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, 2017, 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.
85 enum lov_device_flags {
86 LOV_DEV_INITIALIZED = 1 << 0
93 /* Data-on-MDT array item in lov_device::ld_md_tgts[] */
94 struct lovdom_device {
95 struct cl_device *ldm_mdc;
101 * XXX Locking of lov-private data is missing.
103 struct cl_device ld_cl;
104 struct lov_obd *ld_lov;
105 /** size of lov_device::ld_target[] array */
107 struct lovsub_device **ld_target;
110 /* Data-on-MDT devices */
112 struct lovdom_device *ld_md_tgts;
113 struct obd_device *ld_lmv;
114 /* LU site for subdevices */
115 struct lu_site ld_site;
121 enum lov_layout_type {
122 LLT_EMPTY, /** empty file without body (mknod + truncate) */
123 LLT_RELEASED, /** file with no objects (data in HSM) */
124 LLT_COMP, /** support composite layout */
125 LLT_FOREIGN, /** foreign layout */
129 static inline char *llt2str(enum lov_layout_type llt)
148 * Return lov_layout_entry_type associated with a given composite layout
151 static inline __u32 lov_entry_type(struct lov_stripe_md_entry *lsme)
153 if ((lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_RAID0) ||
154 (lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_MDT))
155 return lov_pattern(lsme->lsme_pattern);
159 struct lov_layout_entry;
163 struct lov_comp_layout_entry_ops {
164 int (*lco_init)(const struct lu_env *env, struct lov_device *dev,
165 struct lov_object *lov, unsigned int index,
166 const struct cl_object_conf *conf,
167 struct lov_layout_entry *lle);
168 void (*lco_fini)(const struct lu_env *env,
169 struct lov_layout_entry *lle);
170 int (*lco_getattr)(const struct lu_env *env, struct lov_object *obj,
171 unsigned int index, struct lov_layout_entry *lle,
172 struct cl_attr **attr);
175 struct lov_layout_raid0 {
178 * When this is true, lov_object::lo_attr contains
179 * valid up to date attributes for a top-level
180 * object. This field is reset to 0 when attributes of
181 * any sub-object change.
185 * Array of sub-objects. Allocated when top-object is
186 * created (lov_init_raid0()).
188 * Top-object is a strict master of its sub-objects:
189 * it is created before them, and outlives its
190 * children (this later is necessary so that basic
191 * functions like cl_object_top() always
192 * work). Top-object keeps a reference on every
195 * When top-object is destroyed (lov_delete_raid0())
196 * it releases its reference to a sub-object and waits
197 * until the latter is finally destroyed.
199 struct lovsub_object **lo_sub;
203 spinlock_t lo_sub_lock;
205 * Cached object attribute, built from sub-object
208 struct cl_attr lo_attr;
211 struct lov_layout_dom {
212 /* keep this always at first place so DOM layout entry
213 * can be addressed also as RAID0 after initialization.
215 struct lov_layout_raid0 lo_dom_r0;
216 struct lovsub_object *lo_dom;
217 struct lov_oinfo *lo_loi;
220 struct lov_layout_entry {
222 unsigned int lle_valid:1;
223 struct lu_extent *lle_extent;
224 struct lov_stripe_md_entry *lle_lsme;
225 struct lov_comp_layout_entry_ops *lle_comp_ops;
227 struct lov_layout_raid0 lle_raid0;
228 struct lov_layout_dom lle_dom;
232 struct lov_mirror_entry {
233 unsigned short lre_mirror_id;
234 unsigned short lre_preferred:1,
235 lre_stale:1, /* set if any components is stale */
236 lre_valid:1; /* set if at least one of components
237 * in this mirror is valid */
238 unsigned short lre_start; /* index to lo_entries, start index of
240 unsigned short lre_end; /* end index of this mirror */
244 * lov-specific file state.
246 * lov object has particular layout type, determining how top-object is built
247 * on top of sub-objects. Layout type can change dynamically. When this
248 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
249 * all state pertaining to the old layout type is destroyed, and new state is
250 * constructed. All object methods take said semaphore in the shared mode,
251 * providing serialization against transition between layout types.
253 * To avoid multiple `if' or `switch' statements, selecting behavior for the
254 * current layout type, object methods perform double-dispatch, invoking
255 * function corresponding to the current layout type.
258 struct cl_object lo_cl;
260 * Serializes object operations with transitions between layout types.
