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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef __LUSTRE_LU_OBJECT_H
34 #define __LUSTRE_LU_OBJECT_H
37 #include <linux/percpu_counter.h>
38 #include "../../include/linux/libcfs/libcfs.h"
39 #include "lustre/lustre_idl.h"
47 * lu_* data-types represent server-side entities shared by data and meta-data
52 * -# support for layering.
54 * Server side object is split into layers, one per device in the
55 * corresponding device stack. Individual layer is represented by struct
56 * lu_object. Compound layered object --- by struct lu_object_header. Most
57 * interface functions take lu_object as an argument and operate on the
58 * whole compound object. This decision was made due to the following
61 * - it's envisaged that lu_object will be used much more often than
64 * - we want lower (non-top) layers to be able to initiate operations
65 * on the whole object.
67 * Generic code supports layering more complex than simple stacking, e.g.,
68 * it is possible that at some layer object "spawns" multiple sub-objects
71 * -# fid-based identification.
73 * Compound object is uniquely identified by its fid. Objects are indexed
74 * by their fids (hash table is used for index).
76 * -# caching and life-cycle management.
78 * Object's life-time is controlled by reference counting. When reference
79 * count drops to 0, object is returned to cache. Cached objects still
80 * retain their identity (i.e., fid), and can be recovered from cache.
82 * Objects are kept in the global LRU list, and lu_site_purge() function
83 * can be used to reclaim given number of unused objects from the tail of
86 * -# avoiding recursion.
88 * Generic code tries to replace recursion through layers by iterations
89 * where possible. Additionally to the end of reducing stack consumption,
90 * data, when practically possible, are allocated through lu_context_key
91 * interface rather than on stack.
98 struct lu_object_header
;
103 * Operations common for data and meta-data devices.
105 struct lu_device_operations
{
107 * Allocate object for the given device (without lower-layer
108 * parts). This is called by lu_object_operations::loo_object_init()
109 * from the parent layer, and should setup at least lu_object::lo_dev
110 * and lu_object::lo_ops fields of resulting lu_object.
112 * Object creation protocol.
114 * Due to design goal of avoiding recursion, object creation (see
115 * lu_object_alloc()) is somewhat involved:
117 * - first, lu_device_operations::ldo_object_alloc() method of the
118 * top-level device in the stack is called. It should allocate top
119 * level object (including lu_object_header), but without any
120 * lower-layer sub-object(s).
122 * - then lu_object_alloc() sets fid in the header of newly created
125 * - then lu_object_operations::loo_object_init() is called. It has
126 * to allocate lower-layer object(s). To do this,
127 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
128 * of the lower-layer device(s).
130 * - for all new objects allocated by
131 * lu_object_operations::loo_object_init() (and inserted into object
132 * stack), lu_object_operations::loo_object_init() is called again
133 * repeatedly, until no new objects are created.
135 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
136 * result->lo_ops != NULL);
138 struct lu_object
*(*ldo_object_alloc
)(const struct lu_env
*env
,
139 const struct lu_object_header
*h
,
140 struct lu_device
*d
);
142 * process config specific for device.
144 int (*ldo_process_config
)(const struct lu_env
*env
,
145 struct lu_device
*, struct lustre_cfg
*);
146 int (*ldo_recovery_complete
)(const struct lu_env
*,
150 * initialize local objects for device. this method called after layer has
151 * been initialized (after LCFG_SETUP stage) and before it starts serving
155 int (*ldo_prepare
)(const struct lu_env
*,
156 struct lu_device
*parent
,
157 struct lu_device
*dev
);
162 * For lu_object_conf flags
165 /* This is a new object to be allocated, or the file
166 * corresponding to the object does not exists.
168 LOC_F_NEW
= 0x00000001,
172 * Object configuration, describing particulars of object being created. On
173 * server this is not used, as server objects are full identified by fid. On
174 * client configuration contains struct lustre_md.
176 struct lu_object_conf
{
178 * Some hints for obj find and alloc.
180 enum loc_flags loc_flags
;
184 * Type of "printer" function used by lu_object_operations::loo_object_print()
187 * Printer function is needed to provide some flexibility in (semi-)debugging
188 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
190 typedef int (*lu_printer_t
)(const struct lu_env
*env
,
191 void *cookie
, const char *format
, ...)
