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_RAID0, /** striped file */
112 LLT_RELEASED, /** file with no objects (data in HSM) */
116 static inline char *llt2str(enum lov_layout_type llt)
133 * lov-specific file state.
135 * lov object has particular layout type, determining how top-object is built
136 * on top of sub-objects. Layout type can change dynamically. When this
137 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
138 * all state pertaining to the old layout type is destroyed, and new state is
139 * constructed. All object methods take said semaphore in the shared mode,
140 * providing serialization against transition between layout types.
142 * To avoid multiple `if' or `switch' statements, selecting behavior for the
143 * current layout type, object methods perform double-dispatch, invoking
144 * function corresponding to the current layout type.
147 struct cl_object lo_cl;
149 * Serializes object operations with transitions between layout types.
151 * This semaphore is taken in shared mode by all object methods, and
152 * is taken in exclusive mode when object type is changed.
154 * \see lov_object::lo_type
156 struct rw_semaphore lo_type_guard;
158 * Type of an object. Protected by lov_object::lo_type_guard.
160 enum lov_layout_type lo_type;
162 * True if layout is invalid. This bit is cleared when layout lock
165 bool lo_layout_invalid;
167 * How many IOs are on going on this object. Layout can be changed
168 * only if there is no active IO.
170 atomic_t lo_active_ios;
172 * Waitq - wait for no one else is using lo_lsm
174 wait_queue_head_t lo_waitq;
176 * Layout metadata. NULL if empty layout.
178 struct lov_stripe_md *lo_lsm;
180 union lov_layout_state {
181 struct lov_layout_raid0 {
184 * When this is true, lov_object::lo_attr contains
185 * valid up to date attributes for a top-level
186 * object. This field is reset to 0 when attributes of
187 * any sub-object change.
191 * Array of sub-objects. Allocated when top-object is
192 * created (lov_init_raid0()).
194 * Top-object is a strict master of its sub-objects:
195 * it is created before them, and outlives its
196 * children (this later is necessary so that basic
197 * functions like cl_object_top() always
198 * work). Top-object keeps a reference on every
201 * When top-object is destroyed (lov_delete_raid0())
202 * it releases its reference to a sub-object and waits
203 * until the latter is finally destroyed.
205 * May be vmalloc'd, must be freed with OBD_FREE_LARGE.
207 struct lovsub_object **lo_sub;
211 spinlock_t lo_sub_lock;
213 * Cached object attribute, built from sub-object
216 struct cl_attr lo_attr;
218 struct lov_layout_state_empty {
220 struct lov_layout_state_released {
224 * Thread that acquired lov_object::lo_type_guard in an exclusive
227 struct task_struct *lo_owner;
231 * State lov_lock keeps for each sub-lock.
233 struct lov_lock_sub {
234 /** sub-lock itself */
235 struct cl_lock sub_lock;
236 /** Set if the sublock has ever been enqueued, meaning it may
237 * 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;
295 struct lov_thread_info {
296 struct cl_object_conf lti_stripe_conf;
297 struct lu_fid lti_fid;
298 struct ost_lvb lti_lvb;
299 struct cl_2queue lti_cl2q;
300 struct cl_page_list lti_plist;
301 wait_queue_t lti_waiter;
305 * State that lov_io maintains for every sub-io.
310 * environment's refcheck.
316 * true, iff cl_io_init() was successfully executed against
317 * lov_io_sub::sub_io.
319 __u16 sub_io_initialized:1,
321 * True, iff lov_io_sub::sub_io and lov_io_sub::sub_env weren't
322 * allocated, but borrowed from a per-device emergency pool.
326 * Linkage into a list (hanging off lov_io::lis_active) of all
327 * sub-io's active for the current IO iteration.
329 struct list_head sub_linkage;
331 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
332 * independently, with lov acting as a scheduler to maximize overall
335 struct cl_io *sub_io;
337 * environment, in which sub-io executes.
339 struct lu_env *sub_env;
343 * IO state private for LOV.
347 struct cl_io_slice lis_cl;
349 * Pointer to the object slice. This is a duplicate of
350 * lov_io::lis_cl::cis_object.
352 struct lov_object *lis_object;
354 * Original end-of-io position for this IO, set by the upper layer as
355 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
356 * changes pos and count to fit IO into a single stripe and uses saved
357 * value to determine when IO iterations have to stop.
359 * This is used only for CIT_READ and CIT_WRITE io's.
361 loff_t lis_io_endpos;
364 * starting position within a file, for the current io loop iteration
365 * (stripe), used by ci_io_loop().
369 * end position with in a file, for the current stripe io. This is
370 * exclusive (i.e., next offset after last byte affected by io).
374 int lis_stripe_count;
375 int lis_active_subios;
378 * the index of ls_single_subio in ls_subios array
380 int lis_single_subio_index;
381 struct cl_io lis_single_subio;
384 * size of ls_subios array, actually the highest stripe #
385 * May be vmalloc'd, must be freed with OBD_FREE_LARGE().
388 struct lov_io_sub *lis_subs;
390 * List of active sub-io's.
