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
96 * XXX Locking of lov-private data is missing.
98 struct cl_device ld_cl;
99 struct lov_obd *ld_lov;
100 /** size of lov_device::ld_target[] array */
102 struct lovsub_device **ld_target;
109 enum lov_layout_type {
110 LLT_EMPTY, /** empty file without body (mknod + truncate) */
111 LLT_RELEASED, /** file with no objects (data in HSM) */
112 LLT_COMP, /** support composite layout */
116 static inline char *llt2str(enum lov_layout_type llt)
132 struct lov_layout_raid0 {
135 * When this is true, lov_object::lo_attr contains
136 * valid up to date attributes for a top-level
137 * object. This field is reset to 0 when attributes of
138 * any sub-object change.
142 * Array of sub-objects. Allocated when top-object is
143 * created (lov_init_raid0()).
145 * Top-object is a strict master of its sub-objects:
146 * it is created before them, and outlives its
147 * children (this later is necessary so that basic
148 * functions like cl_object_top() always
149 * work). Top-object keeps a reference on every
152 * When top-object is destroyed (lov_delete_raid0())
153 * it releases its reference to a sub-object and waits
154 * until the latter is finally destroyed.
156 struct lovsub_object **lo_sub;
160 spinlock_t lo_sub_lock;
162 * Cached object attribute, built from sub-object
165 struct cl_attr lo_attr;
169 * lov-specific file state.
171 * lov object has particular layout type, determining how top-object is built
172 * on top of sub-objects. Layout type can change dynamically. When this
173 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
174 * all state pertaining to the old layout type is destroyed, and new state is
175 * constructed. All object methods take said semaphore in the shared mode,
176 * providing serialization against transition between layout types.
178 * To avoid multiple `if' or `switch' statements, selecting behavior for the
179 * current layout type, object methods perform double-dispatch, invoking
180 * function corresponding to the current layout type.
183 struct cl_object lo_cl;
185 * Serializes object operations with transitions between layout types.
187 * This semaphore is taken in shared mode by all object methods, and
188 * is taken in exclusive mode when object type is changed.
190 * \see lov_object::lo_type
192 struct rw_semaphore lo_type_guard;
194 * Type of an object. Protected by lov_object::lo_type_guard.
196 enum lov_layout_type lo_type;
198 * True if layout is invalid. This bit is cleared when layout lock
201 bool lo_layout_invalid;
203 * How many IOs are on going on this object. Layout can be changed
204 * only if there is no active IO.
206 atomic_t lo_active_ios;
208 * Waitq - wait for no one else is using lo_lsm
210 wait_queue_head_t lo_waitq;
212 * Layout metadata. NULL if empty layout.
214 struct lov_stripe_md *lo_lsm;
216 union lov_layout_state {
217 struct lov_layout_state_empty {
219 struct lov_layout_state_released {
221 struct lov_layout_composite {
223 * Current valid entry count of lo_entries.
225 unsigned int lo_entry_count;
226 struct lov_layout_entry {
227 struct lu_extent lle_extent;
228 struct lov_layout_raid0 lle_raid0;
233 * Thread that acquired lov_object::lo_type_guard in an exclusive
236 struct task_struct *lo_owner;
239 #define lov_foreach_layout_entry(lov, entry) \
240 for (entry = &lov->u.composite.lo_entries[0]; \
241 entry < &lov->u.composite.lo_entries \
242 [lov->u.composite.lo_entry_count]; \
246 * State lov_lock keeps for each sub-lock.
248 struct lov_lock_sub {
249 /** sub-lock itself */
250 struct cl_lock sub_lock;
251 /** Set if the sublock has ever been enqueued, meaning it may
252 * hold resources of underlying layers */
253 unsigned int sub_is_enqueued:1,
259 * lov-specific lock state.
