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, 2015 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 <linux/libcfs/libcfs.h>
48 #include <cl_object.h>
49 #include "lov_internal.h"
52 * Logical object volume layer. This layer implements data striping (raid0).
54 * At the lov layer top-entity (object, page, lock, io) is connected to one or
55 * more sub-entities: top-object, representing a file is connected to a set of
56 * sub-objects, each representing a stripe, file-level top-lock is connected
57 * to a set of per-stripe sub-locks, top-page is connected to a (single)
58 * sub-page, and a top-level IO is connected to a set of (potentially
59 * concurrent) sub-IO's.
61 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
62 * respectively, while a single sub-lock can be part of multiple top-locks.
64 * Reference counting models are different for different types of entities:
66 * - top-object keeps a reference to its sub-objects, and destroys them
67 * when it is destroyed.
69 * - top-page keeps a reference to its sub-page, and destroys it when it
72 * - IO's are not reference counted.
74 * To implement a connection between top and sub entities, lov layer is split
75 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
76 * implementing full set of cl-interfaces. For example, top-object has vvp and
77 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
78 * used to track child-parent relationship.
87 enum lov_device_flags {
88 LOV_DEV_INITIALIZED = 1 << 0
97 * XXX Locking of lov-private data is missing.
99 struct cl_device ld_cl;
100 struct lov_obd *ld_lov;
101 /** size of lov_device::ld_target[] array */
103 struct lovsub_device **ld_target;
110 enum lov_layout_type {
111 LLT_EMPTY, /** empty file without body (mknod + truncate) */
112 LLT_RAID0, /** striped file */
113 LLT_RELEASED, /** file with no objects (data in HSM) */
117 static inline char *llt2str(enum lov_layout_type llt)
134 * lov-specific file state.
136 * lov object has particular layout type, determining how top-object is built
137 * on top of sub-objects. Layout type can change dynamically. When this
138 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
139 * all state pertaining to the old layout type is destroyed, and new state is
140 * constructed. All object methods take said semaphore in the shared mode,
141 * providing serialization against transition between layout types.
143 * To avoid multiple `if' or `switch' statements, selecting behavior for the
144 * current layout type, object methods perform double-dispatch, invoking
145 * function corresponding to the current layout type.
148 struct cl_object lo_cl;
150 * Serializes object operations with transitions between layout types.
152 * This semaphore is taken in shared mode by all object methods, and
153 * is taken in exclusive mode when object type is changed.
155 * \see lov_object::lo_type
157 struct rw_semaphore lo_type_guard;
159 * Type of an object. Protected by lov_object::lo_type_guard.
161 enum lov_layout_type lo_type;
163 * True if layout is invalid. This bit is cleared when layout lock
166 bool lo_layout_invalid;
168 * How many IOs are on going on this object. Layout can be changed
169 * only if there is no active IO.
171 atomic_t lo_active_ios;
173 * Waitq - wait for no one else is using lo_lsm
175 wait_queue_head_t lo_waitq;
177 * Layout metadata. NULL if empty layout.
179 struct lov_stripe_md *lo_lsm;
181 union lov_layout_state {
182 struct lov_layout_raid0 {
185 * When this is true, lov_object::lo_attr contains
186 * valid up to date attributes for a top-level
187 * object. This field is reset to 0 when attributes of
188 * any sub-object change.
192 * Array of sub-objects. Allocated when top-object is
193 * created (lov_init_raid0()).
195 * Top-object is a strict master of its sub-objects:
196 * it is created before them, and outlives its
197 * children (this later is necessary so that basic
198 * functions like cl_object_top() always
199 * work). Top-object keeps a reference on every
202 * When top-object is destroyed (lov_delete_raid0())
203 * it releases its reference to a sub-object and waits
204 * until the latter is finally destroyed.
206 struct lovsub_object **lo_sub;
210 spinlock_t lo_sub_lock;
212 * Cached object attribute, built from sub-object
215 struct cl_attr lo_attr;
217 struct lov_layout_state_empty {
219 struct lov_layout_state_released {
223 * Thread that acquired lov_object::lo_type_guard in an exclusive
226 struct task_struct *lo_owner;
230 * State lov_lock keeps for each sub-lock.
232 struct lov_lock_sub {
233 /** sub-lock itself */
234 struct cl_lock sub_lock;
235 /** Set if the sublock has ever been enqueued, meaning it may
236 * hold resources of underlying layers
238 unsigned int sub_is_enqueued:1,
244 * lov-specific lock state.
