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, 2013, 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
50 # include <libcfs/libcfs.h>
52 # include <liblustre.h>
56 #include <cl_object.h>
57 #include "lov_internal.h"
60 * Logical object volume layer. This layer implements data striping (raid0).
62 * At the lov layer top-entity (object, page, lock, io) is connected to one or
63 * more sub-entities: top-object, representing a file is connected to a set of
64 * sub-objects, each representing a stripe, file-level top-lock is connected
65 * to a set of per-stripe sub-locks, top-page is connected to a (single)
66 * sub-page, and a top-level IO is connected to a set of (potentially
67 * concurrent) sub-IO's.
69 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
70 * respectively, while a single sub-lock can be part of multiple top-locks.
72 * Reference counting models are different for different types of entities:
74 * - top-object keeps a reference to its sub-objects, and destroys them
75 * when it is destroyed.
77 * - top-page keeps a reference to its sub-page, and destroys it when it
80 * - sub-lock keep a reference to its top-locks. Top-lock keeps a
81 * reference (and a hold, see cl_lock_hold()) on its sub-locks when it
82 * actively using them (that is, in cl_lock_state::CLS_QUEUING,
83 * cl_lock_state::CLS_ENQUEUED, cl_lock_state::CLS_HELD states). When
84 * moving into cl_lock_state::CLS_CACHED state, top-lock releases a
85 * hold. From this moment top-lock has only a 'weak' reference to its
86 * sub-locks. This reference is protected by top-lock
87 * cl_lock::cll_guard, and will be automatically cleared by the sub-lock
88 * when the latter is destroyed. When a sub-lock is canceled, a
89 * reference to it is removed from the top-lock array, and top-lock is
90 * moved into CLS_NEW state. It is guaranteed that all sub-locks exist
91 * while their top-lock is in CLS_HELD or CLS_CACHED states.
93 * - IO's are not reference counted.
95 * To implement a connection between top and sub entities, lov layer is split
96 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
97 * implementing full set of cl-interfaces. For example, top-object has vvp and
98 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
99 * used to track child-parent relationship.
104 struct lovsub_device;
105 struct lovsub_object;
108 enum lov_device_flags {
109 LOV_DEV_INITIALIZED = 1 << 0
117 * Resources that are used in memory-cleaning path, and whose allocation
118 * cannot fail even when memory is tight. They are preallocated in sufficient
119 * quantities in lov_device::ld_emerg[], and access to them is serialized
120 * lov_device::ld_mutex.
122 struct lov_device_emerg {
124 * Page list used to submit IO when memory is in pressure.
126 struct cl_page_list emrg_page_list;
128 * sub-io's shared by all threads accessing this device when memory is
129 * too low to allocate sub-io's dynamically.
131 struct cl_io emrg_subio;
133 * Environments used by sub-io's in
134 * lov_device_emerg::emrg_subio.
136 struct lu_env *emrg_env;
138 * Refchecks for lov_device_emerg::emrg_env.
147 * XXX Locking of lov-private data is missing.
149 struct cl_device ld_cl;
150 struct lov_obd *ld_lov;
151 /** size of lov_device::ld_target[] array */
153 struct lovsub_device **ld_target;
156 /** Emergency resources used in memory-cleansing paths. */
157 struct lov_device_emerg **ld_emrg;
159 * Serializes access to lov_device::ld_emrg in low-memory
162 struct mutex ld_mutex;
168 enum lov_layout_type {
169 LLT_EMPTY, /** empty file without body (mknod + truncate) */
170 LLT_RAID0, /** striped file */
171 LLT_RELEASED, /** file with no objects (data in HSM) */
175 static inline char *llt2str(enum lov_layout_type llt)
192 * lov-specific file state.
194 * lov object has particular layout type, determining how top-object is built
195 * on top of sub-objects. Layout type can change dynamically. When this
196 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
197 * all state pertaining to the old layout type is destroyed, and new state is
198 * constructed. All object methods take said semaphore in the shared mode,
199 * providing serialization against transition between layout types.
201 * To avoid multiple `if' or `switch' statements, selecting behavior for the
202 * current layout type, object methods perform double-dispatch, invoking
203 * function corresponding to the current layout type.
206 struct cl_object lo_cl;
208 * Serializes object operations with transitions between layout types.
210 * This semaphore is taken in shared mode by all object methods, and
211 * is taken in exclusive mode when object type is changed.
213 * \see lov_object::lo_type
215 struct rw_semaphore lo_type_guard;
217 * Type of an object. Protected by lov_object::lo_type_guard.
219 enum lov_layout_type lo_type;
221 * True if layout is invalid. This bit is cleared when layout lock
224 bool lo_layout_invalid;
226 * How many IOs are on going on this object. Layout can be changed
227 * only if there is no active IO.
