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 * record the stripe no before the truncate size, used for setting OST
179 * object size for truncate. LU-14128.
181 int lo_trunc_stripeno;
183 * When this is true, lov_object::lo_attr contains
184 * valid up to date attributes for a top-level
185 * object. This field is reset to 0 when attributes of
186 * any sub-object change.
190 * Array of sub-objects. Allocated when top-object is
191 * created (lov_init_raid0()).
193 * Top-object is a strict master of its sub-objects:
194 * it is created before them, and outlives its
195 * children (this later is necessary so that basic
196 * functions like cl_object_top() always
197 * work). Top-object keeps a reference on every
200 * When top-object is destroyed (lov_delete_raid0())
201 * it releases its reference to a sub-object and waits
202 * until the latter is finally destroyed.
204 struct lovsub_object **lo_sub;
208 spinlock_t lo_sub_lock;
210 * Cached object attribute, built from sub-object
213 struct cl_attr lo_attr;
216 struct lov_layout_dom {
217 /* keep this always at first place so DOM layout entry
218 * can be addressed also as RAID0 after initialization.
220 struct lov_layout_raid0 lo_dom_r0;
221 struct lovsub_object *lo_dom;
222 struct lov_oinfo *lo_loi;
225 struct lov_layout_entry {
227 unsigned int lle_valid:1;
228 unsigned int lle_preference;
229 struct lu_extent *lle_extent;
230 struct lov_stripe_md_entry *lle_lsme;
231 struct lov_comp_layout_entry_ops *lle_comp_ops;
233 struct lov_layout_raid0 lle_raid0;
234 struct lov_layout_dom lle_dom;
238 struct lov_mirror_entry {
239 unsigned short lre_mirror_id;
240 unsigned short lre_stale:1, /* set if any components is stale */
241 lre_valid:1, /* set if at least one of components
242 * in this mirror is valid */
243 lre_foreign:1; /* set if it is a foreign component */
244 int lre_preference; /* overall preference of this mirror */
246 unsigned short lre_start; /* index to lo_entries, start index of
248 unsigned short lre_end; /* end index of this mirror */
251 enum lov_object_flags {
252 /* Layout is invalid, set when layout lock is lost */
253 LO_LAYOUT_INVALID = 0x1,
257 * lov-specific file state.
259 * lov object has particular layout type, determining how top-object is built
260 * on top of sub-objects. Layout type can change dynamically. When this
261 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
262 * all state pertaining to the old layout type is destroyed, and new state is
263 * constructed. All object methods take said semaphore in the shared mode,
264 * providing serialization against transition between layout types.
266 * To avoid multiple `if' or `switch' statements, selecting behavior for the
267 * current layout type, object methods perform double-dispatch, invoking
268 * function corresponding to the current layout type.
271 struct cl_object lo_cl;
273 * Serializes object operations with transitions between layout types.
275 * This semaphore is taken in shared mode by all object methods, and
276 * is taken in exclusive mode when object type is changed.
278 * \see lov_object::lo_type
280 struct rw_semaphore lo_type_guard;
282 * Type of an object. Protected by lov_object::lo_type_guard.
284 enum lov_layout_type lo_type;
288 unsigned long lo_obj_flags;
290 * How many IOs are on going on this object. Layout can be changed
291 * only if there is no active IO.
293 atomic_t lo_active_ios;
295 * Waitq - wait for no one else is using lo_lsm
297 wait_queue_head_t lo_waitq;
299 * Layout metadata. NULL if empty layout.
301 struct lov_stripe_md *lo_lsm;
303 union lov_layout_state {
304 struct lov_layout_state_empty {
306 struct lov_layout_state_released {
308 struct lov_layout_composite {
310 * flags of lov_comp_md_v1::lcm_flags. Mainly used
315 * For FLR: index of preferred mirror to read.
316 * Preferred mirror is initialized by the preferred
317 * bit of lsme. It can be changed when the preferred
319 * In order to make lov_lsm_entry() return the same
320 * mirror in the same IO context, it's only possible
321 * to change the preferred mirror when the
322 * lo_active_ios reaches zero.