195 * Operations specific for particular lu_object.
197 struct lu_object_operations
{
199 * Allocate lower-layer parts of the object by calling
200 * lu_device_operations::ldo_object_alloc() of the corresponding
203 * This method is called once for each object inserted into object
204 * stack. It's responsibility of this method to insert lower-layer
205 * object(s) it create into appropriate places of object stack.
207 int (*loo_object_init
)(const struct lu_env
*env
,
209 const struct lu_object_conf
*conf
);
211 * Called (in top-to-bottom order) during object allocation after all
212 * layers were allocated and initialized. Can be used to perform
213 * initialization depending on lower layers.
215 int (*loo_object_start
)(const struct lu_env
*env
,
216 struct lu_object
*o
);
218 * Called before lu_object_operations::loo_object_free() to signal
219 * that object is being destroyed. Dual to
220 * lu_object_operations::loo_object_init().
222 void (*loo_object_delete
)(const struct lu_env
*env
,
223 struct lu_object
*o
);
225 * Dual to lu_device_operations::ldo_object_alloc(). Called when
226 * object is removed from memory.
228 void (*loo_object_free
)(const struct lu_env
*env
,
229 struct lu_object
*o
);
231 * Called when last active reference to the object is released (and
232 * object returns to the cache). This method is optional.
234 void (*loo_object_release
)(const struct lu_env
*env
,
235 struct lu_object
*o
);
237 * Optional debugging helper. Print given object.
239 int (*loo_object_print
)(const struct lu_env
*env
, void *cookie
,
240 lu_printer_t p
, const struct lu_object
*o
);
242 * Optional debugging method. Returns true iff method is internally
245 int (*loo_object_invariant
)(const struct lu_object
*o
);
251 struct lu_device_type
;
254 * Device: a layer in the server side abstraction stacking.
258 * reference count. This is incremented, in particular, on each object
259 * created at this layer.
261 * \todo XXX which means that atomic_t is probably too small.
265 * Pointer to device type. Never modified once set.
267 struct lu_device_type
*ld_type
;
269 * Operation vector for this device.
271 const struct lu_device_operations
*ld_ops
;
273 * Stack this device belongs to.
275 struct lu_site
*ld_site
;
277 /** \todo XXX: temporary back pointer into obd. */
278 struct obd_device
*ld_obd
;
280 * A list of references to this object, for debugging.
282 struct lu_ref ld_reference
;
284 * Link the device to the site.
286 struct list_head ld_linkage
;
289 struct lu_device_type_operations
;
292 * Tag bits for device type. They are used to distinguish certain groups of
296 /** this is meta-data device */
297 LU_DEVICE_MD
= (1 << 0),
298 /** this is data device */
299 LU_DEVICE_DT
= (1 << 1),
300 /** data device in the client stack */
301 LU_DEVICE_CL
= (1 << 2)
307 struct lu_device_type
{
309 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
313 * Name of this class. Unique system-wide. Never modified once set.
317 * Operations for this type.
319 const struct lu_device_type_operations
*ldt_ops
;
321 * \todo XXX: temporary pointer to associated obd_type.
323 struct obd_type
*ldt_obd_type
;
325 * \todo XXX: temporary: context tags used by obd_*() calls.
329 * Number of existing device type instances.
331 atomic_t ldt_device_nr
;
333 * Linkage into a global list of all device types.
335 * \see lu_device_types.
337 struct list_head ldt_linkage
;
341 * Operations on a device type.
343 struct lu_device_type_operations
{
345 * Allocate new device.
347 struct lu_device
*(*ldto_device_alloc
)(const struct lu_env
*env
,
348 struct lu_device_type
*t
,
349 struct lustre_cfg
*lcfg
);
351 * Free device. Dual to
352 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
353 * the next device in the stack.
355 struct lu_device
*(*ldto_device_free
)(const struct lu_env
*,
359 * Initialize the devices after allocation
361 int (*ldto_device_init
)(const struct lu_env
*env
,
362 struct lu_device
*, const char *,
365 * Finalize device. Dual to
366 * lu_device_type_operations::ldto_device_init(). Returns pointer to
367 * the next device in the stack.
