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/
32 * This file is part of Lustre, http://www.lustre.org/
34 * Internal interfaces of LOV layer.
36 * Author: Nikita Danilov <nikita.danilov@sun.com>
37 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
40 #ifndef LOV_CL_INTERNAL_H
41 #define LOV_CL_INTERNAL_H
43 #include <libcfs/libcfs.h>
45 #include <cl_object.h>
46 #include "lov_internal.h"
49 * Logical object volume layer. This layer implements data striping (raid0).
51 * At the lov layer top-entity (object, page, lock, io) is connected to one or
52 * more sub-entities: top-object, representing a file is connected to a set of
53 * sub-objects, each representing a stripe, file-level top-lock is connected
54 * to a set of per-stripe sub-locks, top-page is connected to a (single)
55 * sub-page, and a top-level IO is connected to a set of (potentially
56 * concurrent) sub-IO's.
58 * Sub-object, sub-page, and sub-io have well-defined top-object and top-page
59 * respectively, while a single sub-lock can be part of multiple top-locks.
61 * Reference counting models are different for different types of entities:
63 * - top-object keeps a reference to its sub-objects, and destroys them
64 * when it is destroyed.
66 * - top-page keeps a reference to its sub-page, and destroys it when it
69 * - IO's are not reference counted.
71 * To implement a connection between top and sub entities, lov layer is split
72 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
73 * implementing full set of cl-interfaces. For example, top-object has vvp and
74 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
75 * used to track child-parent relationship.
83 enum lov_device_flags {
84 LOV_DEV_INITIALIZED = BIT(0),
91 /* Data-on-MDT array item in lov_device::ld_md_tgts[] */
92 struct lovdom_device {
93 struct cl_device *ldm_mdc;
99 * XXX Locking of lov-private data is missing.
101 struct cl_device ld_cl;
102 struct lov_obd *ld_lov;
103 /** size of lov_device::ld_target[] array */
105 struct lovsub_device **ld_target;
108 /* Data-on-MDT devices */
110 struct lovdom_device *ld_md_tgts;
111 struct obd_device *ld_lmv;
112 /* LU site for subdevices */
113 struct lu_site ld_site;
119 enum lov_layout_type {
120 LLT_EMPTY, /** empty file without body (mknod + truncate) */
121 LLT_RELEASED, /** file with no objects (data in HSM) */
122 LLT_COMP, /** support composite layout */
123 LLT_FOREIGN, /** foreign layout */
127 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 (lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_FOREIGN))
154 return lov_pattern(lsme->lsme_pattern &
155 ~LOV_PATTERN_OVERSTRIPING);
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 unsigned int lle_preference;
224 struct lu_extent *lle_extent;
225 struct lov_stripe_md_entry *lle_lsme;
226 struct lov_comp_layout_entry_ops *lle_comp_ops;
228 struct lov_layout_raid0 lle_raid0;
229 struct lov_layout_dom lle_dom;
233 struct lov_mirror_entry {
234 unsigned short lre_mirror_id;
235 unsigned short 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 lre_foreign:1; /* set if it is a foreign component */
239 int lre_preference; /* overall preference of this mirror */
241 unsigned short lre_start; /* index to lo_entries, start index of
243 unsigned short lre_end; /* end index of this mirror */
246 enum lov_object_flags {
247 /* Layout is invalid, set when layout lock is lost */
248 LO_LAYOUT_INVALID = 0x1,
252 * lov-specific file state.
254 * lov object has particular layout type, determining how top-object is built
255 * on top of sub-objects. Layout type can change dynamically. When this
256 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
257 * all state pertaining to the old layout type is destroyed, and new state is
258 * constructed. All object methods take said semaphore in the shared mode,
259 * providing serialization against transition between layout types.
261 * To avoid multiple `if' or `switch' statements, selecting behavior for the
262 * current layout type, object methods perform double-dispatch, invoking
263 * function corresponding to the current layout type.
266 struct cl_object lo_cl;
268 * Serializes object operations with transitions between layout types.
270 * This semaphore is taken in shared mode by all object methods, and
271 * is taken in exclusive mode when object type is changed.
273 * \see lov_object::lo_type
275 struct rw_semaphore lo_type_guard;
277 * Type of an object. Protected by lov_object::lo_type_guard.
