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, 2015, 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 <libcfs/libcfs.h>
51 #include <cl_object.h>
52 #include "lov_internal.h"
55 * Logical object volume layer. This layer implements data striping (raid0).
57 * At the lov layer top-entity (object, page, lock, io) is connected to one or
58 * more sub-entities: top-object, representing a file is connected to a set of
59 * sub-objects, each representing a stripe, file-level top-lock is connected
60 * to a set of per-stripe sub-locks, top-page is connected to a (single)
61 * sub-page, and a top-level IO is connected to a set of (potentially
62 * concurrent) sub-IO's.
64 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
65 * respectively, while a single sub-lock can be part of multiple top-locks.
67 * Reference counting models are different for different types of entities:
69 * - top-object keeps a reference to its sub-objects, and destroys them
70 * when it is destroyed.
72 * - top-page keeps a reference to its sub-page, and destroys it when it
75 * - IO's are not reference counted.
77 * To implement a connection between top and sub entities, lov layer is split
78 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
79 * implementing full set of cl-interfaces. For example, top-object has vvp and
80 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
81 * used to track child-parent relationship.
90 enum lov_device_flags {
91 LOV_DEV_INITIALIZED = 1 << 0
99 * Resources that are used in memory-cleaning path, and whose allocation
100 * cannot fail even when memory is tight. They are preallocated in sufficient
101 * quantities in lov_device::ld_emerg[], and access to them is serialized
102 * lov_device::ld_mutex.
104 struct lov_device_emerg {
106 * Page list used to submit IO when memory is in pressure.
108 struct cl_page_list emrg_page_list;
110 * sub-io's shared by all threads accessing this device when memory is
111 * too low to allocate sub-io's dynamically.
113 struct cl_io emrg_subio;
115 * Environments used by sub-io's in
116 * lov_device_emerg::emrg_subio.
118 struct lu_env *emrg_env;
120 * Refchecks for lov_device_emerg::emrg_env.
129 * XXX Locking of lov-private data is missing.
131 struct cl_device ld_cl;
132 struct lov_obd *ld_lov;
133 /** size of lov_device::ld_target[] array */
135 struct lovsub_device **ld_target;
138 /** Emergency resources used in memory-cleansing paths. */
139 struct lov_device_emerg **ld_emrg;
141 * Serializes access to lov_device::ld_emrg in low-memory
144 struct mutex ld_mutex;
150 enum lov_layout_type {
151 LLT_EMPTY, /** empty file without body (mknod + truncate) */
152 LLT_RAID0, /** striped file */
153 LLT_RELEASED, /** file with no objects (data in HSM) */
157 static inline char *llt2str(enum lov_layout_type llt)
174 * lov-specific file state.
176 * lov object has particular layout type, determining how top-object is built
177 * on top of sub-objects. Layout type can change dynamically. When this
178 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
179 * all state pertaining to the old layout type is destroyed, and new state is
180 * constructed. All object methods take said semaphore in the shared mode,
181 * providing serialization against transition between layout types.
183 * To avoid multiple `if' or `switch' statements, selecting behavior for the
184 * current layout type, object methods perform double-dispatch, invoking
185 * function corresponding to the current layout type.
188 struct cl_object lo_cl;
190 * Serializes object operations with transitions between layout types.
192 * This semaphore is taken in shared mode by all object methods, and
193 * is taken in exclusive mode when object type is changed.
195 * \see lov_object::lo_type
197 struct rw_semaphore lo_type_guard;
199 * Type of an object. Protected by lov_object::lo_type_guard.
201 enum lov_layout_type lo_type;
203 * True if layout is invalid. This bit is cleared when layout lock
206 bool lo_layout_invalid;
208 * How many IOs are on going on this object. Layout can be changed
209 * only if there is no active IO.
211 atomic_t lo_active_ios;
213 * Waitq - wait for no one else is using lo_lsm
215 wait_queue_head_t lo_waitq;
217 * Layout metadata. NULL if empty layout.