324 int lo_preferred_mirror;
326 * For FLR: the lock to protect access to
327 * lo_preferred_mirror.
329 spinlock_t lo_write_lock;
331 * For FLR: Number of (valid) mirrors.
333 unsigned lo_mirror_count;
334 struct lov_mirror_entry *lo_mirrors;
336 * Current entry count of lo_entries, include
339 unsigned int lo_entry_count;
340 struct lov_layout_entry *lo_entries;
344 * Thread that acquired lov_object::lo_type_guard in an exclusive
347 struct task_struct *lo_owner;
350 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
352 LASSERT(lov->lo_type == LLT_COMP);
353 LASSERTF(i < lov->u.composite.lo_entry_count,
354 "entry %d entry_count %d\n", i,
355 lov->u.composite.lo_entry_count);
357 return &lov->u.composite.lo_entries[i].lle_raid0;
360 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
362 LASSERT(lov->lo_lsm != NULL);
363 LASSERT(i < lov->lo_lsm->lsm_entry_count);
365 return lov->lo_lsm->lsm_entries[i];
368 static inline unsigned lov_flr_state(const struct lov_object *lov)
370 if (lov->lo_type != LLT_COMP)
373 return lov->u.composite.lo_flags & LCM_FL_FLR_MASK;
376 static inline bool lov_is_flr(const struct lov_object *lov)
378 return lov_flr_state(lov) != LCM_FL_NONE;
381 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
383 LASSERT(lov->lo_type == LLT_COMP);
384 LASSERTF(i < lov->u.composite.lo_entry_count,
385 "entry %d entry_count %d\n", i,
386 lov->u.composite.lo_entry_count);
388 return &lov->u.composite.lo_entries[i];
391 #define lov_for_layout_entry(lov, entry, start, end) \
392 for (entry = lov_entry(lov, start); \
393 entry <= lov_entry(lov, end); entry++)
395 #define lov_foreach_layout_entry(lov, entry) \
396 lov_for_layout_entry(lov, entry, 0, \
397 (lov)->u.composite.lo_entry_count - 1)
399 #define lov_foreach_mirror_layout_entry(lov, entry, lre) \
400 lov_for_layout_entry(lov, entry, (lre)->lre_start, (lre)->lre_end)
402 static inline struct lov_mirror_entry *
403 lov_mirror_entry(struct lov_object *lov, int i)
405 LASSERT(i < lov->u.composite.lo_mirror_count);
406 return &lov->u.composite.lo_mirrors[i];
409 #define lov_foreach_mirror_entry(lov, lre) \
410 for (lre = lov_mirror_entry(lov, 0); \
411 lre <= lov_mirror_entry(lov, \
412 lov->u.composite.lo_mirror_count - 1); \
415 static inline unsigned
416 lov_layout_entry_index(struct lov_object *lov, struct lov_layout_entry *entry)
418 struct lov_layout_entry *first = &lov->u.composite.lo_entries[0];
419 unsigned index = (unsigned)(entry - first);
421 LASSERT(entry >= first);
422 LASSERT(index < lov->u.composite.lo_entry_count);
428 * State lov_lock keeps for each sub-lock.
430 struct lov_lock_sub {
431 /** sub-lock itself */
432 struct cl_lock sub_lock;
433 /** Set if the sublock has ever been enqueued, meaning it may
434 * hold resources of underlying layers */
435 unsigned int sub_is_enqueued:1,
441 * lov-specific lock state.
444 struct cl_lock_slice lls_cl;
445 /** Number of sub-locks in this lock */
448 struct lov_lock_sub lls_sub[0];
452 struct cl_page_slice lps_cl;
459 struct lovsub_device {
460 struct cl_device acid_cl;
461 struct cl_device *acid_next;
464 struct lovsub_object {
465 struct cl_object_header lso_header;
466 struct cl_object lso_cl;
467 struct lov_object *lso_super;
472 * Describe the environment settings for sublocks.