279 enum lov_layout_type lo_type;
283 unsigned long lo_obj_flags;
285 * How many IOs are on going on this object. Layout can be changed
286 * only if there is no active IO.
288 atomic_t lo_active_ios;
290 * Waitq - wait for no one else is using lo_lsm
292 wait_queue_head_t lo_waitq;
294 * Layout metadata. NULL if empty layout.
296 struct lov_stripe_md *lo_lsm;
298 union lov_layout_state {
299 struct lov_layout_state_empty {
301 struct lov_layout_state_released {
303 struct lov_layout_composite {
305 * flags of lov_comp_md_v1::lcm_flags. Mainly used
310 * For FLR: index of preferred mirror to read.
311 * Preferred mirror is initialized by the preferred
312 * bit of lsme. It can be changed when the preferred
314 * In order to make lov_lsm_entry() return the same
315 * mirror in the same IO context, it's only possible
316 * to change the preferred mirror when the
317 * lo_active_ios reaches zero.
319 int lo_preferred_mirror;
321 * For FLR: Number of (valid) mirrors.
323 unsigned lo_mirror_count;
324 struct lov_mirror_entry *lo_mirrors;
326 * Current entry count of lo_entries, include
329 unsigned int lo_entry_count;
330 struct lov_layout_entry *lo_entries;
334 * Thread that acquired lov_object::lo_type_guard in an exclusive
337 struct task_struct *lo_owner;
340 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
342 LASSERT(lov->lo_type == LLT_COMP);
343 LASSERTF(i < lov->u.composite.lo_entry_count,
344 "entry %d entry_count %d\n", i,
345 lov->u.composite.lo_entry_count);
347 return &lov->u.composite.lo_entries[i].lle_raid0;
350 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
352 LASSERT(lov->lo_lsm != NULL);
353 LASSERT(i < lov->lo_lsm->lsm_entry_count);
355 return lov->lo_lsm->lsm_entries[i];
358 static inline unsigned lov_flr_state(const struct lov_object *lov)
360 if (lov->lo_type != LLT_COMP)
363 return lov->u.composite.lo_flags & LCM_FL_FLR_MASK;
366 static inline bool lov_is_flr(const struct lov_object *lov)
368 return lov_flr_state(lov) != LCM_FL_NONE;
371 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
373 LASSERT(lov->lo_type == LLT_COMP);
374 LASSERTF(i < lov->u.composite.lo_entry_count,
375 "entry %d entry_count %d\n", i,
376 lov->u.composite.lo_entry_count);
378 return &lov->u.composite.lo_entries[i];
381 #define lov_for_layout_entry(lov, entry, start, end) \
382 if (lov->u.composite.lo_entries && \
383 lov->u.composite.lo_entry_count > 0) \
384 for (entry = lov_entry(lov, start); \
385 entry <= lov_entry(lov, end); entry++)
387 #define lov_foreach_layout_entry(lov, entry) \
388 lov_for_layout_entry(lov, entry, 0, \
389 (lov)->u.composite.lo_entry_count - 1)
391 #define lov_foreach_mirror_layout_entry(lov, entry, lre) \
392 lov_for_layout_entry(lov, entry, (lre)->lre_start, (lre)->lre_end)
394 static inline struct lov_mirror_entry *
395 lov_mirror_entry(struct lov_object *lov, int i)
397 LASSERT(i < lov->u.composite.lo_mirror_count);
398 return &lov->u.composite.lo_mirrors[i];
401 #define lov_foreach_mirror_entry(lov, lre) \
402 for (lre = lov_mirror_entry(lov, 0); \
403 lre <= lov_mirror_entry(lov, \
404 lov->u.composite.lo_mirror_count - 1); \
407 static inline unsigned
408 lov_layout_entry_index(struct lov_object *lov, struct lov_layout_entry *entry)
410 struct lov_layout_entry *first = &lov->u.composite.lo_entries[0];
411 unsigned index = (unsigned)(entry - first);
413 LASSERT(entry >= first);
414 LASSERT(index < lov->u.composite.lo_entry_count);
420 * State lov_lock keeps for each sub-lock.
