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 (c) 2008, 2010, Oracle and/or its affiliates. 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"
59 * Logical object volume layer. This layer implements data striping (raid0).
61 * At the lov layer top-entity (object, page, lock, io) is connected to one or
62 * more sub-entities: top-object, representing a file is connected to a set of
63 * sub-objects, each representing a stripe, file-level top-lock is connected
64 * to a set of per-stripe sub-locks, top-page is connected to a (single)
65 * sub-page, and a top-level IO is connected to a set of (potentially
66 * concurrent) sub-IO's.
68 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
69 * respectively, while a single sub-lock can be part of multiple top-locks.
71 * Reference counting models are different for different types of entities:
73 * - top-object keeps a reference to its sub-objects, and destroys them
74 * when it is destroyed.
76 * - top-page keeps a reference to its sub-page, and destroys it when it
79 * - sub-lock keep a reference to its top-locks. Top-lock keeps a
80 * reference (and a hold, see cl_lock_hold()) on its sub-locks when it
81 * actively using them (that is, in cl_lock_state::CLS_QUEUING,
82 * cl_lock_state::CLS_ENQUEUED, cl_lock_state::CLS_HELD states). When
83 * moving into cl_lock_state::CLS_CACHED state, top-lock releases a
84 * hold. From this moment top-lock has only a 'weak' reference to its
85 * sub-locks. This reference is protected by top-lock
86 * cl_lock::cll_guard, and will be automatically cleared by the sub-lock
87 * when the latter is destroyed. When a sub-lock is canceled, a
88 * reference to it is removed from the top-lock array, and top-lock is
89 * moved into CLS_NEW state. It is guaranteed that all sub-locks exist
90 * while their top-lock is in CLS_HELD or CLS_CACHED states.
92 * - IO's are not reference counted.
94 * To implement a connection between top and sub entities, lov layer is split
95 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
96 * implementing full set of cl-interfaces. For example, top-object has vvp and
97 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
98 * used to track child-parent relationship.
103 struct lovsub_device;
104 struct lovsub_object;
107 enum lov_device_flags {
108 LOV_DEV_INITIALIZED = 1 << 0
116 * Resources that are used in memory-cleaning path, and whose allocation
117 * cannot fail even when memory is tight. They are preallocated in sufficient
118 * quantities in lov_device::ld_emerg[], and access to them is serialized
119 * lov_device::ld_mutex.
121 struct lov_device_emerg {
123 * Page list used to submit IO when memory is in pressure.
125 struct cl_page_list emrg_page_list;
127 * sub-io's shared by all threads accessing this device when memory is
128 * too low to allocate sub-io's dynamically.
130 struct cl_io emrg_subio;
132 * Environments used by sub-io's in
133 * lov_device_emerg::emrg_subio.
135 struct lu_env *emrg_env;
137 * Refchecks for lov_device_emerg::emrg_env.
146 * XXX Locking of lov-private data is missing.
148 struct cl_device ld_cl;
149 struct lov_obd *ld_lov;
150 /** size of lov_device::ld_target[] array */
152 struct lovsub_device **ld_target;
155 /** Emergency resources used in memory-cleansing paths. */
156 struct lov_device_emerg **ld_emrg;
158 * Serializes access to lov_device::ld_emrg in low-memory
161 cfs_mutex_t ld_mutex;
167 enum lov_layout_type {
168 /** empty file without body */
176 * lov-specific file state.
178 * lov object has particular layout type, determining how top-object is built
179 * on top of sub-objects. Layout type can change dynamically. When this
180 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
181 * all state pertaining to the old layout type is destroyed, and new state is
182 * constructed. All object methods take said semaphore in the shared mode,
183 * providing serialization against transition between layout types.
185 * To avoid multiple `if' or `switch' statements, selecting behavior for the
186 * current layout type, object methods perform double-dispatch, invoking
187 * function corresponding to the current layout type.
190 struct cl_object lo_cl;
192 * Serializes object operations with transitions between layout types.
194 * This semaphore is taken in shared mode by all object methods, and
195 * is taken in exclusive mode when object type is changed.
197 * \see lov_object::lo_type
199 cfs_rw_semaphore_t lo_type_guard;
201 * Type of an object. Protected by lov_object::lo_type_guard.
203 enum lov_layout_type lo_type;
205 union lov_layout_state {
206 struct lov_layout_raid0 {
208 struct lov_stripe_md *lo_lsm;
210 * Array of sub-objects. Allocated when top-object is
211 * created (lov_init_raid0()).
