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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef __LUSTRE_LU_OBJECT_H
38 #define __LUSTRE_LU_OBJECT_H
41 #include "../../include/linux/libcfs/libcfs.h"
42 #include "lustre/lustre_idl.h"
50 * lu_* data-types represent server-side entities shared by data and meta-data
55 * -# support for layering.
57 * Server side object is split into layers, one per device in the
58 * corresponding device stack. Individual layer is represented by struct
59 * lu_object. Compound layered object --- by struct lu_object_header. Most
60 * interface functions take lu_object as an argument and operate on the
61 * whole compound object. This decision was made due to the following
64 * - it's envisaged that lu_object will be used much more often than
67 * - we want lower (non-top) layers to be able to initiate operations
68 * on the whole object.
70 * Generic code supports layering more complex than simple stacking, e.g.,
71 * it is possible that at some layer object "spawns" multiple sub-objects
74 * -# fid-based identification.
76 * Compound object is uniquely identified by its fid. Objects are indexed
77 * by their fids (hash table is used for index).
79 * -# caching and life-cycle management.
81 * Object's life-time is controlled by reference counting. When reference
82 * count drops to 0, object is returned to cache. Cached objects still
83 * retain their identity (i.e., fid), and can be recovered from cache.
85 * Objects are kept in the global LRU list, and lu_site_purge() function
86 * can be used to reclaim given number of unused objects from the tail of
89 * -# avoiding recursion.
91 * Generic code tries to replace recursion through layers by iterations
92 * where possible. Additionally to the end of reducing stack consumption,
93 * data, when practically possible, are allocated through lu_context_key
94 * interface rather than on stack.
101 struct lu_object_header;
106 * Operations common for data and meta-data devices.
108 struct lu_device_operations {
110 * Allocate object for the given device (without lower-layer
111 * parts). This is called by lu_object_operations::loo_object_init()
112 * from the parent layer, and should setup at least lu_object::lo_dev
113 * and lu_object::lo_ops fields of resulting lu_object.
115 * Object creation protocol.
117 * Due to design goal of avoiding recursion, object creation (see
118 * lu_object_alloc()) is somewhat involved:
120 * - first, lu_device_operations::ldo_object_alloc() method of the
121 * top-level device in the stack is called. It should allocate top
122 * level object (including lu_object_header), but without any
123 * lower-layer sub-object(s).
125 * - then lu_object_alloc() sets fid in the header of newly created
128 * - then lu_object_operations::loo_object_init() is called. It has
129 * to allocate lower-layer object(s). To do this,
130 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
131 * of the lower-layer device(s).
133 * - for all new objects allocated by
134 * lu_object_operations::loo_object_init() (and inserted into object
135 * stack), lu_object_operations::loo_object_init() is called again
136 * repeatedly, until no new objects are created.
138 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
139 * result->lo_ops != NULL);
141 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
142 const struct lu_object_header *h,
143 struct lu_device *d);
145 * process config specific for device.
147 int (*ldo_process_config)(const struct lu_env *env,
148 struct lu_device *, struct lustre_cfg *);
149 int (*ldo_recovery_complete)(const struct lu_env *,
153 * initialize local objects for device. this method called after layer has
154 * been initialized (after LCFG_SETUP stage) and before it starts serving
158 int (*ldo_prepare)(const struct lu_env *,
159 struct lu_device *parent,
160 struct lu_device *dev);
165 * For lu_object_conf flags
168 /* This is a new object to be allocated, or the file
169 * corresponding to the object does not exists.
171 LOC_F_NEW = 0x00000001,
175 * Object configuration, describing particulars of object being created. On
176 * server this is not used, as server objects are full identified by fid. On
177 * client configuration contains struct lustre_md.
179 struct lu_object_conf {
181 * Some hints for obj find and alloc.
183 enum loc_flags loc_flags;
187 * Type of "printer" function used by lu_object_operations::loo_object_print()
190 * Printer function is needed to provide some flexibility in (semi-)debugging
191 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
193 typedef int (*lu_printer_t)(const struct lu_env *env,
194 void *cookie, const char *format, ...)
