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
100 * XXX Locking of lov-private data is missing.
102 struct cl_device ld_cl;
103 struct lov_obd *ld_lov;
104 /** size of lov_device::ld_target[] array */
106 struct lovsub_device **ld_target;
113 enum lov_layout_type {
114 LLT_EMPTY, /** empty file without body (mknod + truncate) */
115 LLT_RAID0, /** striped file */
116 LLT_RELEASED, /** file with no objects (data in HSM) */
120 static inline char *llt2str(enum lov_layout_type llt)
137 * lov-specific file state.
139 * lov object has particular layout type, determining how top-object is built
140 * on top of sub-objects. Layout type can change dynamically. When this
141 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
142 * all state pertaining to the old layout type is destroyed, and new state is
143 * constructed. All object methods take said semaphore in the shared mode,
144 * providing serialization against transition between layout types.
146 * To avoid multiple `if' or `switch' statements, selecting behavior for the
147 * current layout type, object methods perform double-dispatch, invoking
148 * function corresponding to the current layout type.
151 struct cl_object lo_cl;
153 * Serializes object operations with transitions between layout types.
155 * This semaphore is taken in shared mode by all object methods, and
156 * is taken in exclusive mode when object type is changed.
158 * \see lov_object::lo_type
160 struct rw_semaphore lo_type_guard;
162 * Type of an object. Protected by lov_object::lo_type_guard.
164 enum lov_layout_type lo_type;
166 * True if layout is invalid. This bit is cleared when layout lock
169 bool lo_layout_invalid;
171 * How many IOs are on going on this object. Layout can be changed
172 * only if there is no active IO.
174 atomic_t lo_active_ios;
176 * Waitq - wait for no one else is using lo_lsm
178 wait_queue_head_t lo_waitq;
180 * Layout metadata. NULL if empty layout.
182 struct lov_stripe_md *lo_lsm;
184 union lov_layout_state {
185 struct lov_layout_raid0 {
188 * When this is true, lov_object::lo_attr contains
189 * valid up to date attributes for a top-level
190 * object. This field is reset to 0 when attributes of
191 * any sub-object change.
195 * Array of sub-objects. Allocated when top-object is
196 * created (lov_init_raid0()).
198 * Top-object is a strict master of its sub-objects:
199 * it is created before them, and outlives its
200 * children (this later is necessary so that basic
201 * functions like cl_object_top() always
202 * work). Top-object keeps a reference on every
205 * When top-object is destroyed (lov_delete_raid0())
206 * it releases its reference to a sub-object and waits
207 * until the latter is finally destroyed.
209 * May be vmalloc'd, must be freed with OBD_FREE_LARGE.
211 struct lovsub_object **lo_sub;
215 spinlock_t lo_sub_lock;
217 * Cached object attribute, built from sub-object
220 struct cl_attr lo_attr;
222 struct lov_layout_state_empty {
224 struct lov_layout_state_released {
228 * Thread that acquired lov_object::lo_type_guard in an exclusive
231 struct task_struct *lo_owner;
235 * State lov_lock keeps for each sub-lock.
237 struct lov_lock_sub {
238 /** sub-lock itself */
239 struct cl_lock sub_lock;
240 /** Set if the sublock has ever been enqueued, meaning it may
241 * hold resources of underlying layers */
242 unsigned int sub_is_enqueued:1,
248 * lov-specific lock state.
251 struct cl_lock_slice lls_cl;
252 /** Number of sub-locks in this lock */
255 struct lov_lock_sub lls_sub[0];
259 struct cl_page_slice lps_cl;
260 unsigned int lps_stripe; /* stripe index */
267 struct lovsub_device {
268 struct cl_device acid_cl;
269 struct cl_device *acid_next;
272 struct lovsub_object {
273 struct cl_object_header lso_header;
274 struct cl_object lso_cl;
275 struct lov_object *lso_super;
280 * Lock state at lovsub layer.
283 struct cl_lock_slice lss_cl;
287 * Describe the environment settings for sublocks.
289 struct lov_sublock_env {
290 const struct lu_env *lse_env;
291 struct cl_io *lse_io;
295 struct cl_page_slice lsb_cl;
299 struct lov_thread_info {
300 struct cl_object_conf lti_stripe_conf;
301 struct lu_fid lti_fid;
302 struct ost_lvb lti_lvb;
303 struct cl_2queue lti_cl2q;
304 struct cl_page_list lti_plist;
305 wait_queue_t lti_waiter;
309 * State that lov_io maintains for every sub-io.
314 * environment's refcheck.
320 * true, iff cl_io_init() was successfully executed against
321 * lov_io_sub::sub_io.
323 __u16 sub_io_initialized:1,
325 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
326 * allocated, but borrowed from a per-device emergency pool.
330 * Linkage into a list (hanging off lov_io::lis_active) of all
331 * sub-io's active for the current IO iteration.
333 struct list_head sub_linkage;
335 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
336 * independently, with lov acting as a scheduler to maximize overall
339 struct cl_io *sub_io;
341 * environment, in which sub-io executes.
343 struct lu_env *sub_env;
347 * IO state private for LOV.
351 struct cl_io_slice lis_cl;
353 * Pointer to the object slice. This is a duplicate of
354 * lov_io::lis_cl::cis_object.
356 struct lov_object *lis_object;
358 * Original end-of-io position for this IO, set by the upper layer as
359 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
360 * changes pos and count to fit IO into a single stripe and uses saved
361 * value to determine when IO iterations have to stop.
363 * This is used only for CIT_READ and CIT_WRITE io's.
365 loff_t lis_io_endpos;
368 * starting position within a file, for the current io loop iteration
369 * (stripe), used by ci_io_loop().
373 * end position with in a file, for the current stripe io. This is
374 * exclusive (i.e., next offset after last byte affected by io).
