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"
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;
226 * When this is true, lov_object::lo_attr contains
227 * valid up to date attributes for a top-level
228 * object. This field is reset to 0 when attributes of
229 * any sub-object change.
233 * Cached object attribute, built from sub-object
236 struct cl_attr lo_attr;
238 struct lov_layout_state_empty {
242 * Thread that acquired lov_object::lo_type_guard in an exclusive
245 cfs_task_t *lo_owner;
249 * Flags that top-lock can set on each of its sub-locks.
252 /** Top-lock acquired a hold (cl_lock_hold()) on a sub-lock. */
257 * State lov_lock keeps for each sub-lock.
259 struct lov_lock_sub {
260 /** sub-lock itself */
261 struct lovsub_lock *sub_lock;
262 /** An array of per-sub-lock flags, taken from enum lov_sub_flags */
265 struct cl_lock_descr sub_descr;
266 struct cl_lock_descr sub_got;
270 * lov-specific lock state.
273 struct cl_lock_slice lls_cl;
274 /** Number of sub-locks in this lock */
277 * Number of existing sub-locks.
279 unsigned lls_nr_filled;
281 * Set when sub-lock was canceled, while top-lock was being
284 int lls_cancel_race:1;
286 * An array of sub-locks
288 * There are two issues with managing sub-locks:
290 * - sub-locks are concurrently canceled, and
292 * - sub-locks are shared with other top-locks.
294 * To manage cancellation, top-lock acquires a hold on a sublock
295 * (lov_sublock_adopt()) when the latter is inserted into
296 * lov_lock::lls_sub[]. This hold is released (lov_sublock_release())
297 * when top-lock is going into CLS_CACHED state or destroyed. Hold
298 * prevents sub-lock from cancellation.
300 * Sub-lock sharing means, among other things, that top-lock that is
301 * in the process of creation (i.e., not yet inserted into lock list)
302 * is already accessible to other threads once at least one of its
303 * sub-locks is created, see lov_lock_sub_init().
305 * Sub-lock can be in one of the following states:
307 * - doesn't exist, lov_lock::lls_sub[]::sub_lock == NULL. Such
308 * sub-lock was either never created (top-lock is in CLS_NEW
309 * state), or it was created, then canceled, then destroyed
310 * (lov_lock_unlink() cleared sub-lock pointer in the top-lock).
312 * - sub-lock exists and is on
313 * hold. (lov_lock::lls_sub[]::sub_flags & LSF_HELD). This is a
314 * normal state of a sub-lock in CLS_HELD and CLS_CACHED states
317 * - sub-lock exists, but is not held by the top-lock. This
318 * happens after top-lock released a hold on sub-locks before
319 * going into cache (lov_lock_unuse()).
321 * \todo To support wide-striping, array has to be replaced with a set
322 * of queues to avoid scanning.
324 struct lov_lock_sub *lls_sub;
326 * Original description with which lock was enqueued.
328 struct cl_lock_descr lls_orig;
332 struct cl_page_slice lps_cl;
340 struct lovsub_device {
341 struct cl_device acid_cl;
342 struct lov_device *acid_super;
344 struct cl_device *acid_next;
347 struct lovsub_object {
348 struct cl_object_header lso_header;
349 struct cl_object lso_cl;
350 struct lov_object *lso_super;
355 * A link between a top-lock and a sub-lock. Separate data-structure is
356 * necessary, because top-locks and sub-locks are in M:N relationship.
358 * \todo This can be optimized for a (by far) most frequent case of a single
359 * top-lock per sub-lock.
361 struct lov_lock_link {
362 struct lov_lock *lll_super;
363 /** An index within parent lock. */
366 * A linkage into per sub-lock list of all corresponding top-locks,
367 * hanging off lovsub_lock::lss_parents.
373 * Lock state at lovsub layer.
376 struct cl_lock_slice lss_cl;
378 * List of top-locks that have given sub-lock as their part. Protected
379 * by cl_lock::cll_guard mutex.
381 cfs_list_t lss_parents;
383 * Top-lock that initiated current operation on this sub-lock. This is
384 * only set during top-to-bottom lock operations like enqueue, and is
385 * used to optimize state change notification. Protected by
386 * cl_lock::cll_guard mutex.
388 * \see lovsub_lock_state_one().
390 struct cl_lock *lss_active;
394 * Describe the environment settings for sublocks.
