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 unsigned int lle_valid:1;
221 struct lu_extent *lle_extent;
222 struct lov_stripe_md_entry *lle_lsme;
223 struct lov_comp_layout_entry_ops *lle_comp_ops;
225 struct lov_layout_raid0 lle_raid0;
226 struct lov_layout_dom lle_dom;
230 struct lov_mirror_entry {
231 unsigned short lre_mirror_id;
232 unsigned short lre_preferred:1,
233 lre_valid:1; /* set if at least one of components
234 * in this mirror is valid */
235 unsigned short lre_start; /* index to lo_entries, start index of
237 unsigned short lre_end; /* end index of this mirror */
241 * lov-specific file state.
243 * lov object has particular layout type, determining how top-object is built
244 * on top of sub-objects. Layout type can change dynamically. When this
245 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
246 * all state pertaining to the old layout type is destroyed, and new state is
247 * constructed. All object methods take said semaphore in the shared mode,
248 * providing serialization against transition between layout types.
250 * To avoid multiple `if' or `switch' statements, selecting behavior for the
251 * current layout type, object methods perform double-dispatch, invoking
252 * function corresponding to the current layout type.
255 struct cl_object lo_cl;
257 * Serializes object operations with transitions between layout types.
259 * This semaphore is taken in shared mode by all object methods, and
260 * is taken in exclusive mode when object type is changed.
262 * \see lov_object::lo_type
264 struct rw_semaphore lo_type_guard;
266 * Type of an object. Protected by lov_object::lo_type_guard.
268 enum lov_layout_type lo_type;
270 * True if layout is invalid. This bit is cleared when layout lock
273 bool lo_layout_invalid;
275 * How many IOs are on going on this object. Layout can be changed
276 * only if there is no active IO.
278 atomic_t lo_active_ios;
280 * Waitq - wait for no one else is using lo_lsm
282 wait_queue_head_t lo_waitq;
284 * Layout metadata. NULL if empty layout.
286 struct lov_stripe_md *lo_lsm;
288 union lov_layout_state {
289 struct lov_layout_state_empty {
291 struct lov_layout_state_released {
293 struct lov_layout_composite {
295 * flags of lov_comp_md_v1::lcm_flags. Mainly used
300 * For FLR: index of preferred mirror to read.
301 * Preferred mirror is initialized by the preferred
302 * bit of lsme. It can be changed when the preferred
304 * In order to make lov_lsm_entry() return the same
305 * mirror in the same IO context, it's only possible
306 * to change the preferred mirror when the
307 * lo_active_ios reaches zero.
309 int lo_preferred_mirror;
311 * For FLR: the lock to protect access to
312 * lo_preferred_mirror.
314 spinlock_t lo_write_lock;
316 * For FLR: Number of (valid) mirrors.
318 unsigned lo_mirror_count;
319 struct lov_mirror_entry *lo_mirrors;
321 * Current entry count of lo_entries, include
324 unsigned int lo_entry_count;
325 struct lov_layout_entry *lo_entries;
329 * Thread that acquired lov_object::lo_type_guard in an exclusive
332 struct task_struct *lo_owner;
335 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
337 LASSERT(lov->lo_type == LLT_COMP);
338 LASSERTF(i < lov->u.composite.lo_entry_count,
339 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
341 return &lov->u.composite.lo_entries[i].lle_raid0;
344 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
346 LASSERT(lov->lo_lsm != NULL);
347 LASSERT(i < lov->lo_lsm->lsm_entry_count);
349 return lov->lo_lsm->lsm_entries[i];
352 static inline unsigned lov_flr_state(const struct lov_object *lov)
354 if (lov->lo_type != LLT_COMP)
355 return LCM_FL_NOT_FLR;
357 return lov->u.composite.lo_flags & LCM_FL_FLR_MASK;
360 static inline bool lov_is_flr(const struct lov_object *lov)
362 return lov_flr_state(lov) != LCM_FL_NOT_FLR;
365 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
367 LASSERT(lov->lo_type == LLT_COMP);
368 LASSERTF(i < lov->u.composite.lo_entry_count,
369 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
371 return &lov->u.composite.lo_entries[i];
374 #define lov_for_layout_entry(lov, entry, start, end) \
375 for (entry = lov_entry(lov, start); \
376 entry <= lov_entry(lov, end); entry++)
378 #define lov_foreach_layout_entry(lov, entry) \
379 lov_for_layout_entry(lov, entry, 0, \
380 (lov)->u.composite.lo_entry_count - 1)
382 #define lov_foreach_mirror_layout_entry(lov, entry, lre) \
383 lov_for_layout_entry(lov, entry, (lre)->lre_start, (lre)->lre_end)
385 static inline struct lov_mirror_entry *
386 lov_mirror_entry(struct lov_object *lov, int i)
388 LASSERT(i < lov->u.composite.lo_mirror_count);
389 return &lov->u.composite.lo_mirrors[i];
392 #define lov_foreach_mirror_entry(lov, lre) \
393 for (lre = lov_mirror_entry(lov, 0); \
394 lre <= lov_mirror_entry(lov, \
395 lov->u.composite.lo_mirror_count - 1); \
398 static inline unsigned
399 lov_layout_entry_index(struct lov_object *lov, struct lov_layout_entry *entry)
401 struct lov_layout_entry *first = &lov->u.composite.lo_entries[0];
402 unsigned index = (unsigned)(entry - first);
404 LASSERT(entry >= first);
405 LASSERT(index < lov->u.composite.lo_entry_count);
411 * State lov_lock keeps for each sub-lock.