213 * Top-object is a strict master of its sub-objects:
214 * it is created before them, and outlives its
215 * children (this later is necessary so that basic
216 * functions like cl_object_top() always
217 * work). Top-object keeps a reference on every
220 * When top-object is destroyed (lov_delete_raid0())
221 * it releases its reference to a sub-object and waits
222 * until the latter is finally destroyed.
224 struct lovsub_object **lo_sub;
228 cfs_spinlock_t lo_sub_lock;
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 {
246 * Thread that acquired lov_object::lo_type_guard in an exclusive
249 cfs_task_t *lo_owner;
253 * Flags that top-lock can set on each of its sub-locks.
256 /** Top-lock acquired a hold (cl_lock_hold()) on a sub-lock. */
261 * State lov_lock keeps for each sub-lock.
263 struct lov_lock_sub {
264 /** sub-lock itself */
265 struct lovsub_lock *sub_lock;
266 /** An array of per-sub-lock flags, taken from enum lov_sub_flags */
269 struct cl_lock_descr sub_descr;
270 struct cl_lock_descr sub_got;
274 * lov-specific lock state.
277 struct cl_lock_slice lls_cl;
278 /** Number of sub-locks in this lock */
281 * Number of existing sub-locks.
283 unsigned lls_nr_filled;
285 * Set when sub-lock was canceled, while top-lock was being
288 int lls_cancel_race:1;
290 * An array of sub-locks
292 * There are two issues with managing sub-locks:
294 * - sub-locks are concurrently canceled, and
296 * - sub-locks are shared with other top-locks.
298 * To manage cancellation, top-lock acquires a hold on a sublock
299 * (lov_sublock_adopt()) when the latter is inserted into
300 * lov_lock::lls_sub[]. This hold is released (lov_sublock_release())
301 * when top-lock is going into CLS_CACHED state or destroyed. Hold
302 * prevents sub-lock from cancellation.
304 * Sub-lock sharing means, among other things, that top-lock that is
305 * in the process of creation (i.e., not yet inserted into lock list)
306 * is already accessible to other threads once at least one of its
307 * sub-locks is created, see lov_lock_sub_init().
309 * Sub-lock can be in one of the following states:
311 * - doesn't exist, lov_lock::lls_sub[]::sub_lock == NULL. Such
312 * sub-lock was either never created (top-lock is in CLS_NEW
313 * state), or it was created, then canceled, then destroyed
314 * (lov_lock_unlink() cleared sub-lock pointer in the top-lock).
316 * - sub-lock exists and is on
317 * hold. (lov_lock::lls_sub[]::sub_flags & LSF_HELD). This is a
318 * normal state of a sub-lock in CLS_HELD and CLS_CACHED states
321 * - sub-lock exists, but is not held by the top-lock. This
322 * happens after top-lock released a hold on sub-locks before
323 * going into cache (lov_lock_unuse()).
325 * \todo To support wide-striping, array has to be replaced with a set
326 * of queues to avoid scanning.
328 struct lov_lock_sub *lls_sub;
330 * Original description with which lock was enqueued.
332 struct cl_lock_descr lls_orig;
336 struct cl_page_slice lps_cl;
344 struct lovsub_device {
345 struct cl_device acid_cl;
346 struct lov_device *acid_super;
348 struct cl_device *acid_next;
351 struct lovsub_object {
352 struct cl_object_header lso_header;
353 struct cl_object lso_cl;
354 struct lov_object *lso_super;
359 * A link between a top-lock and a sub-lock. Separate data-structure is
360 * necessary, because top-locks and sub-locks are in M:N relationship.
362 * \todo This can be optimized for a (by far) most frequent case of a single
363 * top-lock per sub-lock.
365 struct lov_lock_link {
366 struct lov_lock *lll_super;
367 /** An index within parent lock. */
370 * A linkage into per sub-lock list of all corresponding top-locks,
371 * hanging off lovsub_lock::lss_parents.
377 * Lock state at lovsub layer.
380 struct cl_lock_slice lss_cl;
382 * List of top-locks that have given sub-lock as their part. Protected
383 * by cl_lock::cll_guard mutex.
385 cfs_list_t lss_parents;
387 * Top-lock that initiated current operation on this sub-lock. This is
388 * only set during top-to-bottom lock operations like enqueue, and is
389 * used to optimize state change notification. Protected by
390 * cl_lock::cll_guard mutex.
392 * \see lovsub_lock_state_one().
394 struct cl_lock *lss_active;
398 * Describe the environment settings for sublocks.
