lock and a sub-lock. Separate data-structure is
391 * necessary, because top-locks and sub-locks are in M:N relationship.
393 * \todo This can be optimized for a (by far) most frequent case of a single
394 * top-lock per sub-lock.
396 struct lov_lock_link {
397 struct lov_lock *lll_super;
398 /** An index within parent lock. */
401 * A linkage into per sub-lock list of all corresponding top-locks,
402 * hanging off lovsub_lock::lss_parents.
404 struct list_head lll_list;
408 * Lock state at lovsub layer.
411 struct cl_lock_slice lss_cl;
413 * List of top-locks that have given sub-lock as their part. Protected
414 * by cl_lock::cll_guard mutex.
416 struct list_head lss_parents;
418 * Top-lock that initiated current operation on this sub-lock. This is
419 * only set during top-to-bottom lock operations like enqueue, and is
420 * used to optimize state change notification. Protected by
421 * cl_lock::cll_guard mutex.
423 * \see lovsub_lock_state_one().
425 struct cl_lock *lss_active;
429 * Describe the environment settings for sublocks.
431 struct lov_sublock_env {
432 const struct lu_env *lse_env;
433 struct cl_io *lse_io;
434 struct lov_io_sub *lse_sub;
438 struct cl_page_slice lsb_cl;
441 struct lov_thread_info {
442 struct cl_object_conf lti_stripe_conf;
443 struct lu_fid lti_fid;
444 struct cl_lock_descr lti_ldescr;
445 struct ost_lvb lti_lvb;
446 struct cl_2queue lti_cl2q;
447 struct cl_lock_closure lti_closure;
448 wait_queue_t lti_waiter;
452 * State that lov_io maintains for every sub-io.
457 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
458 * independently, with lov acting as a scheduler to maximize overall
461 struct cl_io *sub_io;
463 * Linkage into a list (hanging off lov_io::lis_active) of all
464 * sub-io's active for the current IO iteration.
466 struct list_head sub_linkage;
468 * true, iff cl_io_init() was successfully executed against
469 * lov_io_sub::sub_io.
471 int sub_io_initialized;
473 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
474 * allocated, but borrowed from a per-device emergency pool.
478 * environment, in which sub-io executes.
480 struct lu_env *sub_env;
482 * environment's refcheck.
493 * IO state private for LOV.
497 struct cl_io_slice lis_cl;
499 * Pointer to the object slice. This is a duplicate of
500 * lov_io::lis_cl::cis_object.
502 struct lov_object *lis_object;
504 * Original end-of-io position for this IO, set by the upper layer as
505 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
506 * changes pos and count to fit IO into a single stripe and uses saved
507 * value to determine when IO iterations have to stop.
509 * This is used only for CIT_READ and CIT_WRITE io's.
511 loff_t lis_io_endpos;
514 * starting position within a file, for the current io loop iteration
515 * (stripe), used by ci_io_loop().
519 * end position with in a file, for the current stripe io. This is
520 * exclusive (i.e., next offset after last byte affected by io).
525 int lis_stripe_count;
526 int lis_active_subios;
529 * the index of ls_single_subio in ls_subios array
531 int lis_single_subio_index;
532 struct cl_io lis_single_subio;
535 * size of ls_subios array, actually the highest stripe #
538 struct lov_io_sub *lis_subs;
540 * List of active sub-io's.
542 struct list_head lis_active;
547 struct lov_sublock_env ls_subenv;
551 * State of transfer for lov.
554 struct cl_req_slice lr_cl;
558 * State of transfer for lovsub.
561 struct cl_req_slice lsrq_cl;
564 extern struct lu_device_type lov_device_type;
565 extern struct lu_device_type lovsub_device_type;
567 extern struct lu_context_key lov_key;
568 extern struct lu_context_key lov_session_key;
570 extern struct kmem_cache *lov_lock_kmem;
571 extern struct kmem_cache *lov_object_kmem;
572 extern struct kmem_cache *lov_thread_kmem;
573 extern struct kmem_cache *lov_session_kmem;
574 extern struct kmem_cache *lov_req_kmem;
576 extern struct kmem_cache *lovsub_lock_kmem;
577 extern struct kmem_cache *lovsub_object_kmem;
578 extern struct kmem_cache *lovsub_req_kmem;
580 extern struct kmem_cache *lov_lock_link_kmem;
582 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
583 const struct lu_object_conf *conf);
584 int lovsub_object_init(const struct lu_env *env, struct lu_object *obj,
585 const struct lu_object_conf *conf);
586 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
587 struct cl_lock *lock, const struct cl_io *io);
588 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
590 int lovsub_lock_init(const struct lu_env *env, struct cl_object *obj,
591 struct cl_lock *lock, const struct cl_io *io);
593 int lov_lock_init_raid0(const struct lu_env *env, struct cl_object *obj,
594 struct cl_lock *lock, const struct cl_io *io);
595 int lov_lock_init_empty(const struct lu_env *env, struct cl_object *obj,
596 struct cl_lock *lock, const struct cl_io *io);
597 int lov_io_init_raid0(const struct lu_env *env, struct cl_object *obj,
599 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
601 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
603 void lov_lock_unlink(const struct lu_env *env, struct lov_lock_link *link,
604 struct lovsub_lock *sub);
606 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
608 void lov_sub_put(struct lov_io_sub *sub);
609 int lov_sublock_modify(const struct lu_env *env, struct lov_lock *lov,
610 struct lovsub_lock *sublock,
611 const struct cl_lock_descr *d, int idx);
613 int lov_page_init(const struct lu_env *env, struct cl_object *ob,