219 struct lov_stripe_md *lo_lsm;
221 union lov_layout_state {
222 struct lov_layout_raid0 {
225 * When this is true, lov_object::lo_attr contains
226 * valid up to date attributes for a top-level
227 * object. This field is reset to 0 when attributes of
228 * any sub-object change.
232 * Array of sub-objects. Allocated when top-object is
233 * created (lov_init_raid0()).
235 * Top-object is a strict master of its sub-objects:
236 * it is created before them, and outlives its
237 * children (this later is necessary so that basic
238 * functions like cl_object_top() always
239 * work). Top-object keeps a reference on every
242 * When top-object is destroyed (lov_delete_raid0())
243 * it releases its reference to a sub-object and waits
244 * until the latter is finally destroyed.
246 * May be vmalloc'd, must be freed with OBD_FREE_LARGE.
248 struct lovsub_object **lo_sub;
252 spinlock_t lo_sub_lock;
254 * Cached object attribute, built from sub-object
257 struct cl_attr lo_attr;
259 struct lov_layout_state_empty {
261 struct lov_layout_state_released {
265 * Thread that acquired lov_object::lo_type_guard in an exclusive
268 struct task_struct *lo_owner;
272 * State lov_lock keeps for each sub-lock.
274 struct lov_lock_sub {
275 /** sub-lock itself */
276 struct cl_lock sub_lock;
277 /** Set if the sublock has ever been enqueued, meaning it may
278 * hold resources of underlying layers */
279 unsigned int sub_is_enqueued:1,
285 * lov-specific lock state.
288 struct cl_lock_slice lls_cl;
289 /** Number of sub-locks in this lock */
292 struct lov_lock_sub lls_sub[0];
296 struct cl_page_slice lps_cl;
297 unsigned int lps_stripe; /* stripe index */
304 struct lovsub_device {
305 struct cl_device acid_cl;
306 struct lov_device *acid_super;
308 struct cl_device *acid_next;
311 struct lovsub_object {
312 struct cl_object_header lso_header;
313 struct cl_object lso_cl;
314 struct lov_object *lso_super;
319 * A link between a top-lock and a sub-lock. Separate data-structure is
320 * necessary, because top-locks and sub-locks are in M:N relationship.
322 * \todo This can be optimized for a (by far) most frequent case of a single
323 * top-lock per sub-lock.
325 struct lov_lock_link {
326 struct lov_lock *lll_super;
327 /** An index within parent lock. */
330 * A linkage into per sub-lock list of all corresponding top-locks,
331 * hanging off lovsub_lock::lss_parents.
333 struct list_head lll_list;
337 * Lock state at lovsub layer.
340 struct cl_lock_slice lss_cl;
342 * List of top-locks that have given sub-lock as their part. Protected
343 * by cl_lock::cll_guard mutex.
345 struct list_head lss_parents;
347 * Top-lock that initiated current operation on this sub-lock. This is
348 * only set during top-to-bottom lock operations like enqueue, and is
349 * used to optimize state change notification. Protected by
350 * cl_lock::cll_guard mutex.
352 * \see lovsub_lock_state_one().
354 struct cl_lock *lss_active;
358 * Describe the environment settings for sublocks.
360 struct lov_sublock_env {
361 const struct lu_env *lse_env;
362 struct cl_io *lse_io;
363 struct lov_io_sub *lse_sub;
367 struct cl_page_slice lsb_cl;
371 struct lov_thread_info {
372 struct cl_object_conf lti_stripe_conf;
373 struct lu_fid lti_fid;
374 struct cl_lock_descr lti_ldescr;
375 struct ost_lvb lti_lvb;
376 struct cl_2queue lti_cl2q;
377 struct cl_page_list lti_plist;
378 wait_queue_t lti_waiter;
379 struct cl_attr lti_attr;
383 * State that lov_io maintains for every sub-io.