400 struct lov_sublock_env {
401 const struct lu_env *lse_env;
402 struct cl_io *lse_io;
403 struct lov_io_sub *lse_sub;
407 struct cl_page_slice lsb_cl;
411 struct lov_thread_info {
412 struct cl_object_conf lti_stripe_conf;
413 struct lu_fid lti_fid;
414 struct cl_lock_descr lti_ldescr;
415 struct ost_lvb lti_lvb;
416 struct cl_2queue lti_cl2q;
417 union lov_layout_state lti_state;
418 struct cl_lock_closure lti_closure;
419 cfs_waitlink_t lti_waiter;
423 * State that lov_io maintains for every sub-io.
428 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
429 * independently, with lov acting as a scheduler to maximize overall
432 struct cl_io *sub_io;
434 * Linkage into a list (hanging off lov_io::lis_active) of all
435 * sub-io's active for the current IO iteration.
437 cfs_list_t sub_linkage;
439 * true, iff cl_io_init() was successfully executed against
440 * lov_io_sub::sub_io.
442 int sub_io_initialized;
444 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
445 * allocated, but borrowed from a per-device emergency pool.
449 * environment, in which sub-io executes.
451 struct lu_env *sub_env;
453 * environment's refcheck.
464 * IO state private for LOV.
468 struct cl_io_slice lis_cl;
470 * Pointer to the object slice. This is a duplicate of
471 * lov_io::lis_cl::cis_object.
473 struct lov_object *lis_object;
475 * Original end-of-io position for this IO, set by the upper layer as
476 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
477 * changes pos and count to fit IO into a single stripe and uses saved
478 * value to determine when IO iterations have to stop.
480 * This is used only for CIT_READ and CIT_WRITE io's.
482 loff_t lis_io_endpos;
485 * starting position within a file, for the current io loop iteration
486 * (stripe), used by ci_io_loop().
490 * end position with in a file, for the current stripe io. This is
491 * exclusive (i.e., next offset after last byte affected by io).
496 int lis_stripe_count;
497 int lis_active_subios;
500 * the index of ls_single_subio in ls_subios array
502 int lis_single_subio_index;
503 struct cl_io lis_single_subio;
506 * size of ls_subios array, actually the highest stripe #
509 struct lov_io_sub *lis_subs;
511 * List of active sub-io's.
513 cfs_list_t lis_active;
518 struct lov_sublock_env ls_subenv;
522 * State of transfer for lov.
525 struct cl_req_slice lr_cl;
529 * State of transfer for lovsub.
532 struct cl_req_slice lsrq_cl;
535 extern struct lu_device_type lov_device_type;
536 extern struct lu_device_type lovsub_device_type;
538 extern struct lu_context_key lov_key;
539 extern struct lu_context_key lov_session_key;
541 extern cfs_mem_cache_t *lov_page_kmem;
542 extern cfs_mem_cache_t *lov_lock_kmem;
543 extern cfs_mem_cache_t *lov_object_kmem;
544 extern cfs_mem_cache_t *lov_thread_kmem;
545 extern cfs_mem_cache_t *lov_session_kmem;
546 extern cfs_mem_cache_t *lov_req_kmem;
548 extern cfs_mem_cache_t *lovsub_page_kmem;
549 extern cfs_mem_cache_t *lovsub_lock_kmem;
550 extern cfs_mem_cache_t *lovsub_object_kmem;
551 extern cfs_mem_cache_t *lovsub_req_kmem;
553 extern cfs_mem_cache_t *lov_lock_link_kmem;
555 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
556 const struct lu_object_conf *conf);
557 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
558 const struct lu_object_conf *conf);
559 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
560 struct cl_lock *lock, const struct cl_io *io);
561 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
563 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
564 struct cl_lock *lock, const struct cl_io *io);
566 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
567 struct cl_lock *lock, const struct cl_io *io);
568 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
570 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
572 void lov_lock_unlink (const struct lu_env *env, struct lov_lock_link *link,
573 struct lovsub_lock *sub);
575 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
577 void lov_sub_put (struct lov_io_sub *sub);
578 int lov_sublock_modify (const struct lu_env *env, struct lov_lock *lov,
579 struct lovsub_lock *sublock,
580 const struct cl_lock_descr *d, int idx);
583 struct cl_page *lov_page_init (const struct lu_env *env, struct cl_object *ob,
584 struct cl_page *page, cfs_page_t *vmpage);
585 struct cl_page *lovsub_page_init(const struct lu_env *env, struct cl_object *ob,
586 struct cl_page *page, cfs_page_t *vmpage);
588 struct cl_page *lov_page_init_empty(const struct lu_env *env,
589 struct cl_object *obj,
590 struct cl_page *page, cfs_page_t *vmpage);
591 struct cl_page *lov_page_init_raid0(const struct lu_env *env,
592 struct cl_object *obj,
593 struct cl_page *page, cfs_page_t *vmpage);
594 struct lu_object *lov_object_alloc (const struct lu_env *env,
595 const struct lu_object_header *hdr,
596 struct lu_device *dev);
597 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