262 struct cl_lock_slice lls_cl;
263 /** Number of sub-locks in this lock */
266 struct lov_lock_sub lls_sub[0];
270 struct cl_page_slice lps_cl;
271 /** layout_entry + stripe index, composed using lov_comp_index() */
272 unsigned int lps_index;
279 struct lovsub_device {
280 struct cl_device acid_cl;
281 struct cl_device *acid_next;
284 struct lovsub_object {
285 struct cl_object_header lso_header;
286 struct cl_object lso_cl;
287 struct lov_object *lso_super;
292 * Lock state at lovsub layer.
295 struct cl_lock_slice lss_cl;
299 * Describe the environment settings for sublocks.
301 struct lov_sublock_env {
302 const struct lu_env *lse_env;
303 struct cl_io *lse_io;
307 struct cl_page_slice lsb_cl;
311 struct lov_thread_info {
312 struct cl_object_conf lti_stripe_conf;
313 struct lu_fid lti_fid;
314 struct ost_lvb lti_lvb;
315 struct cl_2queue lti_cl2q;
316 struct cl_page_list lti_plist;
317 wait_queue_t lti_waiter;
321 * State that lov_io maintains for every sub-io.
325 * Linkage into a list (hanging off lov_io::lis_subios)
327 struct list_head sub_list;
329 * Linkage into a list (hanging off lov_io::lis_active) of all
330 * sub-io's active for the current IO iteration.
332 struct list_head sub_linkage;
333 unsigned int sub_subio_index;
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
341 * environment, in which sub-io executes.
343 struct lu_env *sub_env;
345 * environment's refcheck.
354 * IO state private for LOV.
358 struct cl_io_slice lis_cl;
360 * Pointer to the object slice. This is a duplicate of
361 * lov_io::lis_cl::cis_object.
363 struct lov_object *lis_object;
365 * Original end-of-io position for this IO, set by the upper layer as
366 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
367 * changes pos and count to fit IO into a single stripe and uses saved
368 * value to determine when IO iterations have to stop.
370 * This is used only for CIT_READ and CIT_WRITE io's.
372 loff_t lis_io_endpos;
375 * starting position within a file, for the current io loop iteration
376 * (stripe), used by ci_io_loop().
380 * end position with in a file, for the current stripe io. This is
381 * exclusive (i.e., next offset after last byte affected by io).
387 * the index of ls_single_subio in ls_subios array
389 int lis_single_subio_index;
390 struct lov_io_sub lis_single_subio;
393 * List of active sub-io's. Active sub-io's are under the range
394 * of [lis_pos, lis_endpos).
396 struct list_head lis_active;
398 * All sub-io's created in this lov_io.
400 struct list_head lis_subios;
405 struct lov_sublock_env ls_subenv;
408 extern struct lu_device_type lov_device_type;
409 extern struct lu_device_type lovsub_device_type;
411 extern struct lu_context_key lov_key;
412 extern struct lu_context_key lov_session_key;
414 extern struct kmem_cache *lov_lock_kmem;
415 extern struct kmem_cache *lov_object_kmem;
416 extern struct kmem_cache *lov_thread_kmem;
417 extern struct kmem_cache *lov_session_kmem;
419 extern struct kmem_cache *lovsub_lock_kmem;
420 extern struct kmem_cache *lovsub_object_kmem;
422 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
423 const struct lu_object_conf *conf);
424 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
425 const struct lu_object_conf *conf);
426 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
427 struct cl_lock *lock, const struct cl_io *io);
428 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
430 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
431 struct cl_lock *lock, const struct cl_io *io);
433 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
434 struct cl_lock *lock, const struct cl_io *io);
435 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
436 struct cl_lock *lock, const struct cl_io *io);
437 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
439 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
441 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
444 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
447 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
448 struct cl_page *page, pgoff_t index);
449 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
450 struct cl_page *page, pgoff_t index);
451 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
452 struct cl_page *page, pgoff_t index);
453 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
454 struct cl_page *page, pgoff_t index);
455 struct lu_object *lov_object_alloc (const struct lu_env *env,
456 const struct lu_object_header *hdr,
457 struct lu_device *dev);
458 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
459 const struct lu_object_header *hdr,
460 struct lu_device *dev);