229 cfs_atomic_t lo_active_ios;
231 * Waitq - wait for no one else is using lo_lsm
233 cfs_waitq_t lo_waitq;
235 * Layout metadata. NULL if empty layout.
237 struct lov_stripe_md *lo_lsm;
239 union lov_layout_state {
240 struct lov_layout_raid0 {
243 * When this is true, lov_object::lo_attr contains
244 * valid up to date attributes for a top-level
245 * object. This field is reset to 0 when attributes of
246 * any sub-object change.
250 * Array of sub-objects. Allocated when top-object is
251 * created (lov_init_raid0()).
253 * Top-object is a strict master of its sub-objects:
254 * it is created before them, and outlives its
255 * children (this later is necessary so that basic
256 * functions like cl_object_top() always
257 * work). Top-object keeps a reference on every
260 * When top-object is destroyed (lov_delete_raid0())
261 * it releases its reference to a sub-object and waits
262 * until the latter is finally destroyed.
264 struct lovsub_object **lo_sub;
268 spinlock_t lo_sub_lock;
270 * Cached object attribute, built from sub-object
273 struct cl_attr lo_attr;
275 struct lov_layout_state_empty {
277 struct lov_layout_state_released {
281 * Thread that acquired lov_object::lo_type_guard in an exclusive
284 cfs_task_t *lo_owner;
288 * Flags that top-lock can set on each of its sub-locks.
291 /** Top-lock acquired a hold (cl_lock_hold()) on a sub-lock. */
296 * State lov_lock keeps for each sub-lock.
298 struct lov_lock_sub {
299 /** sub-lock itself */
300 struct lovsub_lock *sub_lock;
301 /** An array of per-sub-lock flags, taken from enum lov_sub_flags */
304 struct cl_lock_descr sub_descr;
305 struct cl_lock_descr sub_got;
309 * lov-specific lock state.
312 struct cl_lock_slice lls_cl;
313 /** Number of sub-locks in this lock */
316 * Number of existing sub-locks.
318 unsigned lls_nr_filled;
320 * Set when sub-lock was canceled, while top-lock was being
323 unsigned int lls_cancel_race:1;
325 * An array of sub-locks
327 * There are two issues with managing sub-locks:
329 * - sub-locks are concurrently canceled, and
331 * - sub-locks are shared with other top-locks.
333 * To manage cancellation, top-lock acquires a hold on a sublock
334 * (lov_sublock_adopt()) when the latter is inserted into
335 * lov_lock::lls_sub[]. This hold is released (lov_sublock_release())
336 * when top-lock is going into CLS_CACHED state or destroyed. Hold
337 * prevents sub-lock from cancellation.
339 * Sub-lock sharing means, among other things, that top-lock that is
340 * in the process of creation (i.e., not yet inserted into lock list)
341 * is already accessible to other threads once at least one of its
342 * sub-locks is created, see lov_lock_sub_init().
344 * Sub-lock can be in one of the following states:
346 * - doesn't exist, lov_lock::lls_sub[]::sub_lock == NULL. Such
347 * sub-lock was either never created (top-lock is in CLS_NEW
348 * state), or it was created, then canceled, then destroyed
349 * (lov_lock_unlink() cleared sub-lock pointer in the top-lock).
351 * - sub-lock exists and is on
352 * hold. (lov_lock::lls_sub[]::sub_flags & LSF_HELD). This is a
353 * normal state of a sub-lock in CLS_HELD and CLS_CACHED states
356 * - sub-lock exists, but is not held by the top-lock. This
357 * happens after top-lock released a hold on sub-locks before
358 * going into cache (lov_lock_unuse()).
360 * \todo To support wide-striping, array has to be replaced with a set
361 * of queues to avoid scanning.
363 struct lov_lock_sub *lls_sub;
365 * Original description with which lock was enqueued.
367 struct cl_lock_descr lls_orig;
371 struct cl_page_slice lps_cl;
379 struct lovsub_device {
380 struct cl_device acid_cl;
381 struct lov_device *acid_super;
383 struct cl_device *acid_next;
386 struct lovsub_object {
387 struct cl_object_header lso_header;
388 struct cl_object lso_cl;
389 struct lov_object *lso_super;
394 * A link between a top-lock and a sub-lock. Separate data-structure is
395 * necessary, because top-locks and sub-locks are in M:N relationship.
397 * \todo This can be optimized for a (by far) most frequent case of a single
398 * top-lock per sub-lock.
400 struct lov_lock_link {
401 struct lov_lock *lll_super;
402 /** An index within parent lock. */
405 * A linkage into per sub-lock list of all corresponding top-locks,
406 * hanging off lovsub_lock::lss_parents.
412 * Lock state at lovsub layer.