378 int lis_stripe_count;
379 int lis_active_subios;
382 * the index of ls_single_subio in ls_subios array
384 int lis_single_subio_index;
385 struct cl_io lis_single_subio;
388 * size of ls_subios array, actually the highest stripe #
389 * May be vmalloc'd, must be freed with OBD_FREE_LARGE().
392 struct lov_io_sub *lis_subs;
394 * List of active sub-io's.
396 struct list_head lis_active;
401 struct lov_sublock_env ls_subenv;
404 extern struct lu_device_type lov_device_type;
405 extern struct lu_device_type lovsub_device_type;
407 extern struct lu_context_key lov_key;
408 extern struct lu_context_key lov_session_key;
410 extern struct kmem_cache *lov_lock_kmem;
411 extern struct kmem_cache *lov_object_kmem;
412 extern struct kmem_cache *lov_thread_kmem;
413 extern struct kmem_cache *lov_session_kmem;
415 extern struct kmem_cache *lovsub_lock_kmem;
416 extern struct kmem_cache *lovsub_object_kmem;
418 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
419 const struct lu_object_conf *conf);
420 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
421 const struct lu_object_conf *conf);
422 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
423 struct cl_lock *lock, const struct cl_io *io);
424 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
426 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
427 struct cl_lock *lock, const struct cl_io *io);
429 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
430 struct cl_lock *lock, const struct cl_io *io);
431 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
432 struct cl_lock *lock, const struct cl_io *io);
433 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
435 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
437 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
440 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
443 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
444 struct cl_page *page, pgoff_t index);
445 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
446 struct cl_page *page, pgoff_t index);
447 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
448 struct cl_page *page, pgoff_t index);
449 int lov_page_init_raid0 (const struct lu_env *env, struct cl_object *obj,
450 struct cl_page *page, pgoff_t index);
451 struct lu_object *lov_object_alloc (const struct lu_env *env,
452 const struct lu_object_header *hdr,
453 struct lu_device *dev);
454 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
455 const struct lu_object_header *hdr,
456 struct lu_device *dev);
458 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
459 int lov_page_stripe(const struct cl_page *page);
461 #define lov_foreach_target(lov, var) \
462 for (var = 0; var < lov_targets_nr(lov); ++var)
464 /*****************************************************************************
472 static inline struct lov_session *lov_env_session(const struct lu_env *env)
474 struct lov_session *ses;
476 ses = lu_context_key_get(env->le_ses, &lov_session_key);
477 LASSERT(ses != NULL);
481 static inline struct lov_io *lov_env_io(const struct lu_env *env)
483 return &lov_env_session(env)->ls_io;
486 static inline int lov_is_object(const struct lu_object *obj)
488 return obj->lo_dev->ld_type == &lov_device_type;
491 static inline int lovsub_is_object(const struct lu_object *obj)
493 return obj->lo_dev->ld_type == &lovsub_device_type;
496 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
498 return &lov->ld_cl.cd_lu_dev;
501 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
503 LINVRNT(d->ld_type == &lov_device_type);
504 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
507 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
509 return &lovsub->acid_cl;
512 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
514 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
517 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
519 LINVRNT(d->ld_type == &lovsub_device_type);
520 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
523 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
525 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
526 return container_of0(d, struct lovsub_device, acid_cl);
529 static inline struct lu_object *lov2lu(struct lov_object *lov)
531 return &lov->lo_cl.co_lu;
534 static inline struct cl_object *lov2cl(struct lov_object *lov)
539 static inline struct lov_object *lu2lov(const struct lu_object *obj)
541 LINVRNT(lov_is_object(obj));
542 return container_of0(obj, struct lov_object, lo_cl.co_lu);
545 static inline struct lov_object *cl2lov(const struct cl_object *obj)
547 LINVRNT(lov_is_object(&obj->co_lu));
548 return container_of0(obj, struct lov_object, lo_cl);
551 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
553 return &los->lso_cl.co_lu;
556 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
561 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
563 LINVRNT(lovsub_is_object(&obj->co_lu));
564 return container_of0(obj, struct lovsub_object, lso_cl);
567 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
569 LINVRNT(lovsub_is_object(obj));
570 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
573 static inline struct lovsub_lock *
574 cl2lovsub_lock(const struct cl_lock_slice *slice)
576 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
577 return container_of(slice, struct lovsub_lock, lss_cl);
580 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
582 const struct cl_lock_slice *slice;
584 slice = cl_lock_at(lock, &lovsub_device_type);
585 LASSERT(slice != NULL);
586 return cl2lovsub_lock(slice);
589 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
591 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
592 return container_of(slice, struct lov_lock, lls_cl);
595 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
597 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
598 return container_of0(slice, struct lov_page, lps_cl);
601 static inline struct lovsub_page *
602 cl2lovsub_page(const struct cl_page_slice *slice)
604 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
605 return container_of0(slice, struct lovsub_page, lsb_cl);
608 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
609 const struct cl_io_slice *ios)
613 lio = container_of(ios, struct lov_io, lis_cl);
614 LASSERT(lio == lov_env_io(env));
618 static inline int lov_targets_nr(const struct lov_device *lov)
620 return lov->ld_lov->desc.ld_tgt_count;
623 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
625 struct lov_thread_info *info;
627 info = lu_context_key_get(&env->le_ctx, &lov_key);
628 LASSERT(info != NULL);
632 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
634 LASSERT(lov->lo_type == LLT_RAID0);
635 LASSERT(lov->lo_lsm->lsm_magic == LOV_MAGIC ||
636 lov->lo_lsm->lsm_magic == LOV_MAGIC_V3);
637 return &lov->u.raid0;
641 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
642 struct lov_user_md __user *lump);