247 struct cl_lock_slice lls_cl;
248 /** Number of sub-locks in this lock */
251 struct lov_lock_sub lls_sub[0];
255 struct cl_page_slice lps_cl;
256 unsigned int lps_stripe; /* stripe index */
263 struct lovsub_device {
264 struct cl_device acid_cl;
265 struct cl_device *acid_next;
268 struct lovsub_object {
269 struct cl_object_header lso_header;
270 struct cl_object lso_cl;
271 struct lov_object *lso_super;
276 * Lock state at lovsub layer.
279 struct cl_lock_slice lss_cl;
283 * Describe the environment settings for sublocks.
285 struct lov_sublock_env {
286 const struct lu_env *lse_env;
287 struct cl_io *lse_io;
291 struct cl_page_slice lsb_cl;
294 struct lov_thread_info {
295 struct cl_object_conf lti_stripe_conf;
296 struct lu_fid lti_fid;
297 struct ost_lvb lti_lvb;
298 struct cl_2queue lti_cl2q;
299 struct cl_page_list lti_plist;
300 wait_queue_entry_t lti_waiter;
304 * State that lov_io maintains for every sub-io.
309 * environment's refcheck.
315 * true, iff cl_io_init() was successfully executed against
316 * lov_io_sub::sub_io.
318 u16 sub_io_initialized:1,
320 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
321 * allocated, but borrowed from a per-device emergency pool.
325 * Linkage into a list (hanging off lov_io::lis_active) of all
326 * sub-io's active for the current IO iteration.
328 struct list_head sub_linkage;
330 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
331 * independently, with lov acting as a scheduler to maximize overall
334 struct cl_io *sub_io;
336 * environment, in which sub-io executes.
338 struct lu_env *sub_env;
342 * IO state private for LOV.
346 struct cl_io_slice lis_cl;
348 * Pointer to the object slice. This is a duplicate of
349 * lov_io::lis_cl::cis_object.
351 struct lov_object *lis_object;
353 * Original end-of-io position for this IO, set by the upper layer as
354 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
355 * changes pos and count to fit IO into a single stripe and uses saved
356 * value to determine when IO iterations have to stop.
358 * This is used only for CIT_READ and CIT_WRITE io's.
360 loff_t lis_io_endpos;
363 * starting position within a file, for the current io loop iteration
364 * (stripe), used by ci_io_loop().
368 * end position with in a file, for the current stripe io. This is
369 * exclusive (i.e., next offset after last byte affected by io).
373 int lis_stripe_count;
374 int lis_active_subios;
377 * the index of ls_single_subio in ls_subios array
379 int lis_single_subio_index;
380 struct cl_io lis_single_subio;
383 * size of ls_subios array, actually the highest stripe #
386 struct lov_io_sub *lis_subs;
388 * List of active sub-io's.
390 struct list_head lis_active;
395 struct lov_sublock_env ls_subenv;
398 extern struct lu_device_type lov_device_type;
399 extern struct lu_device_type lovsub_device_type;
401 extern struct lu_context_key lov_key;
402 extern struct lu_context_key lov_session_key;
404 extern struct kmem_cache *lov_lock_kmem;
405 extern struct kmem_cache *lov_object_kmem;
406 extern struct kmem_cache *lov_thread_kmem;
407 extern struct kmem_cache *lov_session_kmem;
409 extern struct kmem_cache *lovsub_lock_kmem;
410 extern struct kmem_cache *lovsub_object_kmem;
412 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
413 const struct lu_object_conf *conf);
414 int lovsub_object_init(const struct lu_env *env, struct lu_object *obj,
415 const struct lu_object_conf *conf);
416 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
417 struct cl_lock *lock, const struct cl_io *io);
418 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
420 int lovsub_lock_init(const struct lu_env *env, struct cl_object *obj,
421 struct cl_lock *lock, const struct cl_io *io);
423 int lov_lock_init_raid0(const struct lu_env *env, struct cl_object *obj,
424 struct cl_lock *lock, const struct cl_io *io);
425 int lov_lock_init_empty(const struct lu_env *env, struct cl_object *obj,
426 struct cl_lock *lock, const struct cl_io *io);
427 int lov_io_init_raid0(const struct lu_env *env, struct cl_object *obj,
429 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
431 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
434 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
437 int lov_page_init(const struct lu_env *env, struct cl_object *ob,
438 struct cl_page *page, pgoff_t index);
439 int lovsub_page_init(const struct lu_env *env, struct cl_object *ob,
440 struct cl_page *page, pgoff_t index);
441 int lov_page_init_empty(const struct lu_env *env, struct cl_object *obj,
442 struct cl_page *page, pgoff_t index);