369 struct lu_device
*(*ldto_device_fini
)(const struct lu_env
*env
,
372 * Initialize device type. This is called on module load.
374 int (*ldto_init
)(struct lu_device_type
*t
);
376 * Finalize device type. Dual to
377 * lu_device_type_operations::ldto_init(). Called on module unload.
379 void (*ldto_fini
)(struct lu_device_type
*t
);
381 * Called when the first device is created.
383 void (*ldto_start
)(struct lu_device_type
*t
);
385 * Called when number of devices drops to 0.
387 void (*ldto_stop
)(struct lu_device_type
*t
);
390 static inline int lu_device_is_md(const struct lu_device
*d
)
392 return ergo(d
, d
->ld_type
->ldt_tags
& LU_DEVICE_MD
);
396 * Common object attributes.
401 /** modification time in seconds since Epoch */
403 /** access time in seconds since Epoch */
405 /** change time in seconds since Epoch */
407 /** 512-byte blocks allocated to object */
409 /** permission bits and file type */
417 /** number of persistent references to this object */
419 /** blk bits of the object*/
421 /** blk size of the object*/
433 /** Bit-mask of valid attributes */
447 LA_BLKSIZE
= 1 << 12,
448 LA_KILL_SUID
= 1 << 13,
449 LA_KILL_SGID
= 1 << 14,
453 * Layer in the layered object.
457 * Header for this object.
459 struct lu_object_header
*lo_header
;
461 * Device for this layer.
463 struct lu_device
*lo_dev
;
465 * Operations for this object.
467 const struct lu_object_operations
*lo_ops
;
469 * Linkage into list of all layers.
471 struct list_head lo_linkage
;
473 * Link to the device, for debugging.
475 struct lu_ref_link lo_dev_ref
;
478 enum lu_object_header_flags
{
480 * Don't keep this object in cache. Object will be destroyed as soon
481 * as last reference to it is released. This flag cannot be cleared
484 LU_OBJECT_HEARD_BANSHEE
= 0,
486 * Mark this object has already been taken out of cache.
488 LU_OBJECT_UNHASHED
= 1,
491 enum lu_object_header_attr
{
492 LOHA_EXISTS
= 1 << 0,
493 LOHA_REMOTE
= 1 << 1,
495 * UNIX file type is stored in S_IFMT bits.
497 LOHA_FT_START
= 001 << 12, /**< S_IFIFO */
498 LOHA_FT_END
= 017 << 12, /**< S_IFMT */
502 * "Compound" object, consisting of multiple layers.
504 * Compound object with given fid is unique with given lu_site.
506 * Note, that object does *not* necessary correspond to the real object in the
507 * persistent storage: object is an anchor for locking and method calling, so
508 * it is created for things like not-yet-existing child created by mkdir or
509 * create calls. lu_object_operations::loo_exists() can be used to check
510 * whether object is backed by persistent storage entity.
512 struct lu_object_header
{
514 * Fid, uniquely identifying this object.
516 struct lu_fid loh_fid
;
518 * Object flags from enum lu_object_header_flags. Set and checked
521 unsigned long loh_flags
;
523 * Object reference count. Protected by lu_site::ls_guard.
527 * Common object attributes, cached for efficiency. From enum
528 * lu_object_header_attr.
532 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
534 struct hlist_node loh_hash
;
536 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
538 struct list_head loh_lru
;
540 * Linkage into list of layers. Never modified once set (except lately
541 * during object destruction). No locking is necessary.
543 struct list_head loh_layers
;
545 * A list of references to this object, for debugging.
547 struct lu_ref loh_reference
;
552 struct lu_site_bkt_data
{
554 * number of object in this bucket on the lsb_lru list.
558 * LRU list, updated on each access to object. Protected by
559 * bucket lock of lu_site::ls_obj_hash.
561 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
562 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
563 * of list_for_each_entry_safe_reverse()).
565 struct list_head lsb_lru
;
567 * Wait-queue signaled when an object in this site is ultimately
568 * destroyed (lu_object_free()). It is used by lu_object_find() to
569 * wait before re-trying when object in the process of destruction is
570 * found in the hash table.
572 * \see htable_lookup().