198 * Operations specific for particular lu_object.
200 struct lu_object_operations {
203 * Allocate lower-layer parts of the object by calling
204 * lu_device_operations::ldo_object_alloc() of the corresponding
207 * This method is called once for each object inserted into object
208 * stack. It's responsibility of this method to insert lower-layer
209 * object(s) it create into appropriate places of object stack.
211 int (*loo_object_init)(const struct lu_env *env,
213 const struct lu_object_conf *conf);
215 * Called (in top-to-bottom order) during object allocation after all
216 * layers were allocated and initialized. Can be used to perform
217 * initialization depending on lower layers.
219 int (*loo_object_start)(const struct lu_env *env,
220 struct lu_object *o);
222 * Called before lu_object_operations::loo_object_free() to signal
223 * that object is being destroyed. Dual to
224 * lu_object_operations::loo_object_init().
226 void (*loo_object_delete)(const struct lu_env *env,
227 struct lu_object *o);
229 * Dual to lu_device_operations::ldo_object_alloc(). Called when
230 * object is removed from memory.
232 void (*loo_object_free)(const struct lu_env *env,
233 struct lu_object *o);
235 * Called when last active reference to the object is released (and
236 * object returns to the cache). This method is optional.
238 void (*loo_object_release)(const struct lu_env *env,
239 struct lu_object *o);
241 * Optional debugging helper. Print given object.
243 int (*loo_object_print)(const struct lu_env *env, void *cookie,
244 lu_printer_t p, const struct lu_object *o);
246 * Optional debugging method. Returns true iff method is internally
249 int (*loo_object_invariant)(const struct lu_object *o);
255 struct lu_device_type;
258 * Device: a layer in the server side abstraction stacking.
262 * reference count. This is incremented, in particular, on each object
263 * created at this layer.
265 * \todo XXX which means that atomic_t is probably too small.
269 * Pointer to device type. Never modified once set.
271 struct lu_device_type *ld_type;
273 * Operation vector for this device.
275 const struct lu_device_operations *ld_ops;
277 * Stack this device belongs to.
279 struct lu_site *ld_site;
281 /** \todo XXX: temporary back pointer into obd. */
282 struct obd_device *ld_obd;
284 * A list of references to this object, for debugging.
286 struct lu_ref ld_reference;
288 * Link the device to the site.
290 struct list_head ld_linkage;
293 struct lu_device_type_operations;
296 * Tag bits for device type. They are used to distinguish certain groups of
300 /** this is meta-data device */
301 LU_DEVICE_MD = (1 << 0),
302 /** this is data device */
303 LU_DEVICE_DT = (1 << 1),
304 /** data device in the client stack */
305 LU_DEVICE_CL = (1 << 2)
311 struct lu_device_type {
313 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
317 * Name of this class. Unique system-wide. Never modified once set.
321 * Operations for this type.
323 const struct lu_device_type_operations *ldt_ops;
325 * \todo XXX: temporary pointer to associated obd_type.
327 struct obd_type *ldt_obd_type;
329 * \todo XXX: temporary: context tags used by obd_*() calls.
333 * Number of existing device type instances.
335 unsigned ldt_device_nr;
337 * Linkage into a global list of all device types.
339 * \see lu_device_types.
341 struct list_head ldt_linkage;
345 * Operations on a device type.
347 struct lu_device_type_operations {
349 * Allocate new device.
351 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
352 struct lu_device_type *t,
353 struct lustre_cfg *lcfg);
355 * Free device. Dual to
356 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
357 * the next device in the stack.
359 struct lu_device *(*ldto_device_free)(const struct lu_env *,
363 * Initialize the devices after allocation
365 int (*ldto_device_init)(const struct lu_env *env,
366 struct lu_device *, const char *,
369 * Finalize device. Dual to
370 * lu_device_type_operations::ldto_device_init(). Returns pointer to
371 * the next device in the stack.
373 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
376 * Initialize device type. This is called on module load.
378 int (*ldto_init)(struct lu_device_type *t);
380 * Finalize device type. Dual to
381 * lu_device_type_operations::ldto_init(). Called on module unload.
