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, 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, and a top-level IO is connected to a set of
55 * (potentially concurrent) sub-IO's.
57 * Sub-object and sub-io have well-defined top-object and top-io
58 * respectively, while a single sub-lock can be part of multiple top-locks.
60 * Reference counting models are different for different types of entities:
62 * - top-object keeps a reference to its sub-objects, and destroys them
63 * when it is destroyed.
65 * - IO's are not reference counted.
67 * To implement a connection between top and sub entities, lov layer is split
68 * into two pieces: lov ("upper half"), and lovsub ("bottom half"), both
69 * implementing full set of cl-interfaces. For example, top-object has vvp and
70 * lov layers, and it's sub-object has lovsub and osc layers. lovsub layer is
71 * used to track child-parent relationship.
79 enum lov_device_flags {
80 LOV_DEV_INITIALIZED = BIT(0),
87 /* Data-on-MDT array item in lov_device::ld_md_tgts[] */
88 struct lovdom_device {
89 struct cl_device *ldm_mdc;
95 * XXX Locking of lov-private data is missing.
97 struct cl_device ld_cl;
98 struct lov_obd *ld_lov;
99 /** size of lov_device::ld_target[] array */
101 struct lovsub_device **ld_target;
104 /* Data-on-MDT devices */
106 struct lovdom_device *ld_md_tgts;
107 struct obd_device *ld_lmv;
108 /* LU site for subdevices */
109 struct lu_site ld_site;
115 enum lov_layout_type {
116 LLT_EMPTY, /** empty file without body (mknod + truncate) */
117 LLT_RELEASED, /** file with no objects (data in HSM) */
118 LLT_COMP, /** support composite layout */
119 LLT_FOREIGN, /** foreign layout */
123 static inline char *llt2str(enum lov_layout_type llt)
142 * Return lov_layout_entry_type associated with a given composite layout
145 static inline __u32 lov_entry_type(struct lov_stripe_md_entry *lsme)
147 if ((lov_pattern(lsme->lsme_pattern) & LOV_PATTERN_RAID0) ||
148 (lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_MDT) ||
149 (lov_pattern(lsme->lsme_pattern) == LOV_PATTERN_FOREIGN))
150 return lov_pattern(lsme->lsme_pattern &
151 ~LOV_PATTERN_OVERSTRIPING);
155 struct lov_layout_entry;
159 struct lov_comp_layout_entry_ops {
160 int (*lco_init)(const struct lu_env *env, struct lov_device *dev,
161 struct lov_object *lov, unsigned int index,
162 const struct cl_object_conf *conf,
163 struct lov_layout_entry *lle);
164 void (*lco_fini)(const struct lu_env *env,
165 struct lov_layout_entry *lle);
166 int (*lco_getattr)(const struct lu_env *env, struct lov_object *obj,
167 unsigned int index, struct lov_layout_entry *lle,
168 struct cl_attr **attr);
171 struct lov_layout_raid0 {
174 * When this is true, lov_object::lo_attr contains
175 * valid up to date attributes for a top-level
176 * object. This field is reset to 0 when attributes of
177 * any sub-object change.
181 * Array of sub-objects. Allocated when top-object is
182 * created (lov_init_raid0()).
184 * Top-object is a strict master of its sub-objects:
185 * it is created before them, and outlives its
186 * children (this later is necessary so that basic
187 * functions like cl_object_top() always
188 * work). Top-object keeps a reference on every
191 * When top-object is destroyed (lov_delete_raid0())
192 * it releases its reference to a sub-object and waits
193 * until the latter is finally destroyed.
195 struct lovsub_object **lo_sub;
199 spinlock_t lo_sub_lock;
201 * Cached object attribute, built from sub-object
204 struct cl_attr lo_attr;
207 struct lov_layout_dom {
208 /* keep this always at first place so DOM layout entry
209 * can be addressed also as RAID0 after initialization.
211 struct lov_layout_raid0 lo_dom_r0;
212 struct lovsub_object *lo_dom;
213 struct lov_oinfo *lo_loi;
216 struct lov_layout_entry {
218 unsigned int lle_valid:1;
219 unsigned int lle_preference;
220 struct lu_extent *lle_extent;
221 struct lov_stripe_md_entry *lle_lsme;
222 struct lov_comp_layout_entry_ops *lle_comp_ops;
224 struct lov_layout_raid0 lle_raid0;
225 struct lov_layout_dom lle_dom;
229 struct lov_mirror_entry {
230 unsigned short lre_mirror_id;
231 unsigned short lre_stale:1, /* set if any components is stale */
232 lre_valid:1, /* set if at least one of components
233 * in this mirror is valid */
234 lre_foreign:1; /* set if it is a foreign component */
235 int lre_preference; /* overall preference of this mirror */
237 unsigned short lre_start; /* index to lo_entries, start index of
239 unsigned short lre_end; /* end index of this mirror */
242 enum lov_object_flags {
243 /* Layout is invalid, set when layout lock is lost */
244 LO_LAYOUT_INVALID = 0x1,
248 * lov-specific file state.