392 struct list_head lis_active;
397 struct lov_sublock_env ls_subenv;
400 extern struct lu_device_type lov_device_type;
401 extern struct lu_device_type lovsub_device_type;
403 extern struct lu_context_key lov_key;
404 extern struct lu_context_key lov_session_key;
406 extern struct kmem_cache *lov_lock_kmem;
407 extern struct kmem_cache *lov_object_kmem;
408 extern struct kmem_cache *lov_thread_kmem;
409 extern struct kmem_cache *lov_session_kmem;
411 extern struct kmem_cache *lovsub_lock_kmem;
412 extern struct kmem_cache *lovsub_object_kmem;
414 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
415 const struct lu_object_conf *conf);
416 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
417 const struct lu_object_conf *conf);
418 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
419 struct cl_lock *lock, const struct cl_io *io);
420 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
422 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
423 struct cl_lock *lock, const struct cl_io *io);
425 int lov_lock_init_raid0 (const struct lu_env *env, struct cl_object *obj,
426 struct cl_lock *lock, const struct cl_io *io);
427 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
428 struct cl_lock *lock, const struct cl_io *io);
429 int lov_io_init_raid0 (const struct lu_env *env, struct cl_object *obj,
431 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
433 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
436 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
439 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
440 struct cl_page *page, pgoff_t index);
441 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
442 struct cl_page *page, pgoff_t index);
443 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
444 struct cl_page *page, pgoff_t index);
445 int lov_page_init_raid0 (const struct lu_env *env, struct cl_object *obj,
446 struct cl_page *page, pgoff_t index);
447 struct lu_object *lov_object_alloc (const struct lu_env *env,
448 const struct lu_object_header *hdr,
449 struct lu_device *dev);
450 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
451 const struct lu_object_header *hdr,
452 struct lu_device *dev);
454 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
455 int lov_page_stripe(const struct cl_page *page);
457 #define lov_foreach_target(lov, var) \
458 for (var = 0; var < lov_targets_nr(lov); ++var)
460 /*****************************************************************************
468 static inline struct lov_session *lov_env_session(const struct lu_env *env)
470 struct lov_session *ses;
472 ses = lu_context_key_get(env->le_ses, &lov_session_key);
473 LASSERT(ses != NULL);
477 static inline struct lov_io *lov_env_io(const struct lu_env *env)
479 return &lov_env_session(env)->ls_io;
482 static inline int lov_is_object(const struct lu_object *obj)
484 return obj->lo_dev->ld_type == &lov_device_type;
487 static inline int lovsub_is_object(const struct lu_object *obj)
489 return obj->lo_dev->ld_type == &lovsub_device_type;
492 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
494 return &lov->ld_cl.cd_lu_dev;
497 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
499 LINVRNT(d->ld_type == &lov_device_type);
500 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
503 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
505 return &lovsub->acid_cl;
508 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
510 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
513 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
515 LINVRNT(d->ld_type == &lovsub_device_type);
516 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
519 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
521 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
522 return container_of0(d, struct lovsub_device, acid_cl);
525 static inline struct lu_object *lov2lu(struct lov_object *lov)
527 return &lov->lo_cl.co_lu;
530 static inline struct cl_object *lov2cl(struct lov_object *lov)
535 static inline struct lov_object *lu2lov(const struct lu_object *obj)
537 LINVRNT(lov_is_object(obj));
538 return container_of0(obj, struct lov_object, lo_cl.co_lu);
541 static inline struct lov_object *cl2lov(const struct cl_object *obj)
543 LINVRNT(lov_is_object(&obj->co_lu));
544 return container_of0(obj, struct lov_object, lo_cl);
547 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
549 return &los->lso_cl.co_lu;
552 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
557 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
559 LINVRNT(lovsub_is_object(&obj->co_lu));
560 return container_of0(obj, struct lovsub_object, lso_cl);
563 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
565 LINVRNT(lovsub_is_object(obj));
566 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
569 static inline struct lovsub_lock *
570 cl2lovsub_lock(const struct cl_lock_slice *slice)
572 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
573 return container_of(slice, struct lovsub_lock, lss_cl);
576 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
578 const struct cl_lock_slice *slice;
580 slice = cl_lock_at(lock, &lovsub_device_type);
581 LASSERT(slice != NULL);
582 return cl2lovsub_lock(slice);
585 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
587 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
588 return container_of(slice, struct lov_lock, lls_cl);
591 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
593 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
594 return container_of0(slice, struct lov_page, lps_cl);
597 static inline struct lovsub_page *
598 cl2lovsub_page(const struct cl_page_slice *slice)
600 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
601 return container_of0(slice, struct lovsub_page, lsb_cl);
604 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
605 const struct cl_io_slice *ios)
609 lio = container_of(ios, struct lov_io, lis_cl);
610 LASSERT(lio == lov_env_io(env));
614 static inline int lov_targets_nr(const struct lov_device *lov)
616 return lov->ld_lov->desc.ld_tgt_count;
619 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
621 struct lov_thread_info *info;
623 info = lu_context_key_get(&env->le_ctx, &lov_key);
624 LASSERT(info != NULL);
628 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov)
630 LASSERT(lov->lo_type == LLT_RAID0);
631 LASSERT(lov->lo_lsm->lsm_magic == LOV_MAGIC ||
632 lov->lo_lsm->lsm_magic == LOV_MAGIC_V3);
633 return &lov->u.raid0;
637 int lov_getstripe(struct lov_object *obj, struct lov_stripe_md *lsm,
638 struct lov_user_md __user *lump);