388 * environment's refcheck.
395 * true, iff cl_io_init() was successfully executed against
396 * lov_io_sub::sub_io.
398 __u16 sub_io_initialized:1,
400 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
401 * allocated, but borrowed from a per-device emergency pool.
405 * Linkage into a list (hanging off lov_io::lis_active) of all
406 * sub-io's active for the current IO iteration.
408 struct list_head sub_linkage;
410 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
411 * independently, with lov acting as a scheduler to maximize overall
414 struct cl_io *sub_io;
416 * environment, in which sub-io executes.
418 struct lu_env *sub_env;
422 * IO state private for LOV.
426 struct cl_io_slice lis_cl;
428 * Pointer to the object slice. This is a duplicate of
429 * lov_io::lis_cl::cis_object.
431 struct lov_object *lis_object;
433 * Original end-of-io position for this IO, set by the upper layer as
434 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
435 * changes pos and count to fit IO into a single stripe and uses saved
436 * value to determine when IO iterations have to stop.
438 * This is used only for CIT_READ and CIT_WRITE io's.
440 loff_t lis_io_endpos;
443 * starting position within a file, for the current io loop iteration
444 * (stripe), used by ci_io_loop().
448 * end position with in a file, for the current stripe io. This is
449 * exclusive (i.e., next offset after last byte affected by io).
454 int lis_stripe_count;
455 int lis_active_subios;
458 * the index of ls_single_subio in ls_subios array
460 int lis_single_subio_index;
461 struct cl_io lis_single_subio;
464 * size of ls_subios array, actually the highest stripe #
465 * May be vmalloc'd, must be freed with OBD_FREE_LARGE().
468 struct lov_io_sub *lis_subs;
470 * List of active sub-io's.
472 struct list_head lis_active;
477 struct lov_sublock_env ls_subenv;
480 extern struct lu_device_type lov_device_type;
481 extern struct lu_device_type lovsub_device_type;
483 extern struct lu_context_key lov_key;
484 extern struct lu_context_key lov_session_key;
486 extern struct kmem_cache *lov_lock_kmem;
487 extern struct kmem_cache *lov_object_kmem;
488 extern struct kmem_cache *lov_thread_kmem;
489 extern struct kmem_cache *lov_session_kmem;
491 extern struct kmem_cache *lovsub_lock_kmem;
492 extern struct kmem_cache *lovsub_object_kmem;
494 extern struct kmem_cache *lov_lock_link_kmem;
496 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
497 const struct lu_object_conf *conf);
498 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
499 const struct lu_object_conf *conf);
500 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
501 struct cl_lock *lock, const struct cl_io *io);
502 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
504 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
505 struct cl_lock *lock, const struct cl_io *io);
507 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
508 struct cl_lock *lock, const struct cl_io *io);
509 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
510 struct cl_lock *lock, const struct cl_io *io);
511 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
513 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
515 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
517 void lov_lock_unlink (const struct lu_env *env, struct lov_lock_link *link,
518 struct lovsub_lock *sub);
520 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
522 void lov_sub_put (struct lov_io_sub *sub);
523 int lov_sublock_modify (const struct lu_env *env, struct lov_lock *lov,
524 struct lovsub_lock *sublock,
525 const struct cl_lock_descr *d, int idx);
528 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
529 struct cl_page *page, pgoff_t index);
530 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
531 struct cl_page *page, pgoff_t index);
532 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
533 struct cl_page *page, pgoff_t index);
534 int lov_page_init_raid0 (const struct lu_env *env, struct cl_object *obj,
535 struct cl_page *page, pgoff_t index);
536 struct lu_object *lov_object_alloc (const struct lu_env *env,
537 const struct lu_object_header *hdr,
538 struct lu_device *dev);
539 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
540 const struct lu_object_header *hdr,
541 struct lu_device *dev);
543 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
544 struct lov_lock *lck,
545 struct lovsub_lock *sub);