422 struct lov_lock_sub {
423 /** sub-lock itself */
424 struct cl_lock sub_lock;
425 /** Set if the sublock has ever been enqueued, meaning it may
426 * hold resources of underlying layers */
427 unsigned int sub_is_enqueued:1,
433 * lov-specific lock state.
436 struct cl_lock_slice lls_cl;
437 /** Number of sub-locks in this lock */
440 struct lov_lock_sub lls_sub[0];
444 struct cl_page_slice lps_cl;
451 struct lovsub_device {
452 struct cl_device acid_cl;
453 struct cl_device *acid_next;
456 struct lovsub_object {
457 struct cl_object_header lso_header;
458 struct cl_object lso_cl;
459 struct lov_object *lso_super;
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;
471 struct lov_thread_info {
472 struct cl_object_conf lti_stripe_conf;
473 struct lu_fid lti_fid;
474 struct ost_lvb lti_lvb;
475 struct cl_2queue lti_cl2q;
476 struct cl_page_list lti_plist;
480 * State that lov_io maintains for every sub-io.
484 * Linkage into a list (hanging off lov_io::lis_subios)
486 struct list_head sub_list;
488 * Linkage into a list (hanging off lov_io::lis_active) of all
489 * sub-io's active for the current IO iteration.
491 struct list_head sub_linkage;
492 unsigned int sub_subio_index;
494 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
495 * independently, with lov acting as a scheduler to maximize overall
500 * environment, in which sub-io executes.
502 struct lu_env *sub_env;
504 * environment's refcheck.
512 * IO state private for LOV.
514 #define LIS_CACHE_ENTRY_NONE -ENOENT
517 struct cl_io_slice lis_cl;
520 * FLR: index to lo_mirrors. Valid only if lov_is_flr() returns true.
522 * The mirror index of this io. Preserved over cl_io_init()
523 * if io->ci_ndelay_tried is greater than zero.
525 int lis_mirror_index;
527 * FLR: the layout gen when lis_mirror_index was cached. The
528 * mirror index makes sense only when the layout gen doesn't
531 int lis_mirror_layout_gen;
534 * fields below this will be initialized in lov_io_init().
536 unsigned lis_preserved;
539 * Pointer to the object slice. This is a duplicate of
540 * lov_io::lis_cl::cis_object.
542 struct lov_object *lis_object;
544 * Original end-of-io position for this IO, set by the upper layer as
545 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
546 * changes pos and count to fit IO into a single stripe and uses saved
547 * value to determine when IO iterations have to stop.
549 * This is used only for CIT_READ and CIT_WRITE io's.
551 loff_t lis_io_endpos;
554 * Record the stripe index before the truncate size, used for setting OST
555 * object size for truncate. LU-14128. lis_trunc_stripe_index[i] refers to
556 * lov_object.u.composite.lo_entries[i].
558 int *lis_trunc_stripe_index;
561 * starting position within a file, for the current io loop iteration
562 * (stripe), used by ci_io_loop().
566 * end position with in a file, for the current stripe io. This is
567 * exclusive (i.e., next offset after last byte affected by io).
573 * the index of ls_single_subio in ls_subios array
575 int lis_single_subio_index;
576 struct lov_io_sub lis_single_subio;
579 * List of active sub-io's. Active sub-io's are under the range
580 * of [lis_pos, lis_endpos).
582 struct list_head lis_active;
584 * All sub-io's created in this lov_io.