250 * lov object has particular layout type, determining how top-object is built
251 * on top of sub-objects. Layout type can change dynamically. When this
252 * happens, lov_object::lo_type_guard semaphore is taken in exclusive mode,
253 * all state pertaining to the old layout type is destroyed, and new state is
254 * constructed. All object methods take said semaphore in the shared mode,
255 * providing serialization against transition between layout types.
257 * To avoid multiple `if' or `switch' statements, selecting behavior for the
258 * current layout type, object methods perform double-dispatch, invoking
259 * function corresponding to the current layout type.
262 struct cl_object lo_cl;
264 * Serializes object operations with transitions between layout types.
266 * This semaphore is taken in shared mode by all object methods, and
267 * is taken in exclusive mode when object type is changed.
269 * \see lov_object::lo_type
271 struct rw_semaphore lo_type_guard;
273 * Type of an object. Protected by lov_object::lo_type_guard.
275 enum lov_layout_type lo_type;
279 unsigned long lo_obj_flags;
281 * How many IOs are on going on this object. Layout can be changed
282 * only if there is no active IO.
284 atomic_t lo_active_ios;
286 * Waitq - wait for no one else is using lo_lsm
288 wait_queue_head_t lo_waitq;
290 * Layout metadata. NULL if empty layout.
292 struct lov_stripe_md *lo_lsm;
294 union lov_layout_state {
295 struct lov_layout_state_empty {
297 struct lov_layout_state_released {
299 struct lov_layout_composite {
301 * flags of lov_comp_md_v1::lcm_flags. Mainly used
306 * For FLR: index of preferred mirror to read.
307 * Preferred mirror is initialized by the preferred
308 * bit of lsme. It can be changed when the preferred
310 * In order to make lov_lsm_entry() return the same
311 * mirror in the same IO context, it's only possible
312 * to change the preferred mirror when the
313 * lo_active_ios reaches zero.
315 int lo_preferred_mirror;
317 * For FLR: Number of (valid) mirrors.
319 unsigned lo_mirror_count;
320 struct lov_mirror_entry *lo_mirrors;
322 * Current entry count of lo_entries, include
325 unsigned int lo_entry_count;
326 struct lov_layout_entry *lo_entries;
330 * Thread that acquired lov_object::lo_type_guard in an exclusive
333 struct task_struct *lo_owner;
336 static inline struct lov_layout_raid0 *lov_r0(struct lov_object *lov, int i)
338 LASSERT(lov->lo_type == LLT_COMP);
339 LASSERTF(i < lov->u.composite.lo_entry_count,
340 "entry %d entry_count %d\n", i,
341 lov->u.composite.lo_entry_count);
343 return &lov->u.composite.lo_entries[i].lle_raid0;
346 static inline struct lov_stripe_md_entry *lov_lse(struct lov_object *lov, int i)
348 LASSERT(lov->lo_lsm != NULL);
349 LASSERT(i < lov->lo_lsm->lsm_entry_count);
351 return lov->lo_lsm->lsm_entries[i];
354 static inline unsigned lov_flr_state(const struct lov_object *lov)
356 if (lov->lo_type != LLT_COMP)
359 return lov->u.composite.lo_flags & LCM_FL_FLR_MASK;
362 static inline bool lov_is_flr(const struct lov_object *lov)
364 return lov_flr_state(lov) != LCM_FL_NONE;
367 static inline struct lov_layout_entry *lov_entry(struct lov_object *lov, int i)
369 LASSERT(lov->lo_type == LLT_COMP);
370 LASSERTF(i < lov->u.composite.lo_entry_count,
371 "entry %d entry_count %d\n", i,
372 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];
441 struct lovsub_device {
442 struct cl_device acid_cl;
443 struct cl_device *acid_next;
446 struct lovsub_object {
447 struct cl_object_header lso_header;
448 struct cl_object lso_cl;
449 struct lov_object *lso_super;
454 * Describe the environment settings for sublocks.
456 struct lov_sublock_env {
457 const struct lu_env *lse_env;
458 struct cl_io *lse_io;
461 struct lov_thread_info {
462 struct cl_object_conf lti_stripe_conf;
463 struct lu_fid lti_fid;
464 struct ost_lvb lti_lvb;
465 struct cl_2queue lti_cl2q;
466 struct cl_page_list lti_plist;
470 * State that lov_io maintains for every sub-io.
474 * Linkage into a list (hanging off lov_io::lis_subios)
476 struct list_head sub_list;
478 * Linkage into a list (hanging off lov_io::lis_active) of all
479 * sub-io's active for the current IO iteration.