413 struct lov_lock_sub {
414 /** sub-lock itself */
415 struct cl_lock sub_lock;
416 /** Set if the sublock has ever been enqueued, meaning it may
417 * hold resources of underlying layers */
418 unsigned int sub_is_enqueued:1,
424 * lov-specific lock state.
427 struct cl_lock_slice lls_cl;
428 /** Number of sub-locks in this lock */
431 struct lov_lock_sub lls_sub[0];
435 struct cl_page_slice lps_cl;
436 /** layout_entry + stripe index, composed using lov_comp_index() */
437 unsigned int lps_index;
444 struct lovsub_device {
445 struct cl_device acid_cl;
446 struct cl_device *acid_next;
449 struct lovsub_object {
450 struct cl_object_header lso_header;
451 struct cl_object lso_cl;
452 struct lov_object *lso_super;
457 * Lock state at lovsub layer.
460 struct cl_lock_slice lss_cl;
464 * Describe the environment settings for sublocks.
466 struct lov_sublock_env {
467 const struct lu_env *lse_env;
468 struct cl_io *lse_io;
472 struct cl_page_slice lsb_cl;
476 struct lov_thread_info {
477 struct cl_object_conf lti_stripe_conf;
478 struct lu_fid lti_fid;
479 struct ost_lvb lti_lvb;
480 struct cl_2queue lti_cl2q;
481 struct cl_page_list lti_plist;
482 wait_queue_t lti_waiter;
486 * State that lov_io maintains for every sub-io.
490 * Linkage into a list (hanging off lov_io::lis_subios)
492 struct list_head sub_list;
494 * Linkage into a list (hanging off lov_io::lis_active) of all
495 * sub-io's active for the current IO iteration.
497 struct list_head sub_linkage;
498 unsigned int sub_subio_index;
500 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
501 * independently, with lov acting as a scheduler to maximize overall
506 * environment, in which sub-io executes.
508 struct lu_env *sub_env;
510 * environment's refcheck.
519 * IO state private for LOV.
523 struct cl_io_slice lis_cl;
526 * FLR: index to lo_mirrors. Valid only if lov_is_flr() returns true.
528 * The mirror index of this io. Preserved over cl_io_init()
529 * if io->ci_ndelay_tried is greater than zero.
531 int lis_mirror_index;
533 * FLR: the layout gen when lis_mirror_index was cached. The
534 * mirror index makes sense only when the layout gen doesn't
537 int lis_mirror_layout_gen;
540 * fields below this will be initialized in lov_io_init().
542 unsigned lis_preserved;
545 * Pointer to the object slice. This is a duplicate of
546 * lov_io::lis_cl::cis_object.
548 struct lov_object *lis_object;
550 * Original end-of-io position for this IO, set by the upper layer as
551 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
552 * changes pos and count to fit IO into a single stripe and uses saved
553 * value to determine when IO iterations have to stop.
555 * This is used only for CIT_READ and CIT_WRITE io's.
557 loff_t lis_io_endpos;
560 * starting position within a file, for the current io loop iteration
561 * (stripe), used by ci_io_loop().
565 * end position with in a file, for the current stripe io. This is
566 * exclusive (i.e., next offset after last byte affected by io).
572 * the index of ls_single_subio in ls_subios array
574 int lis_single_subio_index;
575 struct lov_io_sub lis_single_subio;
578 * List of active sub-io's. Active sub-io's are under the range
579 * of [lis_pos, lis_endpos).
581 struct list_head lis_active;
583 * All sub-io's created in this lov_io.
