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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Internal interfaces of LOV layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #ifndef LOV_CL_INTERNAL_H
43 #define LOV_CL_INTERNAL_H
45 #include <libcfs/libcfs.h>
47 #include <cl_object.h>
48 #include "lov_internal.h"
51 * Logical object volume layer. This layer implements data striping (raid0).
53 * At the lov layer top-entity (object, page, lock, io) is connected to one or
54 * more sub-entities: top-object, representing a file is connected to a set of
55 * sub-objects, each representing a stripe, file-level top-lock is connected
56 * to a set of per-stripe sub-locks, top-page is connected to a (single)
57 * sub-page, and a top-level IO is connected to a set of (potentially
58 * concurrent) sub-IO's.
60 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
61 * respectively, while a single sub-lock can be part of multiple top-locks.
63 * Reference counting models are different for different types of entities:
65 * - top-object keeps a reference to its sub-objects, and destroys them
66 * when it is destroyed.
68 * - top-page keeps a reference to its sub-page, and destroys it when it
71 * - IO's are not reference counted.
73 * To implement a connection between top and sub entities, lov layer is split
74 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
75 * implementing full set of cl-interfaces. For example, top-object has vvp and
76 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
77 * used to track child-parent relationship.
86 enum lov_device_flags {
87 LOV_DEV_INITIALIZED = 1 << 0
94 /* Data-on-MDT array item in lov_device::ld_md_tgts[] */
95 struct lovdom_device {
96 struct cl_device *ldm_mdc;
102 * XXX Locking of lov-private data is missing.
104 struct cl_device ld_cl;
105 struct lov_obd *ld_lov;
106 /** size of lov_device::ld_target[] array */
108 struct lovsub_device **ld_target;
111 /* Data-on-MDT devices */
113 struct lovdom_device *ld_md_tgts;
114 struct obd_device *ld_lmv;
115 /* LU site for subdevices */
116 struct lu_site ld_site;
122 enum lov_layout_type {
123 LLT_EMPTY, /** empty file without body (mknod + truncate) */
124 LLT_RELEASED, /** file with no objects (data in HSM) */
125 LLT_COMP, /** support composite layout */
129 static inline char *llt2str(enum lov_layout_type llt)
146 * Return lov_layout_entry_type associated with a given composite layout
149 static inline __u32 lov_entry_type(struct lov_stripe_md_entry *lsme)
151 if ((lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_RAID0) ||
152 (lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_MDT))
153 return lov_pattern(lsme->lsme_pattern);
157 struct lov_layout_entry;
161 struct lov_comp_layout_entry_ops {
162 int (*lco_init)(const struct lu_env *env, struct lov_device *dev,
163 struct lov_object *lov, unsigned int index,
164 const struct cl_object_conf *conf,
165 struct lov_layout_entry *lle);
166 void (*lco_fini)(const struct lu_env *env,
167 struct lov_layout_entry *lle);
168 int (*lco_getattr)(const struct lu_env *env, struct lov_object *obj,
169 unsigned int index, struct lov_layout_entry *lle,
170 struct cl_attr **attr);
173 struct lov_layout_raid0 {
176 * When this is true, lov_object::lo_attr contains
177 * valid up to date attributes for a top-level
178 * object. This field is reset to 0 when attributes of
179 * any sub-object change.
183 * Array of sub-objects. Allocated when top-object is
184 * created (lov_init_raid0()).
186 * Top-object is a strict master of its sub-objects:
187 * it is created before them, and outlives its
188 * children (this later is necessary so that basic
189 * functions like cl_object_top() always
190 * work). Top-object keeps a reference on every
193 * When top-object is destroyed (lov_delete_raid0())
194 * it releases its reference to a sub-object and waits
195 * until the latter is finally destroyed.
197 struct lovsub_object **lo_sub;
201 spinlock_t lo_sub_lock;
203 * Cached object attribute, built from sub-object
206 struct cl_attr lo_attr;
209 struct lov_layout_dom {
210 /* keep this always at first place so DOM layout entry
211 * can be addressed also as RAID0 after initialization.
213 struct lov_layout_raid0 lo_dom_r0;
214 struct lovsub_object *lo_dom;
215 struct lov_oinfo *lo_loi;
218 struct lov_layout_entry {
220 struct lu_extent lle_extent;
221 struct lov_comp_layout_entry_ops *lle_comp_ops;
223 struct lov_layout_raid0 lle_raid0;
224 struct lov_layout_dom lle_dom;
229 * lov-specific file state.
231 * lov object has particular layout type, determining how top-object is built
232 * on top of sub-objects. Layout type can change dynamically. When this
233 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
234 * all state pertaining to the old layout type is destroyed, and new state is
235 * constructed. All object methods take said semaphore in the shared mode,
236 * providing serialization against transition between layout types.
238 * To avoid multiple `if' or `switch' statements, selecting behavior for the
239 * current layout type, object methods perform double-dispatch, invoking
240 * function corresponding to the current layout type.
243 struct cl_object lo_cl;
245 * Serializes object operations with transitions between layout types.