383 void (*ldto_fini)(struct lu_device_type *t);
385 * Called when the first device is created.
387 void (*ldto_start)(struct lu_device_type *t);
389 * Called when number of devices drops to 0.
391 void (*ldto_stop)(struct lu_device_type *t);
394 static inline int lu_device_is_md(const struct lu_device *d)
396 return ergo(d, d->ld_type->ldt_tags & LU_DEVICE_MD);
400 * Common object attributes.
405 /** modification time in seconds since Epoch */
407 /** access time in seconds since Epoch */
409 /** change time in seconds since Epoch */
411 /** 512-byte blocks allocated to object */
413 /** permission bits and file type */
421 /** number of persistent references to this object */
423 /** blk bits of the object*/
425 /** blk size of the object*/
437 /** Bit-mask of valid attributes */
451 LA_BLKSIZE = 1 << 12,
452 LA_KILL_SUID = 1 << 13,
453 LA_KILL_SGID = 1 << 14,
457 * Layer in the layered object.
461 * Header for this object.
463 struct lu_object_header *lo_header;
465 * Device for this layer.
467 struct lu_device *lo_dev;
469 * Operations for this object.
471 const struct lu_object_operations *lo_ops;
473 * Linkage into list of all layers.
475 struct list_head lo_linkage;
477 * Link to the device, for debugging.
479 struct lu_ref_link lo_dev_ref;
482 enum lu_object_header_flags {
484 * Don't keep this object in cache. Object will be destroyed as soon
485 * as last reference to it is released. This flag cannot be cleared
488 LU_OBJECT_HEARD_BANSHEE = 0,
490 * Mark this object has already been taken out of cache.
492 LU_OBJECT_UNHASHED = 1,
495 enum lu_object_header_attr {
496 LOHA_EXISTS = 1 << 0,
497 LOHA_REMOTE = 1 << 1,
499 * UNIX file type is stored in S_IFMT bits.
501 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
502 LOHA_FT_END = 017 << 12, /**< S_IFMT */
506 * "Compound" object, consisting of multiple layers.
508 * Compound object with given fid is unique with given lu_site.
510 * Note, that object does *not* necessary correspond to the real object in the
511 * persistent storage: object is an anchor for locking and method calling, so
512 * it is created for things like not-yet-existing child created by mkdir or
513 * create calls. lu_object_operations::loo_exists() can be used to check
514 * whether object is backed by persistent storage entity.
516 struct lu_object_header {
518 * Fid, uniquely identifying this object.
520 struct lu_fid loh_fid;
522 * Object flags from enum lu_object_header_flags. Set and checked
525 unsigned long loh_flags;
527 * Object reference count. Protected by lu_site::ls_guard.
531 * Common object attributes, cached for efficiency. From enum
532 * lu_object_header_attr.
536 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
538 struct hlist_node loh_hash;
540 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
542 struct list_head loh_lru;
544 * Linkage into list of layers. Never modified once set (except lately
545 * during object destruction). No locking is necessary.
547 struct list_head loh_layers;
549 * A list of references to this object, for debugging.
551 struct lu_ref loh_reference;
556 struct lu_site_bkt_data {
558 * number of object in this bucket on the lsb_lru list.
562 * LRU list, updated on each access to object. Protected by
563 * bucket lock of lu_site::ls_obj_hash.
565 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
566 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
567 * of list_for_each_entry_safe_reverse()).
569 struct list_head lsb_lru;
571 * Wait-queue signaled when an object in this site is ultimately
572 * destroyed (lu_object_free()). It is used by lu_object_find() to
573 * wait before re-trying when object in the process of destruction is
574 * found in the hash table.
576 * \see htable_lookup().
578 wait_queue_head_t lsb_marche_funebre;
586 LU_SS_CACHE_DEATH_RACE,
588 LU_SS_LRU_LEN, /* # of objects in lsb_lru lists */
593 * lu_site is a "compartment" within which objects are unique, and LRU
594 * discipline is maintained.