462 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
463 int lov_page_stripe(const struct cl_page *page);
464 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
466 #define lov_foreach_target(lov, var) \
467 for (var = 0; var < lov_targets_nr(lov); ++var)
469 /*****************************************************************************
477 static inline struct lov_session *lov_env_session(const struct lu_env *env)
479 struct lov_session *ses;
481 ses = lu_context_key_get(env->le_ses, &lov_session_key);
482 LASSERT(ses != NULL);
486 static inline struct lov_io *lov_env_io(const struct lu_env *env)
488 return &lov_env_session(env)->ls_io;
491 static inline int lov_is_object(const struct lu_object *obj)
493 return obj->lo_dev->ld_type == &lov_device_type;
496 static inline int lovsub_is_object(const struct lu_object *obj)
498 return obj->lo_dev->ld_type == &lovsub_device_type;
501 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
503 return &lov->ld_cl.cd_lu_dev;
506 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
508 LINVRNT(d->ld_type == &lov_device_type);
509 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
512 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
514 return &lovsub->acid_cl;
517 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
519 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
522 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
524 LINVRNT(d->ld_type == &lovsub_device_type);
525 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
528 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
530 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
531 return container_of0(d, struct lovsub_device, acid_cl);
534 static inline struct lu_object *lov2lu(struct lov_object *lov)
536 return &lov->lo_cl.co_lu;
539 static inline struct cl_object *lov2cl(struct lov_object *lov)
544 static inline struct lov_object *lu2lov(const struct lu_object *obj)
546 LINVRNT(lov_is_object(obj));
547 return container_of0(obj, struct lov_object, lo_cl.co_lu);
550 static inline struct lov_object *cl2lov(const struct cl_object *obj)
552 LINVRNT(lov_is_object(&obj->co_lu));
553 return container_of0(obj, struct lov_object, lo_cl);
556 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
558 return &los->lso_cl.co_lu;
561 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
566 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
568 LINVRNT(lovsub_is_object(&obj->co_lu));
569 return container_of0(obj, struct lovsub_object, lso_cl);
572 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
574 LINVRNT(lovsub_is_object(obj));
575 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
578 static inline struct lovsub_lock *
579 cl2lovsub_lock(const struct cl_lock_slice *slice)
581 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
582 return container_of(slice, struct lovsub_lock, lss_cl);
585 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
587 const struct cl_lock_slice *slice;
589 slice = cl_lock_at(lock, &lovsub_device_type);
590 LASSERT(slice != NULL);
591 return cl2lovsub_lock(slice);
594 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
596 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
597 return container_of(slice, struct lov_lock, lls_cl);
600 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
602 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
603 return container_of0(slice, struct lov_page, lps_cl);
606 static inline struct lovsub_page *
607 cl2lovsub_page(const struct cl_page_slice *slice)
609 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
610 return container_of0(slice, struct lovsub_page, lsb_cl);
613 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
614 const struct cl_io_slice *ios)
618 lio = container_of(ios, struct lov_io, lis_cl);
619 LASSERT(lio == lov_env_io(env));
623 static inline int lov_targets_nr(const struct lov_device *lov)
625 return lov->ld_lov->desc.ld_tgt_count;
628 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
630 struct lov_thread_info *info;
632 info = lu_context_key_get(&env->le_ctx, &lov_key);
633 LASSERT(info != NULL);
637 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
639 LASSERT(lov->lo_type == LLT_COMP);
640 LASSERTF(i < lov->u.composite.lo_entry_count,
641 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
643 return &lov->u.composite.lo_entries[i].lle_raid0;
646 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
648 LASSERT(lov->lo_lsm != NULL);
649 LASSERT(i < lov->lo_lsm->lsm_entry_count);
651 return lov->lo_lsm->lsm_entries[i];
655 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
656 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,