574 wait_queue_head_t lsb_marche_funebre
;
582 LU_SS_CACHE_DEATH_RACE
,
588 * lu_site is a "compartment" within which objects are unique, and LRU
589 * discipline is maintained.
591 * lu_site exists so that multiple layered stacks can co-exist in the same
594 * lu_site has the same relation to lu_device as lu_object_header to
601 struct cfs_hash
*ls_obj_hash
;
603 * index of bucket on hash table while purging
605 unsigned int ls_purge_start
;
607 * Top-level device for this stack.
609 struct lu_device
*ls_top_dev
;
611 * Bottom-level device for this stack
613 struct lu_device
*ls_bottom_dev
;
615 * Linkage into global list of sites.
617 struct list_head ls_linkage
;
619 * List for lu device for this site, protected
622 struct list_head ls_ld_linkage
;
623 spinlock_t ls_ld_lock
;
626 * Lock to serialize site purge.
628 struct mutex ls_purge_mutex
;
633 struct lprocfs_stats
*ls_stats
;
635 * XXX: a hack! fld has to find md_site via site, remove when possible
637 struct seq_server_site
*ld_seq_site
;
639 * Number of objects in lsb_lru_lists - used for shrinking
641 struct percpu_counter ls_lru_len_counter
;
644 static inline struct lu_site_bkt_data
*
645 lu_site_bkt_from_fid(struct lu_site
*site
, struct lu_fid
*fid
)
647 struct cfs_hash_bd bd
;
649 cfs_hash_bd_get(site
->ls_obj_hash
, fid
, &bd
);
650 return cfs_hash_bd_extra_get(site
->ls_obj_hash
, &bd
);
653 static inline struct seq_server_site
*lu_site2seq(const struct lu_site
*s
)
655 return s
->ld_seq_site
;
659 * Constructors/destructors.
663 int lu_site_init(struct lu_site
*s
, struct lu_device
*d
);
664 void lu_site_fini(struct lu_site
*s
);
665 int lu_site_init_finish(struct lu_site
*s
);
666 void lu_stack_fini(const struct lu_env
*env
, struct lu_device
*top
);
667 void lu_device_get(struct lu_device
*d
);
668 void lu_device_put(struct lu_device
*d
);
669 int lu_device_init(struct lu_device
*d
, struct lu_device_type
*t
);
670 void lu_device_fini(struct lu_device
*d
);
671 int lu_object_header_init(struct lu_object_header
*h
);
672 void lu_object_header_fini(struct lu_object_header
*h
);
673 int lu_object_init(struct lu_object
*o
,
674 struct lu_object_header
*h
, struct lu_device
*d
);
675 void lu_object_fini(struct lu_object
*o
);
676 void lu_object_add_top(struct lu_object_header
*h
, struct lu_object
*o
);
677 void lu_object_add(struct lu_object
*before
, struct lu_object
*o
);
680 * Helpers to initialize and finalize device types.
683 int lu_device_type_init(struct lu_device_type
*ldt
);
684 void lu_device_type_fini(struct lu_device_type
*ldt
);
689 * Caching and reference counting.
694 * Acquire additional reference to the given object. This function is used to
695 * attain additional reference. To acquire initial reference use
698 static inline void lu_object_get(struct lu_object
*o
)
700 LASSERT(atomic_read(&o
->lo_header
->loh_ref
) > 0);
701 atomic_inc(&o
->lo_header
->loh_ref
);
705 * Return true of object will not be cached after last reference to it is
708 static inline int lu_object_is_dying(const struct lu_object_header
*h
)
710 return test_bit(LU_OBJECT_HEARD_BANSHEE
, &h
->loh_flags
);
713 void lu_object_put(const struct lu_env
*env
, struct lu_object
*o
);
714 void lu_object_unhash(const struct lu_env
*env
, struct lu_object
*o
);
715 int lu_site_purge_objects(const struct lu_env
*env
, struct lu_site
*s
, int nr
,
718 static inline int lu_site_purge(const struct lu_env
*env
, struct lu_site
*s
,
721 return lu_site_purge_objects(env
, s
, nr
, true);
724 void lu_site_print(const struct lu_env
*env
, struct lu_site
*s
, void *cookie
,
725 lu_printer_t printer
);
726 struct lu_object
*lu_object_find_at(const struct lu_env
*env
,
727 struct lu_device
*dev
,
728 const struct lu_fid
*f
,
729 const struct lu_object_conf
*conf
);
730 struct lu_object
*lu_object_find_slice(const struct lu_env
*env
,
731 struct lu_device
*dev
,
732 const struct lu_fid
*f
,
733 const struct lu_object_conf
*conf
);
742 * First (topmost) sub-object of given compound object
744 static inline struct lu_object
*lu_object_top(struct lu_object_header
*h
)
746 LASSERT(!list_empty(&h
->loh_layers
));
747 return container_of0(h
->loh_layers
.next
, struct lu_object
, lo_linkage
);
751 * Next sub-object in the layering
753 static inline struct lu_object
*lu_object_next(const struct lu_object
*o
)
755 return container_of0(o
->lo_linkage
.next
, struct lu_object
, lo_linkage
);
759 * Pointer to the fid of this object.