596 * lu_site exists so that multiple layered stacks can co-exist in the same
599 * lu_site has the same relation to lu_device as lu_object_header to
606 struct cfs_hash *ls_obj_hash;
608 * index of bucket on hash table while purging
612 * Top-level device for this stack.
614 struct lu_device *ls_top_dev;
616 * Bottom-level device for this stack
618 struct lu_device *ls_bottom_dev;
620 * Linkage into global list of sites.
622 struct list_head ls_linkage;
624 * List for lu device for this site, protected
627 struct list_head ls_ld_linkage;
628 spinlock_t ls_ld_lock;
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;
640 static inline struct lu_site_bkt_data *
641 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
643 struct cfs_hash_bd bd;
645 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
646 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
649 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
651 return s->ld_seq_site;
655 * Constructors/destructors.
659 int lu_site_init (struct lu_site *s, struct lu_device *d);
660 void lu_site_fini (struct lu_site *s);
661 int lu_site_init_finish (struct lu_site *s);
662 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
663 void lu_device_get (struct lu_device *d);
664 void lu_device_put (struct lu_device *d);
665 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
666 void lu_device_fini (struct lu_device *d);
667 int lu_object_header_init(struct lu_object_header *h);
668 void lu_object_header_fini(struct lu_object_header *h);
669 int lu_object_init (struct lu_object *o,
670 struct lu_object_header *h, struct lu_device *d);
671 void lu_object_fini (struct lu_object *o);
672 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
673 void lu_object_add (struct lu_object *before, struct lu_object *o);
676 * Helpers to initialize and finalize device types.
679 int lu_device_type_init(struct lu_device_type *ldt);
680 void lu_device_type_fini(struct lu_device_type *ldt);
681 void lu_types_stop(void);
686 * Caching and reference counting.
691 * Acquire additional reference to the given object. This function is used to
692 * attain additional reference. To acquire initial reference use
695 static inline void lu_object_get(struct lu_object *o)
697 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
698 atomic_inc(&o->lo_header->loh_ref);
702 * Return true of object will not be cached after last reference to it is
705 static inline int lu_object_is_dying(const struct lu_object_header *h)
707 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
710 void lu_object_put(const struct lu_env *env, struct lu_object *o);
711 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
713 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
715 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
716 lu_printer_t printer);
717 struct lu_object *lu_object_find_at(const struct lu_env *env,
718 struct lu_device *dev,
719 const struct lu_fid *f,
720 const struct lu_object_conf *conf);
721 struct lu_object *lu_object_find_slice(const struct lu_env *env,
722 struct lu_device *dev,
723 const struct lu_fid *f,
724 const struct lu_object_conf *conf);
733 * First (topmost) sub-object of given compound object
735 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
737 LASSERT(!list_empty(&h->loh_layers));
738 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
742 * Next sub-object in the layering
744 static inline struct lu_object *lu_object_next(const struct lu_object *o)
746 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
750 * Pointer to the fid of this object.
752 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
754 return &o->lo_header->loh_fid;
758 * return device operations vector for this object
760 static inline const struct lu_device_operations *
761 lu_object_ops(const struct lu_object *o)
763 return o->lo_dev->ld_ops;
767 * Given a compound object, find its slice, corresponding to the device type
770 struct lu_object *lu_object_locate(struct lu_object_header *h,
771 const struct lu_device_type *dtype);
774 * Printer function emitting messages through libcfs_debug_msg().
776 int lu_cdebug_printer(const struct lu_env *env,
777 void *cookie, const char *format, ...);
780 * Print object description followed by a user-supplied message.
782 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
784 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
786 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
787 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
788 CDEBUG(mask, format, ## __VA_ARGS__); \
793 * Print short object description followed by a user-supplied message.
795 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
797 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
799 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
800 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
801 (object)->lo_header); \
802 lu_cdebug_printer(env, &msgdata, "\n"); \
803 CDEBUG(mask, format, ## __VA_ARGS__); \
807 void lu_object_print (const struct lu_env *env, void *cookie,
808 lu_printer_t printer, const struct lu_object *o);
809 void lu_object_header_print(const struct lu_env *env, void *cookie,
810 lu_printer_t printer,
811 const struct lu_object_header *hdr);
814 * Check object consistency.
