1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
40 #ifndef __LUSTRE_LU_OBJECT_H
41 #define __LUSTRE_LU_OBJECT_H
48 #include <libcfs/libcfs.h>
50 #include <lustre/lustre_idl.h>
55 struct proc_dir_entry;
60 * lu_* data-types represent server-side entities shared by data and meta-data
65 * -# support for layering.
67 * Server side object is split into layers, one per device in the
68 * corresponding device stack. Individual layer is represented by struct
69 * lu_object. Compound layered object --- by struct lu_object_header. Most
70 * interface functions take lu_object as an argument and operate on the
71 * whole compound object. This decision was made due to the following
74 * - it's envisaged that lu_object will be used much more often than
77 * - we want lower (non-top) layers to be able to initiate operations
78 * on the whole object.
80 * Generic code supports layering more complex than simple stacking, e.g.,
81 * it is possible that at some layer object "spawns" multiple sub-objects
84 * -# fid-based identification.
86 * Compound object is uniquely identified by its fid. Objects are indexed
87 * by their fids (hash table is used for index).
89 * -# caching and life-cycle management.
91 * Object's life-time is controlled by reference counting. When reference
92 * count drops to 0, object is returned to cache. Cached objects still
93 * retain their identity (i.e., fid), and can be recovered from cache.
95 * Objects are kept in the global LRU list, and lu_site_purge() function
96 * can be used to reclaim given number of unused objects from the tail of
99 * -# avoiding recursion.
101 * Generic code tries to replace recursion through layers by iterations
102 * where possible. Additionally to the end of reducing stack consumption,
103 * data, when practically possible, are allocated through lu_context_key
104 * interface rather than on stack.
111 struct lu_object_header;
116 * Operations common for data and meta-data devices.
118 struct lu_device_operations {
120 * Allocate object for the given device (without lower-layer
121 * parts). This is called by lu_object_operations::loo_object_init()
122 * from the parent layer, and should setup at least lu_object::lo_dev
123 * and lu_object::lo_ops fields of resulting lu_object.
125 * Object creation protocol.
127 * Due to design goal of avoiding recursion, object creation (see
128 * lu_object_alloc()) is somewhat involved:
130 * - first, lu_device_operations::ldo_object_alloc() method of the
131 * top-level device in the stack is called. It should allocate top
132 * level object (including lu_object_header), but without any
133 * lower-layer sub-object(s).
135 * - then lu_object_alloc() sets fid in the header of newly created
138 * - then lu_object_operations::loo_object_init() is called. It has
139 * to allocate lower-layer object(s). To do this,
140 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
141 * of the lower-layer device(s).
143 * - for all new objects allocated by
144 * lu_object_operations::loo_object_init() (and inserted into object
145 * stack), lu_object_operations::loo_object_init() is called again
146 * repeatedly, until no new objects are created.
148 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
149 * result->lo_ops != NULL);
151 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
152 const struct lu_object_header *h,
153 struct lu_device *d);
155 * process config specific for device.
157 int (*ldo_process_config)(const struct lu_env *env,
158 struct lu_device *, struct lustre_cfg *);
159 int (*ldo_recovery_complete)(const struct lu_env *,
163 * initialize local objects for device. this method called after layer has
164 * been initialized (after LCFG_SETUP stage) and before it starts serving
168 int (*ldo_prepare)(const struct lu_env *,
169 struct lu_device *parent,
170 struct lu_device *dev);
175 * For lu_object_conf flags
178 /* Currently, only used for client-side object initialization. */
183 * Object configuration, describing particulars of object being created. On
184 * server this is not used, as server objects are full identified by fid. On
185 * client configuration contains struct lustre_md.
187 struct lu_object_conf {
189 * Some hints for obj find and alloc.
191 loc_flags_t loc_flags;
195 * Type of "printer" function used by lu_object_operations::loo_object_print()
198 * Printer function is needed to provide some flexibility in (semi-)debugging
199 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
201 typedef int (*lu_printer_t)(const struct lu_env *env,
202 void *cookie, const char *format, ...)
203 __attribute__ ((format (printf, 3, 4)));
206 * Operations specific for particular lu_object.