481 struct list_head sub_linkage;
482 unsigned int sub_subio_index;
484 * sub-io for a stripe. Ideally sub-io's can be stopped and resumed
485 * independently, with lov acting as a scheduler to maximize overall
490 * environment, in which sub-io executes.
492 struct lu_env *sub_env;
494 * environment's refcheck.
502 * IO state private for LOV.
504 #define LIS_CACHE_ENTRY_NONE -ENOENT
507 struct cl_io_slice lis_cl;
510 * FLR: index to lo_mirrors. Valid only if lov_is_flr() returns true.
512 * The mirror index of this io. Preserved over cl_io_init()
513 * if io->ci_ndelay_tried is greater than zero.
515 int lis_mirror_index;
517 * FLR: the layout gen when lis_mirror_index was cached. The
518 * mirror index makes sense only when the layout gen doesn't
521 int lis_mirror_layout_gen;
524 * fields below this will be initialized in lov_io_init().
526 unsigned lis_preserved;
529 * Pointer to the object slice. This is a duplicate of
530 * lov_io::lis_cl::cis_object.
532 struct lov_object *lis_object;
534 * Original end-of-io position for this IO, set by the upper layer as
535 * cl_io::u::ci_rw::pos + cl_io::u::ci_rw::count. lov remembers this,
536 * changes pos and count to fit IO into a single stripe and uses saved
537 * value to determine when IO iterations have to stop.
539 * This is used only for CIT_READ and CIT_WRITE io's.
541 loff_t lis_io_endpos;
544 * Record the stripe index before the truncate size, used for setting OST
545 * object size for truncate. LU-14128. lis_trunc_stripe_index[i] refers to
546 * lov_object.u.composite.lo_entries[i].
548 int *lis_trunc_stripe_index;
551 * starting position within a file, for the current io loop iteration
552 * (stripe), used by ci_io_loop().
556 * end position with in a file, for the current stripe io. This is
557 * exclusive (i.e., next offset after last byte affected by io).
563 * the index of ls_single_subio in ls_subios array
565 int lis_single_subio_index;
566 struct lov_io_sub lis_single_subio;
569 * List of active sub-io's. Active sub-io's are under the range
570 * of [lis_pos, lis_endpos).
572 struct list_head lis_active;
574 * All sub-io's created in this lov_io.
576 struct list_head lis_subios;
577 /* Cached results from stripe & offset calculations for page init */
578 int lis_cached_entry;
579 int lis_cached_stripe;
580 loff_t lis_cached_off;
581 loff_t lis_cached_suboff;
582 struct lov_io_sub *lis_cached_sub;
587 struct lov_sublock_env ls_subenv;
590 extern struct lu_device_type lov_device_type;
591 extern struct lu_device_type lovsub_device_type;
593 extern struct lu_context_key lov_key;
594 extern struct lu_context_key lov_session_key;
596 extern struct kmem_cache *lov_lock_kmem;
597 extern struct kmem_cache *lov_object_kmem;
598 extern struct kmem_cache *lov_thread_kmem;
599 extern struct kmem_cache *lov_session_kmem;
601 extern struct kmem_cache *lovsub_object_kmem;
603 int lov_lock_init_composite(const struct lu_env *env, struct cl_object *obj,
604 struct cl_lock *lock, const struct cl_io *io);
605 int lov_lock_init_empty (const struct lu_env *env, struct cl_object *obj,
606 struct cl_lock *lock, const struct cl_io *io);
607 int lov_io_init_composite(const struct lu_env *env, struct cl_object *obj,
609 int lov_io_init_empty (const struct lu_env *env, struct cl_object *obj,
611 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
614 struct lov_io_sub *lov_sub_get(const struct lu_env *env, struct lov_io *lio,
618 CP_LOV_INDEX_EMPTY = -1U,
621 static inline bool lov_page_is_empty(const struct cl_page *cp)
623 return cp->cp_lov_index == CP_LOV_INDEX_EMPTY;
626 int lov_page_init_empty (const struct lu_env *env, struct cl_object *obj,
627 struct cl_page *page, pgoff_t index);
628 int lov_page_init_composite(const struct lu_env *env, struct cl_object *obj,
629 struct cl_page *page, pgoff_t index);
630 int lov_page_init_foreign(const struct lu_env *env, struct cl_object *obj,
631 struct cl_page *page, pgoff_t index);
632 struct lu_object *lov_object_alloc (const struct lu_env *env,
633 const struct lu_object_header *hdr,
634 struct lu_device *dev);
636 struct lu_object *lovsub_object_alloc(const struct lu_env *env,
637 const struct lu_object_header *hdr,
638 struct lu_device *dev);
640 int lov_lsm_entry(const struct lov_stripe_md *lsm, __u64 offset);
641 int lov_io_layout_at(struct lov_io *lio, __u64 offset);
643 #define lov_foreach_target(lov, var) \
644 for (var = 0; var < lov_targets_nr(lov); ++var)
646 static inline struct lu_extent *lov_io_extent(struct lov_io *io, int i)
648 return &lov_lse(io->lis_object, i)->lsme_extent;
652 * For layout entries within @ext.