816 int lu_object_invariant(const struct lu_object *o);
819 * Check whether object exists, no matter on local or remote storage.
820 * Note: LOHA_EXISTS will be set once some one created the object,
821 * and it does not needs to be committed to storage.
823 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
826 * Check whether object on the remote storage.
828 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
830 static inline int lu_object_assert_exists(const struct lu_object *o)
832 return lu_object_exists(o);
835 static inline int lu_object_assert_not_exists(const struct lu_object *o)
837 return !lu_object_exists(o);
841 * Attr of this object.
843 static inline __u32 lu_object_attr(const struct lu_object *o)
845 LASSERT(lu_object_exists(o) != 0);
846 return o->lo_header->loh_attr;
849 static inline void lu_object_ref_add(struct lu_object *o,
853 lu_ref_add(&o->lo_header->loh_reference, scope, source);
856 static inline void lu_object_ref_add_at(struct lu_object *o,
857 struct lu_ref_link *link,
861 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
864 static inline void lu_object_ref_del(struct lu_object *o,
865 const char *scope, const void *source)
867 lu_ref_del(&o->lo_header->loh_reference, scope, source);
870 static inline void lu_object_ref_del_at(struct lu_object *o,
871 struct lu_ref_link *link,
872 const char *scope, const void *source)
874 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
877 /** input params, should be filled out by mdt */
881 /** count in bytes */
882 unsigned int rp_count;
883 /** number of pages */
884 unsigned int rp_npages;
885 /** requested attr */
887 /** pointers to pages */
888 struct page **rp_pages;
891 enum lu_xattr_flags {
892 LU_XATTR_REPLACE = (1 << 0),
893 LU_XATTR_CREATE = (1 << 1)
902 /** For lu_context health-checks */
903 enum lu_context_state {
911 * lu_context. Execution context for lu_object methods. Currently associated
914 * All lu_object methods, except device and device type methods (called during
915 * system initialization and shutdown) are executed "within" some
916 * lu_context. This means, that pointer to some "current" lu_context is passed
917 * as an argument to all methods.
919 * All service ptlrpc threads create lu_context as part of their
920 * initialization. It is possible to create "stand-alone" context for other
921 * execution environments (like system calls).
923 * lu_object methods mainly use lu_context through lu_context_key interface
924 * that allows each layer to associate arbitrary pieces of data with each
925 * context (see pthread_key_create(3) for similar interface).
927 * On a client, lu_context is bound to a thread, see cl_env_get().
929 * \see lu_context_key
933 * lu_context is used on the client side too. Yet we don't want to
934 * allocate values of server-side keys for the client contexts and
937 * To achieve this, set of tags in introduced. Contexts and keys are
938 * marked with tags. Key value are created only for context whose set
939 * of tags has non-empty intersection with one for key. Tags are taken
940 * from enum lu_context_tag.
943 enum lu_context_state lc_state;
945 * Pointer to the home service thread. NULL for other execution
948 struct ptlrpc_thread *lc_thread;
950 * Pointer to an array with key values. Internal implementation
955 * Linkage into a list of all remembered contexts. Only
956 * `non-transient' contexts, i.e., ones created for service threads
959 struct list_head lc_remember;
961 * Version counter used to skip calls to lu_context_refill() when no
962 * keys were registered.
972 * lu_context_key interface. Similar to pthread_key.
975 enum lu_context_tag {
977 * Thread on md server
979 LCT_MD_THREAD = 1 << 0,
981 * Thread on dt server
983 LCT_DT_THREAD = 1 << 1,
985 * Context for transaction handle
987 LCT_TX_HANDLE = 1 << 2,
991 LCT_CL_THREAD = 1 << 3,
993 * A per-request session on a server, and a per-system-call session on
996 LCT_SESSION = 1 << 4,
998 * A per-request data on OSP device
1000 LCT_OSP_THREAD = 1 << 5,
1004 LCT_MG_THREAD = 1 << 6,
1006 * Context for local operations
1010 * Set when at least one of keys, having values in this context has
1011 * non-NULL lu_context_key::lct_exit() method. This is used to
1012 * optimize lu_context_exit() call.