761 static inline const struct lu_fid
*lu_object_fid(const struct lu_object
*o
)
763 return &o
->lo_header
->loh_fid
;
767 * return device operations vector for this object
769 static inline const struct lu_device_operations
*
770 lu_object_ops(const struct lu_object
*o
)
772 return o
->lo_dev
->ld_ops
;
776 * Given a compound object, find its slice, corresponding to the device type
779 struct lu_object
*lu_object_locate(struct lu_object_header
*h
,
780 const struct lu_device_type
*dtype
);
783 * Printer function emitting messages through libcfs_debug_msg().
785 int lu_cdebug_printer(const struct lu_env
*env
,
786 void *cookie
, const char *format
, ...);
789 * Print object description followed by a user-supplied message.
791 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
793 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
794 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
795 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
796 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
801 * Print short object description followed by a user-supplied message.
803 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
805 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
806 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
807 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
808 (object)->lo_header); \
809 lu_cdebug_printer(env, &msgdata, "\n"); \
810 CDEBUG(mask, format, ## __VA_ARGS__); \
814 void lu_object_print (const struct lu_env
*env
, void *cookie
,
815 lu_printer_t printer
, const struct lu_object
*o
);
816 void lu_object_header_print(const struct lu_env
*env
, void *cookie
,
817 lu_printer_t printer
,
818 const struct lu_object_header
*hdr
);
821 * Check object consistency.
823 int lu_object_invariant(const struct lu_object
*o
);
826 * Check whether object exists, no matter on local or remote storage.
827 * Note: LOHA_EXISTS will be set once some one created the object,
828 * and it does not needs to be committed to storage.
830 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
833 * Check whether object on the remote storage.
835 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
837 static inline int lu_object_assert_exists(const struct lu_object
*o
)
839 return lu_object_exists(o
);
842 static inline int lu_object_assert_not_exists(const struct lu_object
*o
)
844 return !lu_object_exists(o
);
848 * Attr of this object.
850 static inline __u32
lu_object_attr(const struct lu_object
*o
)
852 LASSERT(lu_object_exists(o
) != 0);
853 return o
->lo_header
->loh_attr
;
856 static inline void lu_object_ref_add(struct lu_object
*o
,
860 lu_ref_add(&o
->lo_header
->loh_reference
, scope
, source
);
863 static inline void lu_object_ref_add_at(struct lu_object
*o
,
864 struct lu_ref_link
*link
,
868 lu_ref_add_at(&o
->lo_header
->loh_reference
, link
, scope
, source
);
871 static inline void lu_object_ref_del(struct lu_object
*o
,
872 const char *scope
, const void *source
)
874 lu_ref_del(&o
->lo_header
->loh_reference
, scope
, source
);
877 static inline void lu_object_ref_del_at(struct lu_object
*o
,
878 struct lu_ref_link
*link
,
879 const char *scope
, const void *source
)
881 lu_ref_del_at(&o
->lo_header
->loh_reference
, link
, scope
, source
);
884 /** input params, should be filled out by mdt */
888 /** count in bytes */
889 unsigned int rp_count
;
890 /** number of pages */
891 unsigned int rp_npages
;
892 /** requested attr */
894 /** pointers to pages */
895 struct page
**rp_pages
;
898 enum lu_xattr_flags
{
899 LU_XATTR_REPLACE
= (1 << 0),
900 LU_XATTR_CREATE
= (1 << 1)
909 /** For lu_context health-checks */
910 enum lu_context_state
{
918 * lu_context. Execution context for lu_object methods. Currently associated
921 * All lu_object methods, except device and device type methods (called during
922 * system initialization and shutdown) are executed "within" some
923 * lu_context. This means, that pointer to some "current" lu_context is passed
924 * as an argument to all methods.