247 * This semaphore is taken in shared mode by all object methods, and
248 * is taken in exclusive mode when object type is changed.
250 * \see lov_object::lo_type
252 struct rw_semaphore lo_type_guard;
254 * Type of an object. Protected by lov_object::lo_type_guard.
256 enum lov_layout_type lo_type;
258 * True if layout is invalid. This bit is cleared when layout lock
261 bool lo_layout_invalid;
263 * How many IOs are on going on this object. Layout can be changed
264 * only if there is no active IO.
266 atomic_t lo_active_ios;
268 * Waitq - wait for no one else is using lo_lsm
270 wait_queue_head_t lo_waitq;
272 * Layout metadata. NULL if empty layout.
274 struct lov_stripe_md *lo_lsm;
276 union lov_layout_state {
277 struct lov_layout_state_empty {
279 struct lov_layout_state_released {
281 struct lov_layout_composite {
283 * Current valid entry count of entries.
285 unsigned int lo_entry_count;
286 struct lov_layout_entry *lo_entries;
290 * Thread that acquired lov_object::lo_type_guard in an exclusive
293 struct task_struct *lo_owner;
296 #define lov_foreach_layout_entry(lov, entry) \
297 for (entry = &lov->u.composite.lo_entries[0]; \
298 entry < &lov->u.composite.lo_entries \
299 [lov->u.composite.lo_entry_count]; \
303 * State lov_lock keeps for each sub-lock.
305 struct lov_lock_sub {
306 /** sub-lock itself */
307 struct cl_lock sub_lock;
308 /** Set if the sublock has ever been enqueued, meaning it may
309 * hold resources of underlying layers */
310 unsigned int sub_is_enqueued:1,
316 * lov-specific lock state.
319 struct cl_lock_slice lls_cl;
320 /** Number of sub-locks in this lock */
323 struct lov_lock_sub lls_sub[0];
327 struct cl_page_slice lps_cl;
328 /** layout_entry + stripe index, composed using lov_comp_index() */
329 unsigned int lps_index;
336 struct lovsub_device {
337 struct cl_device acid_cl;
338 struct cl_device *acid_next;
341 struct lovsub_object {
342 struct cl_object_header lso_header;
343 struct cl_object lso_cl;
344 struct lov_object *lso_super;
349 * Lock state at lovsub layer.
352 struct cl_lock_slice lss_cl;
356 * Describe the environment settings for sublocks.
358 struct lov_sublock_env {
359 const struct lu_env *lse_env;
360 struct cl_io *lse_io;
364 struct cl_page_slice lsb_cl;
368 struct lov_thread_info {
369 struct cl_object_conf lti_stripe_conf;
370 struct lu_fid lti_fid;
371 struct ost_lvb lti_lvb;
372 struct cl_2queue lti_cl2q;
373 struct cl_page_list lti_plist;
374 wait_queue_t lti_waiter;
378 * State that lov_io maintains for every sub-io.
382 * Linkage into a list (hanging off lov_io::lis_subios)
384 struct list_head sub_list;
386 * Linkage into a list (hanging off lov_io::lis_active) of all
387 * sub-io's active for the current IO iteration.
389 struct list_head sub_linkage;
390 unsigned int sub_subio_index;
392 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
393 * independently, with lov acting as a scheduler to maximize overall
398 * environment, in which sub-io executes.
400 struct lu_env *sub_env;
402 * environment's refcheck.
411 * IO state private for LOV.
415 struct cl_io_slice lis_cl;
417 * Pointer to the object slice. This is a duplicate of
418 * lov_io::lis_cl::cis_object.
420 struct lov_object *lis_object;
422 * Original end-of-io position for this IO, set by the upper layer as
423 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
424 * changes pos and count to fit IO into a single stripe and uses saved
425 * value to determine when IO iterations have to stop.
427 * This is used only for CIT_READ and CIT_WRITE io's.
429 loff_t lis_io_endpos;
432 * starting position within a file, for the current io loop iteration
433 * (stripe), used by ci_io_loop().
437 * end position with in a file, for the current stripe io. This is
438 * exclusive (i.e., next offset after last byte affected by io).
444 * the index of ls_single_subio in ls_subios array
446 int lis_single_subio_index;
447 struct lov_io_sub lis_single_subio;
450 * List of active sub-io's. Active sub-io's are under the range
451 * of [lis_pos, lis_endpos).
453 struct list_head lis_active;
455 * All sub-io's created in this lov_io.