208 struct lu_object_operations {
211 * Allocate lower-layer parts of the object by calling
212 * lu_device_operations::ldo_object_alloc() of the corresponding
215 * This method is called once for each object inserted into object
216 * stack. It's responsibility of this method to insert lower-layer
217 * object(s) it create into appropriate places of object stack.
219 int (*loo_object_init)(const struct lu_env *env,
221 const struct lu_object_conf *conf);
223 * Called (in top-to-bottom order) during object allocation after all
224 * layers were allocated and initialized. Can be used to perform
225 * initialization depending on lower layers.
227 int (*loo_object_start)(const struct lu_env *env,
228 struct lu_object *o);
230 * Called before lu_object_operations::loo_object_free() to signal
231 * that object is being destroyed. Dual to
232 * lu_object_operations::loo_object_init().
234 void (*loo_object_delete)(const struct lu_env *env,
235 struct lu_object *o);
237 * Dual to lu_device_operations::ldo_object_alloc(). Called when
238 * object is removed from memory.
240 void (*loo_object_free)(const struct lu_env *env,
241 struct lu_object *o);
243 * Called when last active reference to the object is released (and
244 * object returns to the cache). This method is optional.
246 void (*loo_object_release)(const struct lu_env *env,
247 struct lu_object *o);
249 * Optional debugging helper. Print given object.
251 int (*loo_object_print)(const struct lu_env *env, void *cookie,
252 lu_printer_t p, const struct lu_object *o);
254 * Optional debugging method. Returns true iff method is internally
257 int (*loo_object_invariant)(const struct lu_object *o);
263 struct lu_device_type;
266 * Device: a layer in the server side abstraction stacking.
270 * reference count. This is incremented, in particular, on each object
271 * created at this layer.
273 * \todo XXX which means that atomic_t is probably too small.
277 * Pointer to device type. Never modified once set.
279 struct lu_device_type *ld_type;
281 * Operation vector for this device.
283 const struct lu_device_operations *ld_ops;
285 * Stack this device belongs to.
287 struct lu_site *ld_site;
288 struct proc_dir_entry *ld_proc_entry;
290 /** \todo XXX: temporary back pointer into obd. */
291 struct obd_device *ld_obd;
293 * A list of references to this object, for debugging.
295 struct lu_ref ld_reference;
298 struct lu_device_type_operations;
301 * Tag bits for device type. They are used to distinguish certain groups of
305 /** this is meta-data device */
306 LU_DEVICE_MD = (1 << 0),
307 /** this is data device */
308 LU_DEVICE_DT = (1 << 1),
309 /** data device in the client stack */
310 LU_DEVICE_CL = (1 << 2)
316 struct lu_device_type {
318 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
322 * Name of this class. Unique system-wide. Never modified once set.
326 * Operations for this type.
328 const struct lu_device_type_operations *ldt_ops;
330 * \todo XXX: temporary pointer to associated obd_type.
332 struct obd_type *ldt_obd_type;
334 * \todo XXX: temporary: context tags used by obd_*() calls.
338 * Number of existing device type instances.
340 unsigned ldt_device_nr;
342 * Linkage into a global list of all device types.
344 * \see lu_device_types.
346 cfs_list_t ldt_linkage;
350 * Operations on a device type.
352 struct lu_device_type_operations {
354 * Allocate new device.
356 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
357 struct lu_device_type *t,
358 struct lustre_cfg *lcfg);
360 * Free device. Dual to
361 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
362 * the next device in the stack.
364 struct lu_device *(*ldto_device_free)(const struct lu_env *,
368 * Initialize the devices after allocation
370 int (*ldto_device_init)(const struct lu_env *env,
371 struct lu_device *, const char *,
374 * Finalize device. Dual to
375 * lu_device_type_operations::ldto_device_init(). Returns pointer to
376 * the next device in the stack.
378 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
381 * Initialize device type. This is called on module load.
383 int (*ldto_init)(struct lu_device_type *t);
385 * Finalize device type. Dual to
386 * lu_device_type_operations::ldto_init(). Called on module unload.
388 void (*ldto_fini)(struct lu_device_type *t);
390 * Called when the first device is created.