474 struct lov_sublock_env {
475 const struct lu_env *lse_env;
476 struct cl_io *lse_io;
479 struct lov_thread_info {
480 struct cl_object_conf lti_stripe_conf;
481 struct lu_fid lti_fid;
482 struct ost_lvb lti_lvb;
483 struct cl_2queue lti_cl2q;
484 struct cl_page_list lti_plist;
488 * State that lov_io maintains for every sub-io.
492 * Linkage into a list (hanging off lov_io::lis_subios)
494 struct list_head sub_list;
496 * Linkage into a list (hanging off lov_io::lis_active) of all
497 * sub-io's active for the current IO iteration.
499 struct list_head sub_linkage;
500 unsigned int sub_subio_index;
502 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
503 * independently, with lov acting as a scheduler to maximize overall
508 * environment, in which sub-io executes.
510 struct lu_env *sub_env;
512 * environment's refcheck.
520 * IO state private for LOV.
522 #define LIS_CACHE_ENTRY_NONE -ENOENT
525 struct cl_io_slice lis_cl;
528 * FLR: index to lo_mirrors. Valid only if lov_is_flr() returns true.
530 * The mirror index of this io. Preserved over cl_io_init()
531 * if io->ci_ndelay_tried is greater than zero.
533 int lis_mirror_index;
535 * FLR: the layout gen when lis_mirror_index was cached. The
536 * mirror index makes sense only when the layout gen doesn't
539 int lis_mirror_layout_gen;
542 * fields below this will be initialized in lov_io_init().
544 unsigned lis_preserved;
547 * Pointer to the object slice. This is a duplicate of
548 * lov_io::lis_cl::cis_object.
550 struct lov_object *lis_object;
552 * Original end-of-io position for this IO, set by the upper layer as
553 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
554 * changes pos and count to fit IO into a single stripe and uses saved
555 * value to determine when IO iterations have to stop.
557 * This is used only for CIT_READ and CIT_WRITE io's.
559 loff_t lis_io_endpos;
562 * starting position within a file, for the current io loop iteration
563 * (stripe), used by ci_io_loop().
567 * end position with in a file, for the current stripe io. This is
568 * exclusive (i.e., next offset after last byte affected by io).
574 * the index of ls_single_subio in ls_subios array
576 int lis_single_subio_index;
577 struct lov_io_sub lis_single_subio;
580 * List of active sub-io's. Active sub-io's are under the range
581 * of [lis_pos, lis_endpos).
583 struct list_head lis_active;
585 * All sub-io's created in this lov_io.
587 struct list_head lis_subios;
588 /* Cached results from stripe & offset calculations for page init */
589 int lis_cached_entry;
590 int lis_cached_stripe;
591 loff_t lis_cached_off;
592 loff_t lis_cached_suboff;
593 struct lov_io_sub *lis_cached_sub;
598 struct lov_sublock_env ls_subenv;
601 extern struct lu_device_type lov_device_type;
602 extern struct lu_device_type lovsub_device_type;
604 extern struct lu_context_key lov_key;
605 extern struct lu_context_key lov_session_key;
607 extern struct kmem_cache *lov_lock_kmem;
608 extern struct kmem_cache *lov_object_kmem;
609 extern struct kmem_cache *lov_thread_kmem;
610 extern struct kmem_cache *lov_session_kmem;
612 extern struct kmem_cache *lovsub_object_kmem;
614 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
615 struct cl_lock *lock, const struct cl_io *io);
616 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
617 struct cl_lock *lock, const struct cl_io *io);
618 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
620 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
622 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
625 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
628 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
629 struct cl_page *page, pgoff_t index);
630 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
631 struct cl_page *page, pgoff_t index);
632 int lov_page_init_foreign(const struct lu_env *env, struct cl_object *obj,
633 struct cl_page *page, pgoff_t index);
634 struct lu_object *lov_object_alloc (const struct lu_env *env,
635 const struct lu_object_header *hdr,
636 struct lu_device *dev);
638 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
639 const struct lu_object_header *hdr,
640 struct lu_device *dev);
642 int lov_page_stripe(const struct cl_page *page);
643 bool lov_page_is_empty(const struct cl_page *page);
644 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
645 int lov_io_layout_at(struct lov_io *lio, __u64 offset);
647 #define lov_foreach_target(lov, var) \
648 for (var = 0; var < lov_targets_nr(lov); ++var)
650 static inline struct lu_extent *lov_io_extent(struct lov_io *io, int i)
652 return &lov_lse(io->lis_object, i)->lsme_extent;
656 * For layout entries within @ext.