926 * All service ptlrpc threads create lu_context as part of their
927 * initialization. It is possible to create "stand-alone" context for other
928 * execution environments (like system calls).
930 * lu_object methods mainly use lu_context through lu_context_key interface
931 * that allows each layer to associate arbitrary pieces of data with each
932 * context (see pthread_key_create(3) for similar interface).
934 * On a client, lu_context is bound to a thread, see cl_env_get().
936 * \see lu_context_key
940 * lu_context is used on the client side too. Yet we don't want to
941 * allocate values of server-side keys for the client contexts and
944 * To achieve this, set of tags in introduced. Contexts and keys are
945 * marked with tags. Key value are created only for context whose set
946 * of tags has non-empty intersection with one for key. Tags are taken
947 * from enum lu_context_tag.
950 enum lu_context_state lc_state
;
952 * Pointer to the home service thread. NULL for other execution
955 struct ptlrpc_thread
*lc_thread
;
957 * Pointer to an array with key values. Internal implementation
962 * Linkage into a list of all remembered contexts. Only
963 * `non-transient' contexts, i.e., ones created for service threads
966 struct list_head lc_remember
;
968 * Version counter used to skip calls to lu_context_refill() when no
969 * keys were registered.
979 * lu_context_key interface. Similar to pthread_key.
982 enum lu_context_tag
{
984 * Thread on md server
986 LCT_MD_THREAD
= 1 << 0,
988 * Thread on dt server
990 LCT_DT_THREAD
= 1 << 1,
992 * Context for transaction handle
994 LCT_TX_HANDLE
= 1 << 2,
998 LCT_CL_THREAD
= 1 << 3,
1000 * A per-request session on a server, and a per-system-call session on
1003 LCT_SESSION
= 1 << 4,
1005 * A per-request data on OSP device
1007 LCT_OSP_THREAD
= 1 << 5,
1011 LCT_MG_THREAD
= 1 << 6,
1013 * Context for local operations
1017 * session for server thread
1019 LCT_SERVER_SESSION
= BIT(8),
1021 * Set when at least one of keys, having values in this context has
1022 * non-NULL lu_context_key::lct_exit() method. This is used to
1023 * optimize lu_context_exit() call.
1025 LCT_HAS_EXIT
= 1 << 28,
1027 * Don't add references for modules creating key values in that context.
1028 * This is only for contexts used internally by lu_object framework.
1030 LCT_NOREF
= 1 << 29,
1032 * Key is being prepared for retiring, don't create new values for it.
1034 LCT_QUIESCENT
= 1 << 30,
1036 * Context should be remembered.
1038 LCT_REMEMBER
= 1 << 31,
1040 * Contexts usable in cache shrinker thread.
1042 LCT_SHRINKER
= LCT_MD_THREAD
| LCT_DT_THREAD
| LCT_CL_THREAD
|
1047 * Key. Represents per-context value slot.
1049 * Keys are usually registered when module owning the key is initialized, and
1050 * de-registered when module is unloaded. Once key is registered, all new
1051 * contexts with matching tags, will get key value. "Old" contexts, already
1052 * initialized at the time of key registration, can be forced to get key value
1053 * by calling lu_context_refill().
1055 * Every key value is counted in lu_context_key::lct_used and acquires a
1056 * reference on an owning module. This means, that all key values have to be
1057 * destroyed before module can be unloaded. This is usually achieved by
1058 * stopping threads started by the module, that created contexts in their
1059 * entry functions. Situation is complicated by the threads shared by multiple
1060 * modules, like ptlrpcd daemon on a client. To work around this problem,
1061 * contexts, created in such threads, are `remembered' (see
1062 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1063 * for unloading it does the following:
1065 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1066 * preventing new key values from being allocated in the new contexts,
1069 * - scans a list of remembered contexts, destroying values of module
1070 * keys, thus releasing references to the module.