546 struct lov_io_sub *lov_page_subio (const struct lu_env *env,
548 const struct cl_page_slice *slice);
550 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
551 int lov_page_stripe(const struct cl_page *page);
553 #define lov_foreach_target(lov, var) \
554 for (var = 0; var < lov_targets_nr(lov); ++var)
556 /*****************************************************************************
564 static inline struct lov_session *lov_env_session(const struct lu_env *env)
566 struct lov_session *ses;
568 ses = lu_context_key_get(env->le_ses, &lov_session_key);
569 LASSERT(ses != NULL);
573 static inline struct lov_io *lov_env_io(const struct lu_env *env)
575 return &lov_env_session(env)->ls_io;
578 static inline int lov_is_object(const struct lu_object *obj)
580 return obj->lo_dev->ld_type == &lov_device_type;
583 static inline int lovsub_is_object(const struct lu_object *obj)
585 return obj->lo_dev->ld_type == &lovsub_device_type;
588 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
590 return &lov->ld_cl.cd_lu_dev;
593 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
595 LINVRNT(d->ld_type == &lov_device_type);
596 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
599 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
601 return &lovsub->acid_cl;
604 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
606 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
609 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
611 LINVRNT(d->ld_type == &lovsub_device_type);
612 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
615 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
617 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
618 return container_of0(d, struct lovsub_device, acid_cl);
621 static inline struct lu_object *lov2lu(struct lov_object *lov)
623 return &lov->lo_cl.co_lu;
626 static inline struct cl_object *lov2cl(struct lov_object *lov)
631 static inline struct lov_object *lu2lov(const struct lu_object *obj)
633 LINVRNT(lov_is_object(obj));
634 return container_of0(obj, struct lov_object, lo_cl.co_lu);
637 static inline struct lov_object *cl2lov(const struct cl_object *obj)
639 LINVRNT(lov_is_object(&obj->co_lu));
640 return container_of0(obj, struct lov_object, lo_cl);
643 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
645 return &los->lso_cl.co_lu;
648 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
653 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
655 LINVRNT(lovsub_is_object(&obj->co_lu));
656 return container_of0(obj, struct lovsub_object, lso_cl);
659 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
661 LINVRNT(lovsub_is_object(obj));
662 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
665 static inline struct lovsub_lock *
666 cl2lovsub_lock(const struct cl_lock_slice *slice)
668 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
669 return container_of(slice, struct lovsub_lock, lss_cl);
672 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
674 const struct cl_lock_slice *slice;
676 slice = cl_lock_at(lock, &lovsub_device_type);
677 LASSERT(slice != NULL);
678 return cl2lovsub_lock(slice);
681 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
683 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
684 return container_of(slice, struct lov_lock, lls_cl);
687 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
689 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
690 return container_of0(slice, struct lov_page, lps_cl);
693 static inline struct lovsub_page *
694 cl2lovsub_page(const struct cl_page_slice *slice)
696 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
697 return container_of0(slice, struct lovsub_page, lsb_cl);
700 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
701 const struct cl_io_slice *ios)
705 lio = container_of(ios, struct lov_io, lis_cl);
706 LASSERT(lio == lov_env_io(env));
710 static inline int lov_targets_nr(const struct lov_device *lov)
712 return lov->ld_lov->desc.ld_tgt_count;
715 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
717 struct lov_thread_info *info;
719 info = lu_context_key_get(&env->le_ctx, &lov_key);
720 LASSERT(info != NULL);
724 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
726 LASSERT(lov->lo_type == LLT_RAID0);
727 LASSERT(lov->lo_lsm->lsm_magic == LOV_MAGIC ||
728 lov->lo_lsm->lsm_magic == LOV_MAGIC_V3);
729 return &lov->u.raid0;
733 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
734 struct lov_user_md __user *lump);