1014 LCT_HAS_EXIT = 1 << 28,
1016 * Don't add references for modules creating key values in that context.
1017 * This is only for contexts used internally by lu_object framework.
1019 LCT_NOREF = 1 << 29,
1021 * Key is being prepared for retiring, don't create new values for it.
1023 LCT_QUIESCENT = 1 << 30,
1025 * Context should be remembered.
1027 LCT_REMEMBER = 1 << 31,
1029 * Contexts usable in cache shrinker thread.
1031 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1035 * Key. Represents per-context value slot.
1037 * Keys are usually registered when module owning the key is initialized, and
1038 * de-registered when module is unloaded. Once key is registered, all new
1039 * contexts with matching tags, will get key value. "Old" contexts, already
1040 * initialized at the time of key registration, can be forced to get key value
1041 * by calling lu_context_refill().
1043 * Every key value is counted in lu_context_key::lct_used and acquires a
1044 * reference on an owning module. This means, that all key values have to be
1045 * destroyed before module can be unloaded. This is usually achieved by
1046 * stopping threads started by the module, that created contexts in their
1047 * entry functions. Situation is complicated by the threads shared by multiple
1048 * modules, like ptlrpcd daemon on a client. To work around this problem,
1049 * contexts, created in such threads, are `remembered' (see
1050 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1051 * for unloading it does the following:
1053 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1054 * preventing new key values from being allocated in the new contexts,
1057 * - scans a list of remembered contexts, destroying values of module
1058 * keys, thus releasing references to the module.
1060 * This is done by lu_context_key_quiesce(). If module is re-activated
1061 * before key has been de-registered, lu_context_key_revive() call clears
1062 * `quiescent' marker.
1064 * lu_context code doesn't provide any internal synchronization for these
1065 * activities---it's assumed that startup (including threads start-up) and
1066 * shutdown are serialized by some external means.
1070 struct lu_context_key {
1072 * Set of tags for which values of this key are to be instantiated.
1076 * Value constructor. This is called when new value is created for a
1077 * context. Returns pointer to new value of error pointer.
1079 void *(*lct_init)(const struct lu_context *ctx,
1080 struct lu_context_key *key);
1082 * Value destructor. Called when context with previously allocated
1083 * value of this slot is destroyed. \a data is a value that was returned
1084 * by a matching call to lu_context_key::lct_init().
1086 void (*lct_fini)(const struct lu_context *ctx,
1087 struct lu_context_key *key, void *data);
1089 * Optional method called on lu_context_exit() for all allocated
1090 * keys. Can be used by debugging code checking that locks are
1093 void (*lct_exit)(const struct lu_context *ctx,
1094 struct lu_context_key *key, void *data);
1096 * Internal implementation detail: index within lu_context::lc_value[]
1097 * reserved for this key.
1101 * Internal implementation detail: number of values created for this
1106 * Internal implementation detail: module for this key.
1108 struct module *lct_owner;
1110 * References to this key. For debugging.