415 struct cl_lock_slice lss_cl;
417 * List of top-locks that have given sub-lock as their part. Protected
418 * by cl_lock::cll_guard mutex.
420 cfs_list_t lss_parents;
422 * Top-lock that initiated current operation on this sub-lock. This is
423 * only set during top-to-bottom lock operations like enqueue, and is
424 * used to optimize state change notification. Protected by
425 * cl_lock::cll_guard mutex.
427 * \see lovsub_lock_state_one().
429 struct cl_lock *lss_active;
433 * Describe the environment settings for sublocks.
435 struct lov_sublock_env {
436 const struct lu_env *lse_env;
437 struct cl_io *lse_io;
438 struct lov_io_sub *lse_sub;
442 struct cl_page_slice lsb_cl;
446 struct lov_thread_info {
447 struct cl_object_conf lti_stripe_conf;
448 struct lu_fid lti_fid;
449 struct cl_lock_descr lti_ldescr;
450 struct ost_lvb lti_lvb;
451 struct cl_2queue lti_cl2q;
452 struct cl_lock_closure lti_closure;
453 cfs_waitlink_t lti_waiter;
457 * State that lov_io maintains for every sub-io.
462 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
463 * independently, with lov acting as a scheduler to maximize overall
466 struct cl_io *sub_io;
468 * Linkage into a list (hanging off lov_io::lis_active) of all
469 * sub-io's active for the current IO iteration.
471 cfs_list_t sub_linkage;
473 * true, iff cl_io_init() was successfully executed against
474 * lov_io_sub::sub_io.
476 int sub_io_initialized;
478 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
479 * allocated, but borrowed from a per-device emergency pool.
483 * environment, in which sub-io executes.
485 struct lu_env *sub_env;
487 * environment's refcheck.
498 * IO state private for LOV.
502 struct cl_io_slice lis_cl;
504 * Pointer to the object slice. This is a duplicate of
505 * lov_io::lis_cl::cis_object.
507 struct lov_object *lis_object;
509 * Original end-of-io position for this IO, set by the upper layer as
510 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
511 * changes pos and count to fit IO into a single stripe and uses saved
512 * value to determine when IO iterations have to stop.
514 * This is used only for CIT_READ and CIT_WRITE io's.
516 loff_t lis_io_endpos;
519 * starting position within a file, for the current io loop iteration
520 * (stripe), used by ci_io_loop().
524 * end position with in a file, for the current stripe io. This is
525 * exclusive (i.e., next offset after last byte affected by io).
530 int lis_stripe_count;
531 int lis_active_subios;
534 * the index of ls_single_subio in ls_subios array
536 int lis_single_subio_index;
537 struct cl_io lis_single_subio;
540 * size of ls_subios array, actually the highest stripe #
543 struct lov_io_sub *lis_subs;
545 * List of active sub-io's.
547 cfs_list_t lis_active;
552 struct lov_sublock_env ls_subenv;
556 * State of transfer for lov.
559 struct cl_req_slice lr_cl;
563 * State of transfer for lovsub.
566 struct cl_req_slice lsrq_cl;
569 extern struct lu_device_type lov_device_type;
570 extern struct lu_device_type lovsub_device_type;
572 extern struct lu_context_key lov_key;
573 extern struct lu_context_key lov_session_key;
575 extern struct kmem_cache *lov_lock_kmem;
576 extern struct kmem_cache *lov_object_kmem;
577 extern struct kmem_cache *lov_thread_kmem;
578 extern struct kmem_cache *lov_session_kmem;
579 extern struct kmem_cache *lov_req_kmem;
581 extern struct kmem_cache *lovsub_lock_kmem;
582 extern struct kmem_cache *lovsub_object_kmem;
583 extern struct kmem_cache *lovsub_req_kmem;
585 extern struct kmem_cache *lov_lock_link_kmem;
587 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
588 const struct lu_object_conf *conf);
589 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
590 const struct lu_object_conf *conf);
591 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
592 struct cl_lock *lock, const struct cl_io *io);
593 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
595 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
596 struct cl_lock *lock, const struct cl_io *io);
598 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
599 struct cl_lock *lock, const struct cl_io *io);
600 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
601 struct cl_lock *lock, const struct cl_io *io);
602 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
604 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
606 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
608 void lov_lock_unlink (const struct lu_env *env, struct lov_lock_link *link,
609 struct lovsub_lock *sub);
611 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
613 void lov_sub_put (struct lov_io_sub *sub);
614 int lov_sublock_modify (const struct lu_env *env, struct lov_lock *lov,
615 struct lovsub_lock *sublock,
616 const struct cl_lock_descr *d, int idx);
619 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
620 struct cl_page *page, struct page *vmpage);
621 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
622 struct cl_page *page, struct page *vmpage);
624 int lov_page_init_empty (const struct lu_env *env,
625 struct cl_object *obj,
626 struct cl_page *page, struct page *vmpage);
627 int lov_page_init_raid0 (const struct lu_env *env,
628 struct cl_object *obj,
629 struct cl_page *page, struct page *vmpage);