614 struct cl_page *page, struct page *vmpage);
615 int lovsub_page_init(const struct lu_env *env, struct cl_object *ob,
616 struct cl_page *page, struct page *vmpage);
618 int lov_page_init_empty(const struct lu_env *env,
619 struct cl_object *obj,
620 struct cl_page *page, struct page *vmpage);
621 int lov_page_init_raid0(const struct lu_env *env,
622 struct cl_object *obj,
623 struct cl_page *page, struct page *vmpage);
624 struct lu_object *lov_object_alloc(const struct lu_env *env,
625 const struct lu_object_header *hdr,
626 struct lu_device *dev);
627 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
628 const struct lu_object_header *hdr,
629 struct lu_device *dev);
631 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
632 struct lov_lock *lck,
633 struct lovsub_lock *sub);
634 struct lov_io_sub *lov_page_subio(const struct lu_env *env,
636 const struct cl_page_slice *slice);
638 #define lov_foreach_target(lov, var) \
639 for (var = 0; var < lov_targets_nr(lov); ++var)
641 /*****************************************************************************
649 static inline struct lov_session *lov_env_session(const struct lu_env *env)
651 struct lov_session *ses;
653 ses = lu_context_key_get(env->le_ses, &lov_session_key);
654 LASSERT(ses != NULL);
658 static inline struct lov_io *lov_env_io(const struct lu_env *env)
660 return &lov_env_session(env)->ls_io;
663 static inline int lov_is_object(const struct lu_object *obj)
665 return obj->lo_dev->ld_type == &lov_device_type;
668 static inline int lovsub_is_object(const struct lu_object *obj)
670 return obj->lo_dev->ld_type == &lovsub_device_type;
673 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
675 return &lov->ld_cl.cd_lu_dev;
678 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
680 LINVRNT(d->ld_type == &lov_device_type);
681 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
684 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
686 return &lovsub->acid_cl;
689 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
691 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
694 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
696 LINVRNT(d->ld_type == &lovsub_device_type);
697 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
700 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
702 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
703 return container_of0(d, struct lovsub_device, acid_cl);
706 static inline struct lu_object *lov2lu(struct lov_object *lov)
708 return &lov->lo_cl.co_lu;
711 static inline struct cl_object *lov2cl(struct lov_object *lov)
716 static inline struct lov_object *lu2lov(const struct lu_object *obj)
718 LINVRNT(lov_is_object(obj));
719 return container_of0(obj, struct lov_object, lo_cl.co_lu);
722 static inline struct lov_object *cl2lov(const struct cl_object *obj)
724 LINVRNT(lov_is_object(&obj->co_lu));
725 return container_of0(obj, struct lov_object, lo_cl);
728 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
730 return &los->lso_cl.co_lu;
733 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
738 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
740 LINVRNT(lovsub_is_object(&obj->co_lu));
741 return container_of0(obj, struct lovsub_object, lso_cl);
744 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
746 LINVRNT(lovsub_is_object(obj));
747 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
750 static inline struct lovsub_lock *
751 cl2lovsub_lock(const struct cl_lock_slice *slice)
753 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
754 return container_of(slice, struct lovsub_lock, lss_cl);
757 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
759 const struct cl_lock_slice *slice;
761 slice = cl_lock_at(lock, &lovsub_device_type);
762 LASSERT(slice != NULL);
763 return cl2lovsub_lock(slice);
766 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
768 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
769 return container_of(slice, struct lov_lock, lls_cl);
772 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
774 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
775 return container_of0(slice, struct lov_page, lps_cl);
778 static inline struct lov_req *cl2lov_req(const struct cl_req_slice *slice)
780 return container_of0(slice, struct lov_req, lr_cl);
783 static inline struct lovsub_page *
784 cl2lovsub_page(const struct cl_page_slice *slice)
786 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
787 return container_of0(slice, struct lovsub_page, lsb_cl);
790 static inline struct lovsub_req *cl2lovsub_req(const struct cl_req_slice *slice)
792 return container_of0(slice, struct lovsub_req, lsrq_cl);
795 static inline struct cl_page *lov_sub_page(const struct cl_page_slice *slice)
797 return slice->cpl_page->cp_child;
800 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
801 const struct cl_io_slice *ios)
805 lio = container_of(ios, struct lov_io, lis_cl);
806 LASSERT(lio == lov_env_io(env));
810 static inline int lov_targets_nr(const struct lov_device *lov)
812 return lov->ld_lov->desc.ld_tgt_count;
815 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
817 struct lov_thread_info *info;
819 info = lu_context_key_get(&env->le_ctx, &lov_key);
820 LASSERT(info != NULL);
824 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
826 LASSERT(lov->lo_type == LLT_RAID0);
827 LASSERT(lov->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC ||
828 lov->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC_V3);
829 return &lov->u.raid0;