392 void (*ldto_start)(struct lu_device_type *t);
394 * Called when number of devices drops to 0.
396 void (*ldto_stop)(struct lu_device_type *t);
400 * Flags for the object layers.
402 enum lu_object_flags {
404 * this flags is set if lu_object_operations::loo_object_init() has
405 * been called for this layer. Used by lu_object_alloc().
407 LU_OBJECT_ALLOCATED = (1 << 0)
411 * Common object attributes.
416 /** modification time in seconds since Epoch */
418 /** access time in seconds since Epoch */
420 /** change time in seconds since Epoch */
422 /** 512-byte blocks allocated to object */
424 /** permission bits and file type */
432 /** number of persistent references to this object */
434 /** blk bits of the object*/
436 /** blk size of the object*/
448 /** Bit-mask of valid attributes */
462 LA_BLKSIZE = 1 << 12,
463 LA_KILL_SUID = 1 << 13,
464 LA_KILL_SGID = 1 << 14,
468 * Layer in the layered object.
472 * Header for this object.
474 struct lu_object_header *lo_header;
476 * Device for this layer.
478 struct lu_device *lo_dev;
480 * Operations for this object.
482 const struct lu_object_operations *lo_ops;
484 * Linkage into list of all layers.
486 cfs_list_t lo_linkage;
488 * Depth. Top level layer depth is 0.
492 * Flags from enum lu_object_flags.
494 unsigned long lo_flags;
496 * Link to the device, for debugging.
498 struct lu_ref_link *lo_dev_ref;
501 enum lu_object_header_flags {
503 * Don't keep this object in cache. Object will be destroyed as soon
504 * as last reference to it is released. This flag cannot be cleared
507 LU_OBJECT_HEARD_BANSHEE = 0
510 enum lu_object_header_attr {
511 LOHA_EXISTS = 1 << 0,
512 LOHA_REMOTE = 1 << 1,
514 * UNIX file type is stored in S_IFMT bits.
516 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
517 LOHA_FT_END = 017 << 12, /**< S_IFMT */
521 * "Compound" object, consisting of multiple layers.
523 * Compound object with given fid is unique with given lu_site.
525 * Note, that object does *not* necessary correspond to the real object in the
526 * persistent storage: object is an anchor for locking and method calling, so
527 * it is created for things like not-yet-existing child created by mkdir or
528 * create calls. lu_object_operations::loo_exists() can be used to check
529 * whether object is backed by persistent storage entity.
531 struct lu_object_header {
533 * Object flags from enum lu_object_header_flags. Set and checked
536 unsigned long loh_flags;
538 * Object reference count. Protected by lu_site::ls_guard.
540 cfs_atomic_t loh_ref;
542 * Fid, uniquely identifying this object.
544 struct lu_fid loh_fid;
546 * Common object attributes, cached for efficiency. From enum
547 * lu_object_header_attr.
551 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
553 cfs_hlist_node_t loh_hash;
555 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
559 * Linkage into list of layers. Never modified once set (except lately
560 * during object destruction). No locking is necessary.
562 cfs_list_t loh_layers;
564 * A list of references to this object, for debugging.
566 struct lu_ref loh_reference;
571 struct lu_site_bkt_data {
573 * number of busy object on this bucket
577 * LRU list, updated on each access to object. Protected by
578 * bucket lock of lu_site::ls_obj_hash.
580 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
581 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
582 * of list_for_each_entry_safe_reverse()).
586 * Wait-queue signaled when an object in this site is ultimately
587 * destroyed (lu_object_free()). It is used by lu_object_find() to
588 * wait before re-trying when object in the process of destruction is
589 * found in the hash table.
591 * \see htable_lookup().
593 cfs_waitq_t lsb_marche_funebre;
601 LU_SS_CACHE_DEATH_RACE,
607 * lu_site is a "compartment" within which objects are unique, and LRU
608 * discipline is maintained.
610 * lu_site exists so that multiple layered stacks can co-exist in the same
613 * lu_site has the same relation to lu_device as lu_object_header to
620 cfs_hash_t *ls_obj_hash;
622 * index of bucket on hash table while purging
626 * Top-level device for this stack.