1072 * This is done by lu_context_key_quiesce(). If module is re-activated
1073 * before key has been de-registered, lu_context_key_revive() call clears
1074 * `quiescent' marker.
1076 * lu_context code doesn't provide any internal synchronization for these
1077 * activities---it's assumed that startup (including threads start-up) and
1078 * shutdown are serialized by some external means.
1082 struct lu_context_key
{
1084 * Set of tags for which values of this key are to be instantiated.
1088 * Value constructor. This is called when new value is created for a
1089 * context. Returns pointer to new value of error pointer.
1091 void *(*lct_init
)(const struct lu_context
*ctx
,
1092 struct lu_context_key
*key
);
1094 * Value destructor. Called when context with previously allocated
1095 * value of this slot is destroyed. \a data is a value that was returned
1096 * by a matching call to lu_context_key::lct_init().
1098 void (*lct_fini
)(const struct lu_context
*ctx
,
1099 struct lu_context_key
*key
, void *data
);
1101 * Optional method called on lu_context_exit() for all allocated
1102 * keys. Can be used by debugging code checking that locks are
1105 void (*lct_exit
)(const struct lu_context
*ctx
,
1106 struct lu_context_key
*key
, void *data
);
1108 * Internal implementation detail: index within lu_context::lc_value[]
1109 * reserved for this key.
1113 * Internal implementation detail: number of values created for this
1118 * Internal implementation detail: module for this key.
1120 struct module
*lct_owner
;
1122 * References to this key. For debugging.
1124 struct lu_ref lct_reference
;
1127 #define LU_KEY_INIT(mod, type) \
1128 static void *mod##_key_init(const struct lu_context *ctx, \
1129 struct lu_context_key *key) \
1133 BUILD_BUG_ON(PAGE_SIZE < sizeof(*value)); \
1135 value = kzalloc(sizeof(*value), GFP_NOFS); \
1137 value = ERR_PTR(-ENOMEM); \
1141 struct __##mod##__dummy_init {; } /* semicolon catcher */
1143 #define LU_KEY_FINI(mod, type) \
1144 static void mod##_key_fini(const struct lu_context *ctx, \
1145 struct lu_context_key *key, void *data) \
1147 type *info = data; \
1151 struct __##mod##__dummy_fini {; } /* semicolon catcher */
1153 #define LU_KEY_INIT_FINI(mod, type) \
1154 LU_KEY_INIT(mod, type); \
1155 LU_KEY_FINI(mod, type)
1157 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1158 struct lu_context_key mod##_thread_key = { \
1160 .lct_init = mod##_key_init, \
1161 .lct_fini = mod##_key_fini \
1164 #define LU_CONTEXT_KEY_INIT(key) \
1166 (key)->lct_owner = THIS_MODULE; \
1169 int lu_context_key_register(struct lu_context_key
*key
);
1170 void lu_context_key_degister(struct lu_context_key
*key
);
1171 void *lu_context_key_get(const struct lu_context
*ctx
,
1172 const struct lu_context_key
*key
);
1173 void lu_context_key_quiesce(struct lu_context_key
*key
);
1174 void lu_context_key_revive(struct lu_context_key
*key
);
1177 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1181 #define LU_KEY_INIT_GENERIC(mod) \
1182 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1184 struct lu_context_key *key = k; \
1187 va_start(args, k); \
1189 LU_CONTEXT_KEY_INIT(key); \
1190 key = va_arg(args, struct lu_context_key *); \
1195 #define LU_TYPE_INIT(mod, ...) \
1196 LU_KEY_INIT_GENERIC(mod) \
1197 static int mod##_type_init(struct lu_device_type *t) \
1199 mod##_key_init_generic(__VA_ARGS__, NULL); \
1200 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1202 struct __##mod##_dummy_type_init {; }
1204 #define LU_TYPE_FINI(mod, ...) \
1205 static void mod##_type_fini(struct lu_device_type *t) \
1207 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1209 struct __##mod##_dummy_type_fini {; }
1211 #define LU_TYPE_START(mod, ...) \
1212 static void mod##_type_start(struct lu_device_type *t) \
1214 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1216 struct __##mod##_dummy_type_start {; }
1218 #define LU_TYPE_STOP(mod, ...) \
1219 static void mod##_type_stop(struct lu_device_type *t) \
1221 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1223 struct __##mod##_dummy_type_stop {; }
1225 #define LU_TYPE_INIT_FINI(mod, ...) \
1226 LU_TYPE_INIT(mod, __VA_ARGS__); \
1227 LU_TYPE_FINI(mod, __VA_ARGS__); \
1228 LU_TYPE_START(mod, __VA_ARGS__); \
1229 LU_TYPE_STOP(mod, __VA_ARGS__)
1231 int lu_context_init(struct lu_context
*ctx
, __u32 tags
);
1232 void lu_context_fini(struct lu_context
*ctx
);
1233 void lu_context_enter(struct lu_context
*ctx
);
1234 void lu_context_exit(struct lu_context
*ctx
);
1235 int lu_context_refill(struct lu_context
*ctx
);
1238 * Helper functions to operate on multiple keys. These are used by the default
1239 * device type operations, defined by LU_TYPE_INIT_FINI().