630 struct lu_object *lov_object_alloc (const struct lu_env *env,
631 const struct lu_object_header *hdr,
632 struct lu_device *dev);
633 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
634 const struct lu_object_header *hdr,
635 struct lu_device *dev);
637 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
638 struct lov_lock *lck,
639 struct lovsub_lock *sub);
640 struct lov_io_sub *lov_page_subio (const struct lu_env *env,
642 const struct cl_page_slice *slice);
644 void lov_lsm_decref(struct lov_object *lov, struct lov_stripe_md *lsm);
645 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
647 #define lov_foreach_target(lov, var) \
648 for (var = 0; var < lov_targets_nr(lov); ++var)
650 /*****************************************************************************
658 static inline struct lov_session *lov_env_session(const struct lu_env *env)
660 struct lov_session *ses;
662 ses = lu_context_key_get(env->le_ses, &lov_session_key);
663 LASSERT(ses != NULL);
667 static inline struct lov_io *lov_env_io(const struct lu_env *env)
669 return &lov_env_session(env)->ls_io;
672 static inline int lov_is_object(const struct lu_object *obj)
674 return obj->lo_dev->ld_type == &lov_device_type;
677 static inline int lovsub_is_object(const struct lu_object *obj)
679 return obj->lo_dev->ld_type == &lovsub_device_type;
682 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
684 return &lov->ld_cl.cd_lu_dev;
687 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
689 LINVRNT(d->ld_type == &lov_device_type);
690 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
693 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
695 return &lovsub->acid_cl;
698 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
700 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
703 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
705 LINVRNT(d->ld_type == &lovsub_device_type);
706 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
709 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
711 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
712 return container_of0(d, struct lovsub_device, acid_cl);
715 static inline struct lu_object *lov2lu(struct lov_object *lov)
717 return &lov->lo_cl.co_lu;
720 static inline struct cl_object *lov2cl(struct lov_object *lov)
725 static inline struct lov_object *lu2lov(const struct lu_object *obj)
727 LINVRNT(lov_is_object(obj));
728 return container_of0(obj, struct lov_object, lo_cl.co_lu);
731 static inline struct lov_object *cl2lov(const struct cl_object *obj)
733 LINVRNT(lov_is_object(&obj->co_lu));
734 return container_of0(obj, struct lov_object, lo_cl);
737 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
739 return &los->lso_cl.co_lu;
742 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
747 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
749 LINVRNT(lovsub_is_object(&obj->co_lu));
750 return container_of0(obj, struct lovsub_object, lso_cl);
753 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
755 LINVRNT(lovsub_is_object(obj));
756 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
759 static inline struct lovsub_lock *
760 cl2lovsub_lock(const struct cl_lock_slice *slice)
762 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
763 return container_of(slice, struct lovsub_lock, lss_cl);
766 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
768 const struct cl_lock_slice *slice;
770 slice = cl_lock_at(lock, &lovsub_device_type);
771 LASSERT(slice != NULL);
772 return cl2lovsub_lock(slice);
775 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
777 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
778 return container_of(slice, struct lov_lock, lls_cl);
781 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
783 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
784 return container_of0(slice, struct lov_page, lps_cl);
787 static inline struct lov_req *cl2lov_req(const struct cl_req_slice *slice)
789 return container_of0(slice, struct lov_req, lr_cl);
792 static inline struct lovsub_page *
793 cl2lovsub_page(const struct cl_page_slice *slice)
795 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
796 return container_of0(slice, struct lovsub_page, lsb_cl);
799 static inline struct lovsub_req *cl2lovsub_req(const struct cl_req_slice *slice)
801 return container_of0(slice, struct lovsub_req, lsrq_cl);
804 static inline struct cl_page *lov_sub_page(const struct cl_page_slice *slice)
806 return slice->cpl_page->cp_child;
809 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
810 const struct cl_io_slice *ios)
814 lio = container_of(ios, struct lov_io, lis_cl);
815 LASSERT(lio == lov_env_io(env));
819 static inline int lov_targets_nr(const struct lov_device *lov)
821 return lov->ld_lov->desc.ld_tgt_count;
824 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
826 struct lov_thread_info *info;
828 info = lu_context_key_get(&env->le_ctx, &lov_key);
829 LASSERT(info != NULL);
833 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
835 LASSERT(lov->lo_type == LLT_RAID0);
836 LASSERT(lov->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC ||
837 lov->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC_V3);
838 return &lov->u.raid0;