598 const struct lu_object_header *hdr,
599 struct lu_device *dev);
601 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
602 struct lov_lock *lck,
603 struct lovsub_lock *sub);
604 struct lov_io_sub *lov_page_subio (const struct lu_env *env,
606 const struct cl_page_slice *slice);
609 #define lov_foreach_target(lov, var) \
610 for (var = 0; var < lov_targets_nr(lov); ++var)
612 /*****************************************************************************
620 static inline struct lov_session *lov_env_session(const struct lu_env *env)
622 struct lov_session *ses;
624 ses = lu_context_key_get(env->le_ses, &lov_session_key);
625 LASSERT(ses != NULL);
629 static inline struct lov_io *lov_env_io(const struct lu_env *env)
631 return &lov_env_session(env)->ls_io;
634 static inline int lov_is_object(const struct lu_object *obj)
636 return obj->lo_dev->ld_type == &lov_device_type;
639 static inline int lovsub_is_object(const struct lu_object *obj)
641 return obj->lo_dev->ld_type == &lovsub_device_type;
644 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
646 return &lov->ld_cl.cd_lu_dev;
649 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
651 LINVRNT(d->ld_type == &lov_device_type);
652 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
655 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
657 return &lovsub->acid_cl;
660 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
662 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
665 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
667 LINVRNT(d->ld_type == &lovsub_device_type);
668 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
671 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
673 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
674 return container_of0(d, struct lovsub_device, acid_cl);
677 static inline struct lu_object *lov2lu(struct lov_object *lov)
679 return &lov->lo_cl.co_lu;
682 static inline struct cl_object *lov2cl(struct lov_object *lov)
687 static inline struct lov_object *lu2lov(const struct lu_object *obj)
689 LINVRNT(lov_is_object(obj));
690 return container_of0(obj, struct lov_object, lo_cl.co_lu);
693 static inline struct lov_object *cl2lov(const struct cl_object *obj)
695 LINVRNT(lov_is_object(&obj->co_lu));
696 return container_of0(obj, struct lov_object, lo_cl);
699 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
701 return &los->lso_cl.co_lu;
704 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
709 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
711 LINVRNT(lovsub_is_object(&obj->co_lu));
712 return container_of0(obj, struct lovsub_object, lso_cl);
715 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
717 LINVRNT(lovsub_is_object(obj));
718 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
721 static inline struct lovsub_lock *
722 cl2lovsub_lock(const struct cl_lock_slice *slice)
724 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
725 return container_of(slice, struct lovsub_lock, lss_cl);
728 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
730 const struct cl_lock_slice *slice;
732 slice = cl_lock_at(lock, &lovsub_device_type);
733 LASSERT(slice != NULL);
734 return cl2lovsub_lock(slice);
737 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
739 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
740 return container_of(slice, struct lov_lock, lls_cl);
743 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
745 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
746 return container_of0(slice, struct lov_page, lps_cl);
749 static inline struct lov_req *cl2lov_req(const struct cl_req_slice *slice)
751 return container_of0(slice, struct lov_req, lr_cl);
754 static inline struct lovsub_page *
755 cl2lovsub_page(const struct cl_page_slice *slice)
757 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
758 return container_of0(slice, struct lovsub_page, lsb_cl);
761 static inline struct lovsub_req *cl2lovsub_req(const struct cl_req_slice *slice)
763 return container_of0(slice, struct lovsub_req, lsrq_cl);
766 static inline struct cl_page *lov_sub_page(const struct cl_page_slice *slice)
768 return slice->cpl_page->cp_child;
771 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
772 const struct cl_io_slice *ios)
776 lio = container_of(ios, struct lov_io, lis_cl);
777 LASSERT(lio == lov_env_io(env));
781 static inline int lov_targets_nr(const struct lov_device *lov)
783 return lov->ld_lov->desc.ld_tgt_count;
786 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
788 struct lov_thread_info *info;
790 info = lu_context_key_get(&env->le_ctx, &lov_key);
791 LASSERT(info != NULL);
795 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
797 struct lov_layout_raid0 *raid0;
799 LASSERT(lov->lo_type == LLT_RAID0);
800 raid0 = &lov->u.raid0;
801 LASSERT(raid0->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC ||
802 raid0->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC_V3);