262 * This semaphore is taken in shared mode by all object methods, and
263 * is taken in exclusive mode when object type is changed.
265 * \see lov_object::lo_type
267 struct rw_semaphore lo_type_guard;
269 * Type of an object. Protected by lov_object::lo_type_guard.
271 enum lov_layout_type lo_type;
273 * True if layout is invalid. This bit is cleared when layout lock
276 bool lo_layout_invalid;
278 * How many IOs are on going on this object. Layout can be changed
279 * only if there is no active IO.
281 atomic_t lo_active_ios;
283 * Waitq - wait for no one else is using lo_lsm
285 wait_queue_head_t lo_waitq;
287 * Layout metadata. NULL if empty layout.
289 struct lov_stripe_md *lo_lsm;
291 union lov_layout_state {
292 struct lov_layout_state_empty {
294 struct lov_layout_state_released {
296 struct lov_layout_composite {
298 * flags of lov_comp_md_v1::lcm_flags. Mainly used
303 * For FLR: index of preferred mirror to read.
304 * Preferred mirror is initialized by the preferred
305 * bit of lsme. It can be changed when the preferred
307 * In order to make lov_lsm_entry() return the same
308 * mirror in the same IO context, it's only possible
309 * to change the preferred mirror when the
310 * lo_active_ios reaches zero.
312 int lo_preferred_mirror;
314 * For FLR: the lock to protect access to
315 * lo_preferred_mirror.
317 spinlock_t lo_write_lock;
319 * For FLR: Number of (valid) mirrors.
321 unsigned lo_mirror_count;
322 struct lov_mirror_entry *lo_mirrors;
324 * Current entry count of lo_entries, include
327 unsigned int lo_entry_count;
328 struct lov_layout_entry *lo_entries;
332 * Thread that acquired lov_object::lo_type_guard in an exclusive
335 struct task_struct *lo_owner;
338 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
340 LASSERT(lov->lo_type == LLT_COMP);
341 LASSERTF(i < lov->u.composite.lo_entry_count,
342 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
344 return &lov->u.composite.lo_entries[i].lle_raid0;
347 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
349 LASSERT(lov->lo_lsm != NULL);
350 LASSERT(i < lov->lo_lsm->lsm_entry_count);
352 return lov->lo_lsm->lsm_entries[i];
355 static inline unsigned lov_flr_state(const struct lov_object *lov)
357 if (lov->lo_type != LLT_COMP)
360 return lov->u.composite.lo_flags & LCM_FL_FLR_MASK;
363 static inline bool lov_is_flr(const struct lov_object *lov)
365 return lov_flr_state(lov) != LCM_FL_NONE;
368 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
370 LASSERT(lov->lo_type == LLT_COMP);
371 LASSERTF(i < lov->u.composite.lo_entry_count,
372 "entry %d entry_count %d", i, lov->u.composite.lo_entry_count);
374 return &lov->u.composite.lo_entries[i];
377 #define lov_for_layout_entry(lov, entry, start, end) \
378 for (entry = lov_entry(lov, start); \
379 entry <= lov_entry(lov, end); entry++)
381 #define lov_foreach_layout_entry(lov, entry) \
382 lov_for_layout_entry(lov, entry, 0, \
383 (lov)->u.composite.lo_entry_count - 1)
385 #define lov_foreach_mirror_layout_entry(lov, entry, lre) \
386 lov_for_layout_entry(lov, entry, (lre)->lre_start, (lre)->lre_end)
388 static inline struct lov_mirror_entry *
389 lov_mirror_entry(struct lov_object *lov, int i)
391 LASSERT(i < lov->u.composite.lo_mirror_count);
392 return &lov->u.composite.lo_mirrors[i];
395 #define lov_foreach_mirror_entry(lov, lre) \
396 for (lre = lov_mirror_entry(lov, 0); \
397 lre <= lov_mirror_entry(lov, \
398 lov->u.composite.lo_mirror_count - 1); \
401 static inline unsigned
402 lov_layout_entry_index(struct lov_object *lov, struct lov_layout_entry *entry)
404 struct lov_layout_entry *first = &lov->u.composite.lo_entries[0];
405 unsigned index = (unsigned)(entry - first);
407 LASSERT(entry >= first);
408 LASSERT(index < lov->u.composite.lo_entry_count);
414 * State lov_lock keeps for each sub-lock.