443 int lov_page_init_raid0(const struct lu_env *env, struct cl_object *obj,
444 struct cl_page *page, pgoff_t index);
445 struct lu_object *lov_object_alloc(const struct lu_env *env,
446 const struct lu_object_header *hdr,
447 struct lu_device *dev);
448 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
449 const struct lu_object_header *hdr,
450 struct lu_device *dev);
452 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
453 int lov_page_stripe(const struct cl_page *page);
455 #define lov_foreach_target(lov, var) \
456 for (var = 0; var < lov_targets_nr(lov); ++var)
458 /*****************************************************************************
466 static inline struct lov_session *lov_env_session(const struct lu_env *env)
468 struct lov_session *ses;
470 ses = lu_context_key_get(env->le_ses, &lov_session_key);
475 static inline struct lov_io *lov_env_io(const struct lu_env *env)
477 return &lov_env_session(env)->ls_io;
480 static inline int lov_is_object(const struct lu_object *obj)
482 return obj->lo_dev->ld_type == &lov_device_type;
485 static inline int lovsub_is_object(const struct lu_object *obj)
487 return obj->lo_dev->ld_type == &lovsub_device_type;
490 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
492 return &lov->ld_cl.cd_lu_dev;
495 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
497 LINVRNT(d->ld_type == &lov_device_type);
498 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
501 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
503 return &lovsub->acid_cl;
506 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
508 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
511 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
513 LINVRNT(d->ld_type == &lovsub_device_type);
514 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
517 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
519 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
520 return container_of0(d, struct lovsub_device, acid_cl);
523 static inline struct lu_object *lov2lu(struct lov_object *lov)
525 return &lov->lo_cl.co_lu;
528 static inline struct cl_object *lov2cl(struct lov_object *lov)
533 static inline struct lov_object *lu2lov(const struct lu_object *obj)
535 LINVRNT(lov_is_object(obj));
536 return container_of0(obj, struct lov_object, lo_cl.co_lu);
539 static inline struct lov_object *cl2lov(const struct cl_object *obj)
541 LINVRNT(lov_is_object(&obj->co_lu));
542 return container_of0(obj, struct lov_object, lo_cl);
545 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
547 return &los->lso_cl.co_lu;
550 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
555 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
557 LINVRNT(lovsub_is_object(&obj->co_lu));
558 return container_of0(obj, struct lovsub_object, lso_cl);
561 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
563 LINVRNT(lovsub_is_object(obj));
564 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
567 static inline struct lovsub_lock *
568 cl2lovsub_lock(const struct cl_lock_slice *slice)
570 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
571 return container_of(slice, struct lovsub_lock, lss_cl);
574 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
576 const struct cl_lock_slice *slice;
578 slice = cl_lock_at(lock, &lovsub_device_type);
580 return cl2lovsub_lock(slice);
583 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
585 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
586 return container_of(slice, struct lov_lock, lls_cl);
589 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
591 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
592 return container_of0(slice, struct lov_page, lps_cl);
595 static inline struct lovsub_page *
596 cl2lovsub_page(const struct cl_page_slice *slice)
598 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
599 return container_of0(slice, struct lovsub_page, lsb_cl);
602 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
603 const struct cl_io_slice *ios)
607 lio = container_of(ios, struct lov_io, lis_cl);
608 LASSERT(lio == lov_env_io(env));
612 static inline int lov_targets_nr(const struct lov_device *lov)
614 return lov->ld_lov->desc.ld_tgt_count;
617 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
619 struct lov_thread_info *info;
621 info = lu_context_key_get(&env->le_ctx, &lov_key);
626 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
628 LASSERT(lov->lo_type == LLT_RAID0);
629 LASSERT(lov->lo_lsm->lsm_magic == LOV_MAGIC ||
630 lov->lo_lsm->lsm_magic == LOV_MAGIC_V3);
631 return &lov->u.raid0;
635 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
636 struct lov_user_md __user *lump);