586 struct list_head lis_subios;
587 /* Cached results from stripe & offset calculations for page init */
588 int lis_cached_entry;
589 int lis_cached_stripe;
590 loff_t lis_cached_off;
591 loff_t lis_cached_suboff;
592 struct lov_io_sub *lis_cached_sub;
597 struct lov_sublock_env ls_subenv;
600 extern struct lu_device_type lov_device_type;
601 extern struct lu_device_type lovsub_device_type;
603 extern struct lu_context_key lov_key;
604 extern struct lu_context_key lov_session_key;
606 extern struct kmem_cache *lov_lock_kmem;
607 extern struct kmem_cache *lov_object_kmem;
608 extern struct kmem_cache *lov_thread_kmem;
609 extern struct kmem_cache *lov_session_kmem;
611 extern struct kmem_cache *lovsub_object_kmem;
613 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
614 struct cl_lock *lock, const struct cl_io *io);
615 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
616 struct cl_lock *lock, const struct cl_io *io);
617 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
619 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
621 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
624 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
627 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
628 struct cl_page *page, pgoff_t index);
629 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
630 struct cl_page *page, pgoff_t index);
631 int lov_page_init_foreign(const struct lu_env *env, struct cl_object *obj,
632 struct cl_page *page, pgoff_t index);
633 struct lu_object *lov_object_alloc (const struct lu_env *env,
634 const struct lu_object_header *hdr,
635 struct lu_device *dev);
637 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
638 const struct lu_object_header *hdr,
639 struct lu_device *dev);
641 int lov_page_stripe(const struct cl_page *page);
642 bool lov_page_is_empty(const struct cl_page *page);
643 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
644 int lov_io_layout_at(struct lov_io *lio, __u64 offset);
646 #define lov_foreach_target(lov, var) \
647 for (var = 0; var < lov_targets_nr(lov); ++var)
649 static inline struct lu_extent *lov_io_extent(struct lov_io *io, int i)
651 return &lov_lse(io->lis_object, i)->lsme_extent;
655 * For layout entries within @ext.
657 #define lov_foreach_io_layout(ind, lio, ext) \
658 for (ind = lov_io_layout_at(lio, (ext)->e_start); \
660 lu_extent_is_overlapped(lov_io_extent(lio, ind), ext); \
661 ind = lov_io_layout_at(lio, lov_io_extent(lio, ind)->e_end))
663 /*****************************************************************************
671 static inline struct lov_session *lov_env_session(const struct lu_env *env)
673 struct lov_session *ses;
675 ses = lu_context_key_get(env->le_ses, &lov_session_key);
676 LASSERT(ses != NULL);
680 static inline struct lov_io *lov_env_io(const struct lu_env *env)
682 return &lov_env_session(env)->ls_io;
685 static inline int lov_is_object(const struct lu_object *obj)
687 return obj->lo_dev->ld_type == &lov_device_type;
690 static inline int lovsub_is_object(const struct lu_object *obj)
692 return obj->lo_dev->ld_type == &lovsub_device_type;
695 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
697 return &lov->ld_cl.cd_lu_dev;
700 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
702 LINVRNT(d->ld_type == &lov_device_type);
703 return container_of(d, struct lov_device, ld_cl.cd_lu_dev);
706 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
708 return &lovsub->acid_cl;
711 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
713 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
716 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
718 LINVRNT(d->ld_type == &lovsub_device_type);
719 return container_of(d, struct lovsub_device, acid_cl.cd_lu_dev);
722 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
724 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
725 return container_of(d, struct lovsub_device, acid_cl);
728 static inline struct lu_object *lov2lu(struct lov_object *lov)
730 return &lov->lo_cl.co_lu;
733 static inline struct cl_object *lov2cl(struct lov_object *lov)
738 static inline struct lov_object *lu2lov(const struct lu_object *obj)
740 LINVRNT(lov_is_object(obj));
741 return container_of(obj, struct lov_object, lo_cl.co_lu);
744 static inline struct lov_object *cl2lov(const struct cl_object *obj)
746 LINVRNT(lov_is_object(&obj->co_lu));
747 return container_of(obj, struct lov_object, lo_cl);
750 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
752 return &los->lso_cl.co_lu;
755 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
760 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
762 LINVRNT(lovsub_is_object(&obj->co_lu));
763 return container_of(obj, struct lovsub_object, lso_cl);
766 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
768 LINVRNT(lovsub_is_object(obj));
769 return container_of(obj, struct lovsub_object, lso_cl.co_lu);
772 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
774 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
775 return container_of(slice, struct lov_lock, lls_cl);
778 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
780 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
781 return container_of(slice, struct lov_page, lps_cl);
784 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
785 const struct cl_io_slice *ios)
789 lio = container_of(ios, struct lov_io, lis_cl);
790 LASSERT(lio == lov_env_io(env));
794 static inline int lov_targets_nr(const struct lov_device *lov)
796 return lov->ld_lov->desc.ld_tgt_count;
799 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
801 struct lov_thread_info *info;
803 info = lu_context_key_get(&env->le_ctx, &lov_key);
804 LASSERT(info != NULL);
809 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
810 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,