416 struct lov_lock_sub {
417 /** sub-lock itself */
418 struct cl_lock sub_lock;
419 /** Set if the sublock has ever been enqueued, meaning it may
420 * hold resources of underlying layers */
421 unsigned int sub_is_enqueued:1,
427 * lov-specific lock state.
430 struct cl_lock_slice lls_cl;
431 /** Number of sub-locks in this lock */
434 struct lov_lock_sub lls_sub[0];
438 struct cl_page_slice lps_cl;
439 /** layout_entry + stripe index, composed using lov_comp_index() */
440 unsigned int lps_index;
441 /* the layout gen when this page was created */
442 __u32 lps_layout_gen;
449 struct lovsub_device {
450 struct cl_device acid_cl;
451 struct cl_device *acid_next;
454 struct lovsub_object {
455 struct cl_object_header lso_header;
456 struct cl_object lso_cl;
457 struct lov_object *lso_super;
462 * Describe the environment settings for sublocks.
464 struct lov_sublock_env {
465 const struct lu_env *lse_env;
466 struct cl_io *lse_io;
469 struct lov_thread_info {
470 struct cl_object_conf lti_stripe_conf;
471 struct lu_fid lti_fid;
472 struct ost_lvb lti_lvb;
473 struct cl_2queue lti_cl2q;
474 struct cl_page_list lti_plist;
475 wait_queue_entry_t lti_waiter;
479 * State that lov_io maintains for every sub-io.
483 * Linkage into a list (hanging off lov_io::lis_subios)
485 struct list_head sub_list;
487 * Linkage into a list (hanging off lov_io::lis_active) of all
488 * sub-io's active for the current IO iteration.
490 struct list_head sub_linkage;
491 unsigned int sub_subio_index;
493 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
494 * independently, with lov acting as a scheduler to maximize overall
499 * environment, in which sub-io executes.
501 struct lu_env *sub_env;
503 * environment's refcheck.
512 * IO state private for LOV.
516 struct cl_io_slice lis_cl;
519 * FLR: index to lo_mirrors. Valid only if lov_is_flr() returns true.
521 * The mirror index of this io. Preserved over cl_io_init()
522 * if io->ci_ndelay_tried is greater than zero.
524 int lis_mirror_index;
526 * FLR: the layout gen when lis_mirror_index was cached. The
527 * mirror index makes sense only when the layout gen doesn't
530 int lis_mirror_layout_gen;
533 * fields below this will be initialized in lov_io_init().
535 unsigned lis_preserved;
538 * Pointer to the object slice. This is a duplicate of
539 * lov_io::lis_cl::cis_object.
541 struct lov_object *lis_object;
543 * Original end-of-io position for this IO, set by the upper layer as
544 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
545 * changes pos and count to fit IO into a single stripe and uses saved
546 * value to determine when IO iterations have to stop.
548 * This is used only for CIT_READ and CIT_WRITE io's.
550 loff_t lis_io_endpos;
553 * starting position within a file, for the current io loop iteration
554 * (stripe), used by ci_io_loop().
558 * end position with in a file, for the current stripe io. This is
559 * exclusive (i.e., next offset after last byte affected by io).
565 * the index of ls_single_subio in ls_subios array
567 int lis_single_subio_index;
568 struct lov_io_sub lis_single_subio;
571 * List of active sub-io's. Active sub-io's are under the range
572 * of [lis_pos, lis_endpos).
574 struct list_head lis_active;
576 * All sub-io's created in this lov_io.