658 #define lov_foreach_io_layout(ind, lio, ext) \
659 for (ind = lov_io_layout_at(lio, (ext)->e_start); \
661 lu_extent_is_overlapped(lov_io_extent(lio, ind), ext); \
662 ind = lov_io_layout_at(lio, lov_io_extent(lio, ind)->e_end))
664 /*****************************************************************************
672 static inline struct lov_session *lov_env_session(const struct lu_env *env)
674 struct lov_session *ses;
676 ses = lu_context_key_get(env->le_ses, &lov_session_key);
677 LASSERT(ses != NULL);
681 static inline struct lov_io *lov_env_io(const struct lu_env *env)
683 return &lov_env_session(env)->ls_io;
686 static inline int lov_is_object(const struct lu_object *obj)
688 return obj->lo_dev->ld_type == &lov_device_type;
691 static inline int lovsub_is_object(const struct lu_object *obj)
693 return obj->lo_dev->ld_type == &lovsub_device_type;
696 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
698 return &lov->ld_cl.cd_lu_dev;
701 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
703 LINVRNT(d->ld_type == &lov_device_type);
704 return container_of(d, struct lov_device, ld_cl.cd_lu_dev);
707 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
709 return &lovsub->acid_cl;
712 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
714 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
717 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
719 LINVRNT(d->ld_type == &lovsub_device_type);
720 return container_of(d, struct lovsub_device, acid_cl.cd_lu_dev);
723 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
725 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
726 return container_of(d, struct lovsub_device, acid_cl);
729 static inline struct lu_object *lov2lu(struct lov_object *lov)
731 return &lov->lo_cl.co_lu;
734 static inline struct cl_object *lov2cl(struct lov_object *lov)
739 static inline struct lov_object *lu2lov(const struct lu_object *obj)
741 LINVRNT(lov_is_object(obj));
742 return container_of(obj, struct lov_object, lo_cl.co_lu);
745 static inline struct lov_object *cl2lov(const struct cl_object *obj)
747 LINVRNT(lov_is_object(&obj->co_lu));
748 return container_of(obj, struct lov_object, lo_cl);
751 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
753 return &los->lso_cl.co_lu;
756 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
761 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
763 LINVRNT(lovsub_is_object(&obj->co_lu));
764 return container_of(obj, struct lovsub_object, lso_cl);
767 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
769 LINVRNT(lovsub_is_object(obj));
770 return container_of(obj, struct lovsub_object, lso_cl.co_lu);
773 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
775 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
776 return container_of(slice, struct lov_lock, lls_cl);
779 static inline struct lov_page *cl2lov_page(const struct cl_page_slice *slice)
781 LINVRNT(lov_is_object(&slice->cpl_obj->co_lu));
782 return container_of(slice, struct lov_page, lps_cl);
785 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
786 const struct cl_io_slice *ios)
790 lio = container_of(ios, struct lov_io, lis_cl);
791 LASSERT(lio == lov_env_io(env));
795 static inline int lov_targets_nr(const struct lov_device *lov)
797 return lov->ld_lov->desc.ld_tgt_count;
800 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
802 struct lov_thread_info *info;
804 info = lu_context_key_get(&env->le_ctx, &lov_key);
805 LASSERT(info != NULL);
810 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
811 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,