1242 int lu_context_key_register_many(struct lu_context_key
*k
, ...);
1243 void lu_context_key_degister_many(struct lu_context_key
*k
, ...);
1244 void lu_context_key_revive_many(struct lu_context_key
*k
, ...);
1245 void lu_context_key_quiesce_many(struct lu_context_key
*k
, ...);
1252 * "Local" context, used to store data instead of stack.
1254 struct lu_context le_ctx
;
1256 * "Session" context for per-request data.
1258 struct lu_context
*le_ses
;
1261 int lu_env_init(struct lu_env
*env
, __u32 tags
);
1262 void lu_env_fini(struct lu_env
*env
);
1263 int lu_env_refill(struct lu_env
*env
);
1265 /** @} lu_context */
1268 * Output site statistical counters into a buffer. Suitable for
1269 * ll_rd_*()-style functions.
1271 int lu_site_stats_print(const struct lu_site
*s
, struct seq_file
*m
);
1274 * Common name structure to be passed around for various name related methods.
1277 const char *ln_name
;
1282 * Validate names (path components)
1284 * To be valid \a name must be non-empty, '\0' terminated of length \a
1285 * name_len, and not contain '/'. The maximum length of a name (before
1286 * say -ENAMETOOLONG will be returned) is really controlled by llite
1287 * and the server. We only check for something insane coming from bad
1288 * integer handling here.
1290 static inline bool lu_name_is_valid_2(const char *name
, size_t name_len
)
1292 return name
&& name_len
> 0 && name_len
< INT_MAX
&&
1293 name
[name_len
] == '\0' && strlen(name
) == name_len
&&
1294 !memchr(name
, '/', name_len
);
1298 * Common buffer structure to be passed around for various xattr_{s,g}et()
1306 #define DLUBUF "(%p %zu)"
1307 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1309 * One-time initializers, called at obdclass module initialization, not
1314 * Initialization of global lu_* data.
1316 int lu_global_init(void);
1319 * Dual to lu_global_init().
1321 void lu_global_fini(void);
1323 struct lu_kmem_descr
{
1324 struct kmem_cache
**ckd_cache
;
1325 const char *ckd_name
;
1326 const size_t ckd_size
;
1329 int lu_kmem_init(struct lu_kmem_descr
*caches
);
1330 void lu_kmem_fini(struct lu_kmem_descr
*caches
);
1332 void lu_buf_free(struct lu_buf
*buf
);
1333 void lu_buf_alloc(struct lu_buf
*buf
, size_t size
);
1334 void lu_buf_realloc(struct lu_buf
*buf
, size_t size
);
1336 int lu_buf_check_and_grow(struct lu_buf
*buf
, size_t len
);
1337 struct lu_buf
*lu_buf_check_and_alloc(struct lu_buf
*buf
, size_t len
);
1339 extern __u32 lu_context_tags_default
;
1340 extern __u32 lu_session_tags_default
;
1343 #endif /* __LUSTRE_LU_OBJECT_H */