578 struct list_head lis_subios;
584 struct lov_sublock_env ls_subenv;
587 extern struct lu_device_type lov_device_type;
588 extern struct lu_device_type lovsub_device_type;
590 extern struct lu_context_key lov_key;
591 extern struct lu_context_key lov_session_key;
593 extern struct kmem_cache *lov_lock_kmem;
594 extern struct kmem_cache *lov_object_kmem;
595 extern struct kmem_cache *lov_thread_kmem;
596 extern struct kmem_cache *lov_session_kmem;
598 extern struct kmem_cache *lovsub_object_kmem;
600 int lov_object_init (const struct lu_env *env, struct lu_object *obj,
601 const struct lu_object_conf *conf);
602 int lovsub_object_init (const struct lu_env *env, struct lu_object *obj,
603 const struct lu_object_conf *conf);
604 int lov_lock_init (const struct lu_env *env, struct cl_object *obj,
605 struct cl_lock *lock, const struct cl_io *io);
606 int lov_io_init (const struct lu_env *env, struct cl_object *obj,
609 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
610 struct cl_lock *lock, const struct cl_io *io);
611 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
612 struct cl_lock *lock, const struct cl_io *io);
613 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
615 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
617 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
620 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
623 int lov_page_init (const struct lu_env *env, struct cl_object *ob,
624 struct cl_page *page, pgoff_t index);
625 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
626 struct cl_page *page, pgoff_t index);
627 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
628 struct cl_page *page, pgoff_t index);
629 int lov_page_init_foreign(const struct lu_env *env, struct cl_object *obj,
630 struct cl_page *page, pgoff_t index);
631 struct lu_object *lov_object_alloc (const struct lu_env *env,
632 const struct lu_object_header *hdr,
633 struct lu_device *dev);
635 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
636 const struct lu_object_header *hdr,
637 struct lu_device *dev);
639 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
640 int lov_page_stripe(const struct cl_page *page);
641 bool lov_page_is_empty(const struct cl_page *page);
642 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
643 int lov_io_layout_at(struct lov_io *lio, __u64 offset);
645 #define lov_foreach_target(lov, var) \
646 for (var = 0; var < lov_targets_nr(lov); ++var)
648 static inline struct lu_extent *lov_io_extent(struct lov_io *io, int i)
650 return &lov_lse(io->lis_object, i)->lsme_extent;
654 * For layout entries within @ext.
656 #define lov_foreach_io_layout(ind, lio, ext) \
657 for (ind = lov_io_layout_at(lio, (ext)->e_start); \
659 lu_extent_is_overlapped(lov_io_extent(lio, ind), ext); \
660 ind = lov_io_layout_at(lio, lov_io_extent(lio, ind)->e_end))
662 /*****************************************************************************
670 static inline struct lov_session *lov_env_session(const struct lu_env *env)
672 struct lov_session *ses;
674 ses = lu_context_key_get(env->le_ses, &lov_session_key);
675 LASSERT(ses != NULL);
679 static inline struct lov_io *lov_env_io(const struct lu_env *env)
681 return &lov_env_session(env)->ls_io;
684 static inline int lov_is_object(const struct lu_object *obj)
686 return obj->lo_dev->ld_type == &lov_device_type;
689 static inline int lovsub_is_object(const struct lu_object *obj)
691 return obj->lo_dev->ld_type == &lovsub_device_type;
694 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
696 return &lov->ld_cl.cd_lu_dev;
699 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
701 LINVRNT(d->ld_type == &lov_device_type);
702 return container_of0(d, struct lov_device, ld_cl.cd_lu_dev);
705 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
707 return &lovsub->acid_cl;
710 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
712 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
715 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
717 LINVRNT(d->ld_type == &lovsub_device_type);
718 return container_of0(d, struct lovsub_device, acid_cl.cd_lu_dev);
721 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
723 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
724 return container_of0(d, struct lovsub_device, acid_cl);
727 static inline struct lu_object *lov2lu(struct lov_object *lov)
729 return &lov->lo_cl.co_lu;
732 static inline struct cl_object *lov2cl(struct lov_object *lov)
737 static inline struct lov_object *lu2lov(const struct lu_object *obj)
739 LINVRNT(lov_is_object(obj));
740 return container_of0(obj, struct lov_object, lo_cl.co_lu);
743 static inline struct lov_object *cl2lov(const struct cl_object *obj)
745 LINVRNT(lov_is_object(&obj->co_lu));
746 return container_of0(obj, struct lov_object, lo_cl);
749 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
751 return &los->lso_cl.co_lu;
754 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
759 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
761 LINVRNT(lovsub_is_object(&obj->co_lu));
762 return container_of0(obj, struct lovsub_object, lso_cl);
765 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
767 LINVRNT(lovsub_is_object(obj));
768 return container_of0(obj, struct lovsub_object, lso_cl.co_lu);
771 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
773 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
774 return container_of(slice, struct lov_lock, lls_cl);
777 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
779 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
780 return container_of0(slice, struct lov_page, lps_cl);
783 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
784 const struct cl_io_slice *ios)
788 lio = container_of(ios, struct lov_io, lis_cl);
789 LASSERT(lio == lov_env_io(env));
793 static inline int lov_targets_nr(const struct lov_device *lov)
795 return lov->ld_lov->desc.ld_tgt_count;
798 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
800 struct lov_thread_info *info;
802 info = lu_context_key_get(&env->le_ctx, &lov_key);
803 LASSERT(info != NULL);
808 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
809 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,