585 struct list_head lis_subios;
591 struct lov_sublock_env ls_subenv;
594 extern struct lu_device_type lov_device_type;
595 extern struct lu_device_type lovsub_device_type;
597 extern struct lu_context_key lov_key;
598 extern struct lu_context_key lov_session_key;
600 extern struct kmem_cache *lov_lock_kmem;
601 extern struct kmem_cache *lov_object_kmem;
602 extern struct kmem_cache *lov_thread_kmem;
603 extern struct kmem_cache *lov_session_kmem;
605 extern struct kmem_cache *lovsub_lock_kmem;
606 extern struct kmem_cache *lovsub_object_kmem;
608 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
609 const struct lu_object_conf *conf);
610 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
611 const struct lu_object_conf *conf);
612 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
613 struct cl_lock *lock, const struct cl_io *io);
614 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
616 int lovsub_lock_init (const struct lu_env *env, struct cl_object *obj,
617 struct cl_lock *lock, const struct cl_io *io);
619 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
620 struct cl_lock *lock, const struct cl_io *io);
621 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
622 struct cl_lock *lock, const struct cl_io *io);
623 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
625 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
627 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
630 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
633 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
634 struct cl_page *page, pgoff_t index);
635 int lovsub_page_init (const struct lu_env *env, struct cl_object *ob,
636 struct cl_page *page, pgoff_t index);
637 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
638 struct cl_page *page, pgoff_t index);
639 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
640 struct cl_page *page, pgoff_t index);
641 struct lu_object *lov_object_alloc (const struct lu_env *env,
642 const struct lu_object_header *hdr,
643 struct lu_device *dev);
644 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
645 const struct lu_object_header *hdr,
646 struct lu_device *dev);
648 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
649 int lov_page_stripe(const struct cl_page *page);
650 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
651 int lov_io_layout_at(struct lov_io *lio, __u64 offset);
653 #define lov_foreach_target(lov, var) \
654 for (var = 0; var < lov_targets_nr(lov); ++var)
656 static inline struct lu_extent *lov_io_extent(struct lov_io *io, int i)
658 return &lov_lse(io->lis_object, i)->lsme_extent;
662 * For layout entries within @ext.
664 #define lov_foreach_io_layout(ind, lio, ext) \
665 for (ind = lov_io_layout_at(lio, (ext)->e_start); \
667 lu_extent_is_overlapped(lov_io_extent(lio, ind), ext); \
668 ind = lov_io_layout_at(lio, lov_io_extent(lio, ind)->e_end))
670 /*****************************************************************************
678 static inline struct lov_session *lov_env_session(const struct lu_env *env)
680 struct lov_session *ses;
682 ses = lu_context_key_get(env->le_ses, &lov_session_key);
683 LASSERT(ses != NULL);
687 static inline struct lov_io *lov_env_io(const struct lu_env *env)
689 return &lov_env_session(env)->ls_io;
692 static inline int lov_is_object(const struct lu_object *obj)
694 return obj->lo_dev->ld_type == &lov_device_type;
697 static inline int lovsub_is_object(const struct lu_object *obj)
699 return obj->lo_dev->ld_type == &lovsub_device_type;
702 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
704 return &lov->ld_cl.cd_lu_dev;
707 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
709 LINVRNT(d->ld_type == &lov_device_type);
710 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
713 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
715 return &lovsub->acid_cl;
718 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
720 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
723 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
725 LINVRNT(d->ld_type == &lovsub_device_type);
726 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
729 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
731 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
732 return container_of0(d, struct lovsub_device, acid_cl);
735 static inline struct lu_object *lov2lu(struct lov_object *lov)
737 return &lov->lo_cl.co_lu;
740 static inline struct cl_object *lov2cl(struct lov_object *lov)
745 static inline struct lov_object *lu2lov(const struct lu_object *obj)
747 LINVRNT(lov_is_object(obj));
748 return container_of0(obj, struct lov_object, lo_cl.co_lu);
751 static inline struct lov_object *cl2lov(const struct cl_object *obj)
753 LINVRNT(lov_is_object(&obj->co_lu));
754 return container_of0(obj, struct lov_object, lo_cl);
757 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
759 return &los->lso_cl.co_lu;
762 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
767 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
769 LINVRNT(lovsub_is_object(&obj->co_lu));
770 return container_of0(obj, struct lovsub_object, lso_cl);
773 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
775 LINVRNT(lovsub_is_object(obj));
776 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
779 static inline struct lovsub_lock *
780 cl2lovsub_lock(const struct cl_lock_slice *slice)
782 LINVRNT(lovsub_is_object(&slice->cls_obj->co_lu));
783 return container_of(slice, struct lovsub_lock, lss_cl);
786 static inline struct lovsub_lock *cl2sub_lock(const struct cl_lock *lock)
788 const struct cl_lock_slice *slice;
790 slice = cl_lock_at(lock, &lovsub_device_type);
791 LASSERT(slice != NULL);
792 return cl2lovsub_lock(slice);
795 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
797 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
798 return container_of(slice, struct lov_lock, lls_cl);
801 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
803 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
804 return container_of0(slice, struct lov_page, lps_cl);
807 static inline struct lovsub_page *
808 cl2lovsub_page(const struct cl_page_slice *slice)
810 LINVRNT(lovsub_is_object(&slice->cpl_obj->co_lu));
811 return container_of0(slice, struct lovsub_page, lsb_cl);
814 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
815 const struct cl_io_slice *ios)
819 lio = container_of(ios, struct lov_io, lis_cl);
820 LASSERT(lio == lov_env_io(env));
824 static inline int lov_targets_nr(const struct lov_device *lov)
826 return lov->ld_lov->desc.ld_tgt_count;
829 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
831 struct lov_thread_info *info;
833 info = lu_context_key_get(&env->le_ctx, &lov_key);
834 LASSERT(info != NULL);
839 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
840 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,