396 struct lov_sublock_env {
397 const struct lu_env *lse_env;
398 struct cl_io *lse_io;
399 struct lov_io_sub *lse_sub;
403 struct cl_page_slice lsb_cl;
407 struct lov_thread_info {
408 struct cl_object_conf lti_stripe_conf;
409 struct lu_fid lti_fid;
410 struct cl_lock_descr lti_ldescr;
411 struct ost_lvb lti_lvb;
412 struct cl_2queue lti_cl2q;
413 union lov_layout_state lti_state;
414 struct cl_lock_closure lti_closure;
415 cfs_waitlink_t lti_waiter;
419 * State that lov_io maintains for every sub-io.
424 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
425 * independently, with lov acting as a scheduler to maximize overall
428 struct cl_io *sub_io;
430 * Linkage into a list (hanging off lov_io::lis_active) of all
431 * sub-io's active for the current IO iteration.
433 cfs_list_t sub_linkage;
435 * true, iff cl_io_init() was successfully executed against
436 * lov_io_sub::sub_io.
438 int sub_io_initialized;
440 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
441 * allocated, but borrowed from a per-device emergency pool.
445 * environment, in which sub-io executes.
447 struct lu_env *sub_env;
449 * environment's refcheck.
460 * IO state private for LOV.
464 struct cl_io_slice lis_cl;
466 * Pointer to the object slice. This is a duplicate of
467 * lov_io::lis_cl::cis_object.
469 struct lov_object *lis_object;
471 * Original end-of-io position for this IO, set by the upper layer as
472 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
473 * changes pos and count to fit IO into a single stripe and uses saved
474 * value to determine when IO iterations have to stop.
476 * This is used only for CIT_READ and CIT_WRITE io's.
478 loff_t lis_io_endpos;
481 * starting position within a file, for the current io loop iteration
482 * (stripe), used by ci_io_loop().
486 * end position with in a file, for the current stripe io. This is
487 * exclusive (i.e., next offset after last byte affected by io).
492 int lis_stripe_count;
493 int lis_active_subios;
496 * the index of ls_single_subio in ls_subios array
498 int lis_single_subio_index;
499 struct cl_io lis_single_subio;
502 * size of ls_subios array, actually the highest stripe #
505 struct lov_io_sub *lis_subs;
507 * List of active sub-io's.
509 cfs_list_t lis_active;
514 struct lov_sublock_env ls_subenv;
518 * State of transfer for lov.
521 struct cl_req_slice lr_cl;
525 * State of transfer for lovsub.
528 struct cl_req_slice lsrq_cl;
531 extern struct lu_device_type lov_device_type;
532 extern struct lu_device_type lovsub_device_type;
534 extern struct lu_context_key lov_key;
535 extern struct lu_context_key lov_session_key;
537 extern cfs_mem_cache_t *lov_page_kmem;
538 extern cfs_mem_cache_t *lov_lock_kmem;
539 extern cfs_mem_cache_t *lov_object_kmem;
540 extern cfs_mem_cache_t *lov_thread_kmem;
541 extern cfs_mem_cache_t *lov_session_kmem;
542 extern cfs_mem_cache_t *lov_req_kmem;
544 extern cfs_mem_cache_t *lovsub_page_kmem;
545 extern cfs_mem_cache_t *lovsub_lock_kmem;
546 extern cfs_mem_cache_t *lovsub_object_kmem;
547 extern cfs_mem_cache_t *lovsub_req_kmem;
549 extern cfs_mem_cache_t *lov_lock_link_kmem;
551 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
552 const struct lu_object_conf *conf);
553 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
554 const struct lu_object_conf *conf);
555 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
556 struct cl_lock *lock, const struct cl_io *io);
557 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
559 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
560 struct cl_lock *lock, const struct cl_io *io);
562 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
563 struct cl_lock *lock, const struct cl_io *io);
564 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
566 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
568 void lov_lock_unlink (const struct lu_env *env, struct lov_lock_link *link,
569 struct lovsub_lock *sub);
571 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
573 void lov_sub_put (struct lov_io_sub *sub);
574 int lov_sublock_modify (const struct lu_env *env, struct lov_lock *lov,
575 struct lovsub_lock *sublock,
576 const struct cl_lock_descr *d, int idx);
579 struct cl_page *lov_page_init (const struct lu_env *env, struct cl_object *ob,
580 struct cl_page *page, cfs_page_t *vmpage);
581 struct cl_page *lovsub_page_init(const struct lu_env *env, struct cl_object *ob,
582 struct cl_page *page, cfs_page_t *vmpage);
584 struct cl_page *lov_page_init_empty(const struct lu_env *env,
585 struct cl_object *obj,
586 struct cl_page *page, cfs_page_t *vmpage);
587 struct cl_page *lov_page_init_raid0(const struct lu_env *env,
588 struct cl_object *obj,
589 struct cl_page *page, cfs_page_t *vmpage);
590 struct lu_object *lov_object_alloc (const struct lu_env *env,
591 const struct lu_object_header *hdr,
592 struct lu_device *dev);
593 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