1112 struct lu_ref lct_reference;
1115 #define LU_KEY_INIT(mod, type) \
1116 static void *mod##_key_init(const struct lu_context *ctx, \
1117 struct lu_context_key *key) \
1121 CLASSERT(PAGE_SIZE >= sizeof (*value)); \
1123 value = kzalloc(sizeof(*value), GFP_NOFS); \
1125 value = ERR_PTR(-ENOMEM); \
1129 struct __##mod##__dummy_init {; } /* semicolon catcher */
1131 #define LU_KEY_FINI(mod, type) \
1132 static void mod##_key_fini(const struct lu_context *ctx, \
1133 struct lu_context_key *key, void *data) \
1135 type *info = data; \
1139 struct __##mod##__dummy_fini {; } /* semicolon catcher */
1141 #define LU_KEY_INIT_FINI(mod, type) \
1142 LU_KEY_INIT(mod, type); \
1143 LU_KEY_FINI(mod, type)
1145 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1146 struct lu_context_key mod##_thread_key = { \
1148 .lct_init = mod##_key_init, \
1149 .lct_fini = mod##_key_fini \
1152 #define LU_CONTEXT_KEY_INIT(key) \
1154 (key)->lct_owner = THIS_MODULE; \
1157 int lu_context_key_register(struct lu_context_key *key);
1158 void lu_context_key_degister(struct lu_context_key *key);
1159 void *lu_context_key_get (const struct lu_context *ctx,
1160 const struct lu_context_key *key);
1161 void lu_context_key_quiesce (struct lu_context_key *key);
1162 void lu_context_key_revive (struct lu_context_key *key);
1165 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1169 #define LU_KEY_INIT_GENERIC(mod) \
1170 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1172 struct lu_context_key *key = k; \
1175 va_start(args, k); \
1177 LU_CONTEXT_KEY_INIT(key); \
1178 key = va_arg(args, struct lu_context_key *); \
1183 #define LU_TYPE_INIT(mod, ...) \
1184 LU_KEY_INIT_GENERIC(mod) \
1185 static int mod##_type_init(struct lu_device_type *t) \
1187 mod##_key_init_generic(__VA_ARGS__, NULL); \
1188 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1190 struct __##mod##_dummy_type_init {; }
1192 #define LU_TYPE_FINI(mod, ...) \
1193 static void mod##_type_fini(struct lu_device_type *t) \
1195 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1197 struct __##mod##_dummy_type_fini {; }
1199 #define LU_TYPE_START(mod, ...) \
1200 static void mod##_type_start(struct lu_device_type *t) \
1202 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1204 struct __##mod##_dummy_type_start {; }
1206 #define LU_TYPE_STOP(mod, ...) \
1207 static void mod##_type_stop(struct lu_device_type *t) \
1209 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1211 struct __##mod##_dummy_type_stop {; }
1213 #define LU_TYPE_INIT_FINI(mod, ...) \
1214 LU_TYPE_INIT(mod, __VA_ARGS__); \
1215 LU_TYPE_FINI(mod, __VA_ARGS__); \
1216 LU_TYPE_START(mod, __VA_ARGS__); \
1217 LU_TYPE_STOP(mod, __VA_ARGS__)
1219 int lu_context_init (struct lu_context *ctx, __u32 tags);
1220 void lu_context_fini (struct lu_context *ctx);
1221 void lu_context_enter (struct lu_context *ctx);
1222 void lu_context_exit (struct lu_context *ctx);
1223 int lu_context_refill(struct lu_context *ctx);
1226 * Helper functions to operate on multiple keys. These are used by the default
1227 * device type operations, defined by LU_TYPE_INIT_FINI().
1230 int lu_context_key_register_many(struct lu_context_key *k, ...);
1231 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1232 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1233 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1240 * "Local" context, used to store data instead of stack.
1242 struct lu_context le_ctx;
1244 * "Session" context for per-request data.
1246 struct lu_context *le_ses;
1249 int lu_env_init (struct lu_env *env, __u32 tags);
1250 void lu_env_fini (struct lu_env *env);
1251 int lu_env_refill(struct lu_env *env);
1253 /** @} lu_context */
1256 * Output site statistical counters into a buffer. Suitable for
1257 * ll_rd_*()-style functions.
1259 int lu_site_stats_print(const struct lu_site *s, struct seq_file *m);
1262 * Common name structure to be passed around for various name related methods.
1265 const char *ln_name;
1270 * Common buffer structure to be passed around for various xattr_{s,g}et()
1278 #define DLUBUF "(%p %zu)"
1279 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1281 * One-time initializers, called at obdclass module initialization, not
1286 * Initialization of global lu_* data.
1288 int lu_global_init(void);
1291 * Dual to lu_global_init().
1293 void lu_global_fini(void);
1295 struct lu_kmem_descr {
1296 struct kmem_cache **ckd_cache;
1297 const char *ckd_name;
1298 const size_t ckd_size;
1301 int lu_kmem_init(struct lu_kmem_descr *caches);
1302 void lu_kmem_fini(struct lu_kmem_descr *caches);
1305 #endif /* __LUSTRE_LU_OBJECT_H */