654 #define lov_foreach_io_layout(ind, lio, ext) \
655 for (ind = lov_io_layout_at(lio, (ext)->e_start); \
657 lu_extent_is_overlapped(lov_io_extent(lio, ind), ext); \
658 ind = lov_io_layout_at(lio, lov_io_extent(lio, ind)->e_end))
660 /*****************************************************************************
668 static inline struct lov_session *lov_env_session(const struct lu_env *env)
670 struct lov_session *ses;
672 ses = lu_context_key_get(env->le_ses, &lov_session_key);
673 LASSERT(ses != NULL);
677 static inline struct lov_io *lov_env_io(const struct lu_env *env)
679 return &lov_env_session(env)->ls_io;
682 static inline int lov_is_object(const struct lu_object *obj)
684 return obj->lo_dev->ld_type == &lov_device_type;
687 static inline int lovsub_is_object(const struct lu_object *obj)
689 return obj->lo_dev->ld_type == &lovsub_device_type;
692 static inline struct lu_device *lov2lu_dev(struct lov_device *lov)
694 return &lov->ld_cl.cd_lu_dev;
697 static inline struct lov_device *lu2lov_dev(const struct lu_device *d)
699 LINVRNT(d->ld_type == &lov_device_type);
700 return container_of(d, struct lov_device, ld_cl.cd_lu_dev);
703 static inline struct cl_device *lovsub2cl_dev(struct lovsub_device *lovsub)
705 return &lovsub->acid_cl;
708 static inline struct lu_device *lovsub2lu_dev(struct lovsub_device *lovsub)
710 return &lovsub2cl_dev(lovsub)->cd_lu_dev;
713 static inline struct lovsub_device *lu2lovsub_dev(const struct lu_device *d)
715 LINVRNT(d->ld_type == &lovsub_device_type);
716 return container_of(d, struct lovsub_device, acid_cl.cd_lu_dev);
719 static inline struct lovsub_device *cl2lovsub_dev(const struct cl_device *d)
721 LINVRNT(d->cd_lu_dev.ld_type == &lovsub_device_type);
722 return container_of(d, struct lovsub_device, acid_cl);
725 static inline struct lu_object *lov2lu(struct lov_object *lov)
727 return &lov->lo_cl.co_lu;
730 static inline struct cl_object *lov2cl(struct lov_object *lov)
735 static inline struct lov_object *lu2lov(const struct lu_object *obj)
737 LINVRNT(lov_is_object(obj));
738 return container_of(obj, struct lov_object, lo_cl.co_lu);
741 static inline struct lov_object *cl2lov(const struct cl_object *obj)
743 LINVRNT(lov_is_object(&obj->co_lu));
744 return container_of(obj, struct lov_object, lo_cl);
747 static inline struct lu_object *lovsub2lu(struct lovsub_object *los)
749 return &los->lso_cl.co_lu;
752 static inline struct cl_object *lovsub2cl(struct lovsub_object *los)
757 static inline struct lovsub_object *cl2lovsub(const struct cl_object *obj)
759 LINVRNT(lovsub_is_object(&obj->co_lu));
760 return container_of(obj, struct lovsub_object, lso_cl);
763 static inline struct lovsub_object *lu2lovsub(const struct lu_object *obj)
765 LINVRNT(lovsub_is_object(obj));
766 return container_of(obj, struct lovsub_object, lso_cl.co_lu);
769 static inline struct lov_lock *cl2lov_lock(const struct cl_lock_slice *slice)
771 LINVRNT(lov_is_object(&slice->cls_obj->co_lu));
772 return container_of(slice, struct lov_lock, lls_cl);
775 static inline struct lov_io *cl2lov_io(const struct lu_env *env,
776 const struct cl_io_slice *ios)
780 lio = container_of(ios, struct lov_io, lis_cl);
781 LASSERT(lio == lov_env_io(env));
785 static inline int lov_targets_nr(const struct lov_device *lov)
787 return lov->ld_lov->desc.ld_tgt_count;
790 static inline struct lov_thread_info *lov_env_info(const struct lu_env *env)
792 struct lov_thread_info *info;
794 info = lu_context_key_get(&env->le_ctx, &lov_key);
795 LASSERT(info != NULL);
800 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
801 struct lov_stripe_md *lsm, struct lov_user_md __user *lump,