594 const struct lu_object_header *hdr,
595 struct lu_device *dev);
597 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
598 struct lov_lock *lck,
599 struct lovsub_lock *sub);
600 struct lov_io_sub *lov_page_subio (const struct lu_env *env,
602 const struct cl_page_slice *slice);
605 #define lov_foreach_target(lov, var) \
606 for (var = 0; var < lov_targets_nr(lov); ++var)
608 /*****************************************************************************
616 static inline struct lov_session *lov_env_session(const struct lu_env *env)
618 struct lov_session *ses;
620 ses = lu_context_key_get(env->le_ses, &lov_session_key);
621 LASSERT(ses != NULL);
625 static inline struct lov_io *lov_env_io(const struct lu_env *env)
627 return &lov_env_session(env)->ls_io;
630 static inline int lov_is_object(const struct lu_object *obj)
632 return obj->lo_dev->ld_type == &lov_device_type;
635 static inline int lovsub_is_object(const struct lu_object *obj)
637 return obj->lo_dev->ld_type == &lovsub_device_type;
640 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
642 return &lov->ld_cl.cd_lu_dev;
645 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
647 LINVRNT(d->ld_type == &lov_device_type);
648 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
651 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
653 return &lovsub->acid_cl;
656 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
658 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
661 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
663 LINVRNT(d->ld_type == &lovsub_device_type);
664 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
667 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
669 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
670 return container_of0(d, struct lovsub_device, acid_cl);
673 static inline struct lu_object *lov2lu(struct lov_object *lov)
675 return &lov->lo_cl.co_lu;
678 static inline struct cl_object *lov2cl(struct lov_object *lov)
683 static inline struct lov_object *lu2lov(const struct lu_object *obj)
685 LINVRNT(lov_is_object(obj));
686 return container_of0(obj, struct lov_object, lo_cl.co_lu);
689 static inline struct lov_object *cl2lov(const struct cl_object *obj)
691 LINVRNT(lov_is_object(&obj->co_lu));
692 return container_of0(obj, struct lov_object, lo_cl);
695 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
697 return &los->lso_cl.co_lu;
700 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
705 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
707 LINVRNT(lovsub_is_object(&obj->co_lu));
708 return container_of0(obj, struct lovsub_object, lso_cl);
711 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
713 LINVRNT(lovsub_is_object(obj));
714 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
717 static inline struct lovsub_lock *
718 cl2lovsub_lock(const struct cl_lock_slice *slice)
720 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
721 return container_of(slice, struct lovsub_lock, lss_cl);
724 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
726 const struct cl_lock_slice *slice;
728 slice = cl_lock_at(lock, &lovsub_device_type);
729 LASSERT(slice != NULL);
730 return cl2lovsub_lock(slice);
733 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
735 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
736 return container_of(slice, struct lov_lock, lls_cl);
739 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
741 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
742 return container_of0(slice, struct lov_page, lps_cl);
745 static inline struct lov_req *cl2lov_req(const struct cl_req_slice *slice)
747 return container_of0(slice, struct lov_req, lr_cl);
750 static inline struct lovsub_page *
751 cl2lovsub_page(const struct cl_page_slice *slice)
753 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
754 return container_of0(slice, struct lovsub_page, lsb_cl);
757 static inline struct lovsub_req *cl2lovsub_req(const struct cl_req_slice *slice)
759 return container_of0(slice, struct lovsub_req, lsrq_cl);
762 static inline struct cl_page *lov_sub_page(const struct cl_page_slice *slice)
764 return slice->cpl_page->cp_child;
767 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
768 const struct cl_io_slice *ios)
772 lio = container_of(ios, struct lov_io, lis_cl);
773 LASSERT(lio == lov_env_io(env));
777 static inline int lov_targets_nr(const struct lov_device *lov)
779 return lov->ld_lov->desc.ld_tgt_count;
782 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
784 struct lov_thread_info *info;
786 info = lu_context_key_get(&env->le_ctx, &lov_key);
787 LASSERT(info != NULL);
791 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
793 struct lov_layout_raid0 *raid0;
795 LASSERT(lov->lo_type == LLT_RAID0);
796 raid0 = &lov->u.raid0;
797 LASSERT(raid0->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC ||
798 raid0->lo_lsm->lsm_wire.lw_magic == LOV_MAGIC_V3);