457 struct list_head lis_subios;
462 struct lov_sublock_env ls_subenv;
465 extern struct lu_device_type lov_device_type;
466 extern struct lu_device_type lovsub_device_type;
468 extern struct lu_context_key lov_key;
469 extern struct lu_context_key lov_session_key;
471 extern struct kmem_cache *lov_lock_kmem;
472 extern struct kmem_cache *lov_object_kmem;
473 extern struct kmem_cache *lov_thread_kmem;
474 extern struct kmem_cache *lov_session_kmem;
476 extern struct kmem_cache *lovsub_lock_kmem;
477 extern struct kmem_cache *lovsub_object_kmem;
479 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
480 const struct lu_object_conf *conf);
481 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
482 const struct lu_object_conf *conf);
483 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
484 struct cl_lock *lock, const struct cl_io *io);
485 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
487 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
488 struct cl_lock *lock, const struct cl_io *io);
490 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
491 struct cl_lock *lock, const struct cl_io *io);
492 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
493 struct cl_lock *lock, const struct cl_io *io);
494 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
496 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
498 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
501 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
504 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
505 struct cl_page *page, pgoff_t index);
506 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
507 struct cl_page *page, pgoff_t index);
508 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
509 struct cl_page *page, pgoff_t index);
510 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
511 struct cl_page *page, pgoff_t index);
512 struct lu_object *lov_object_alloc (const struct lu_env *env,
513 const struct lu_object_header *hdr,
514 struct lu_device *dev);
515 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
516 const struct lu_object_header *hdr,
517 struct lu_device *dev);
519 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
520 int lov_page_stripe(const struct cl_page *page);
521 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
523 #define lov_foreach_target(lov, var) \
524 for (var = 0; var < lov_targets_nr(lov); ++var)
526 /*****************************************************************************
534 static inline struct lov_session *lov_env_session(const struct lu_env *env)
536 struct lov_session *ses;
538 ses = lu_context_key_get(env->le_ses, &lov_session_key);
539 LASSERT(ses != NULL);
543 static inline struct lov_io *lov_env_io(const struct lu_env *env)
545 return &lov_env_session(env)->ls_io;
548 static inline int lov_is_object(const struct lu_object *obj)
550 return obj->lo_dev->ld_type == &lov_device_type;
553 static inline int lovsub_is_object(const struct lu_object *obj)
555 return obj->lo_dev->ld_type == &lovsub_device_type;
558 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
560 return &lov->ld_cl.cd_lu_dev;
563 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
565 LINVRNT(d->ld_type == &lov_device_type);
566 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
569 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
571 return &lovsub->acid_cl;
574 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
576 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
579 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
581 LINVRNT(d->ld_type == &lovsub_device_type);
582 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
585 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
587 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
588 return container_of0(d, struct lovsub_device, acid_cl);
591 static inline struct lu_object *lov2lu(struct lov_object *lov)
593 return &lov->lo_cl.co_lu;
596 static inline struct cl_object *lov2cl(struct lov_object *lov)
601 static inline struct lov_object *lu2lov(const struct lu_object *obj)
603 LINVRNT(lov_is_object(obj));
604 return container_of0(obj, struct lov_object, lo_cl.co_lu);
607 static inline struct lov_object *cl2lov(const struct cl_object *obj)
609 LINVRNT(lov_is_object(&obj->co_lu));
610 return container_of0(obj, struct lov_object, lo_cl);
613 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
615 return &los->lso_cl.co_lu;
618 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
623 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
625 LINVRNT(lovsub_is_object(&obj->co_lu));
626 return container_of0(obj, struct lovsub_object, lso_cl);
629 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
631 LINVRNT(lovsub_is_object(obj));
632 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
635 static inline struct lovsub_lock *
636 cl2lovsub_lock(const struct cl_lock_slice *slice)
638 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
639 return container_of(slice, struct lovsub_lock, lss_cl);
642 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
644 const struct cl_lock_slice *slice;
646 slice = cl_lock_at(lock, &lovsub_device_type);
647 LASSERT(slice != NULL);
648 return cl2lovsub_lock(slice);
651 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
653 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
654 return container_of(slice, struct lov_lock, lls_cl);
657 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
659 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
660 return container_of0(slice, struct lov_page, lps_cl);
663 static inline struct lovsub_page *
664 cl2lovsub_page(const struct cl_page_slice *slice)
666 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
667 return container_of0(slice, struct lovsub_page, lsb_cl);
670 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
671 const struct cl_io_slice *ios)
675 lio = container_of(ios, struct lov_io, lis_cl);
676 LASSERT(lio == lov_env_io(env));
680 static inline int lov_targets_nr(const struct lov_device *lov)
682 return lov->ld_lov->desc.ld_tgt_count;
685 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
687 struct lov_thread_info *info;
689 info = lu_context_key_get(&env->le_ctx, &lov_key);
690 LASSERT(info != NULL);
694 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
696 LASSERT(lov->lo_type == LLT_COMP);
697 LASSERTF(i < lov->u.composite.lo_entry_count,
698 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
700 return &lov->u.composite.lo_entries[i];
703 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
705 LASSERT(lov->lo_type == LLT_COMP);
706 LASSERTF(i < lov->u.composite.lo_entry_count,
707 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
709 return &lov->u.composite.lo_entries[i].lle_raid0;
712 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
714 LASSERT(lov->lo_lsm != NULL);
715 LASSERT(i < lov->lo_lsm->lsm_entry_count);
717 return lov->lo_lsm->lsm_entries[i];
721 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
722 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,