628 struct lu_device *ls_top_dev;
630 * Linkage into global list of sites.
632 cfs_list_t ls_linkage;
636 struct lprocfs_stats *ls_stats;
637 struct lprocfs_stats *ls_time_stats;
640 static inline struct lu_site_bkt_data *
641 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
645 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
646 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
650 * Constructors/destructors.
654 int lu_site_init (struct lu_site *s, struct lu_device *d);
655 void lu_site_fini (struct lu_site *s);
656 int lu_site_init_finish (struct lu_site *s);
657 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
658 void lu_device_get (struct lu_device *d);
659 void lu_device_put (struct lu_device *d);
660 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
661 void lu_device_fini (struct lu_device *d);
662 int lu_object_header_init(struct lu_object_header *h);
663 void lu_object_header_fini(struct lu_object_header *h);
664 int lu_object_init (struct lu_object *o,
665 struct lu_object_header *h, struct lu_device *d);
666 void lu_object_fini (struct lu_object *o);
667 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
668 void lu_object_add (struct lu_object *before, struct lu_object *o);
671 * Helpers to initialize and finalize device types.
674 int lu_device_type_init(struct lu_device_type *ldt);
675 void lu_device_type_fini(struct lu_device_type *ldt);
676 void lu_types_stop(void);
681 * Caching and reference counting.
686 * Acquire additional reference to the given object. This function is used to
687 * attain additional reference. To acquire initial reference use
690 static inline void lu_object_get(struct lu_object *o)
692 LASSERT(cfs_atomic_read(&o->lo_header->loh_ref) > 0);
693 cfs_atomic_inc(&o->lo_header->loh_ref);
697 * Return true of object will not be cached after last reference to it is
700 static inline int lu_object_is_dying(const struct lu_object_header *h)
702 return cfs_test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
705 void lu_object_put(const struct lu_env *env, struct lu_object *o);
707 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
709 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
710 lu_printer_t printer);
711 struct lu_object *lu_object_find(const struct lu_env *env,
712 struct lu_device *dev, const struct lu_fid *f,
713 const struct lu_object_conf *conf);
714 struct lu_object *lu_object_find_at(const struct lu_env *env,
715 struct lu_device *dev,
716 const struct lu_fid *f,
717 const struct lu_object_conf *conf);
718 struct lu_object *lu_object_find_slice(const struct lu_env *env,
719 struct lu_device *dev,
720 const struct lu_fid *f,
721 const struct lu_object_conf *conf);
730 * First (topmost) sub-object of given compound object
732 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
734 LASSERT(!cfs_list_empty(&h->loh_layers));
735 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
739 * Next sub-object in the layering
741 static inline struct lu_object *lu_object_next(const struct lu_object *o)
743 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
747 * Pointer to the fid of this object.
749 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
751 return &o->lo_header->loh_fid;
755 * return device operations vector for this object
757 static const inline struct lu_device_operations *
758 lu_object_ops(const struct lu_object *o)
760 return o->lo_dev->ld_ops;
764 * Given a compound object, find its slice, corresponding to the device type
767 struct lu_object *lu_object_locate(struct lu_object_header *h,
768 const struct lu_device_type *dtype);
770 struct lu_cdebug_print_info {
773 const char *lpi_file;
779 * Printer function emitting messages through libcfs_debug_msg().
781 int lu_cdebug_printer(const struct lu_env *env,
782 void *cookie, const char *format, ...);
784 #define DECLARE_LU_CDEBUG_PRINT_INFO(var, mask) \
785 struct lu_cdebug_print_info var = { \
786 .lpi_subsys = DEBUG_SUBSYSTEM, \
787 .lpi_mask = (mask), \
788 .lpi_file = __FILE__, \
789 .lpi_fn = __FUNCTION__, \
790 .lpi_line = __LINE__ \
794 * Print object description followed by a user-supplied message.
796 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
798 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
800 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
801 lu_object_print(env, &__info, lu_cdebug_printer, object); \
802 CDEBUG(mask, format , ## __VA_ARGS__); \
807 * Print short object description followed by a user-supplied message.
809 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
811 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
813 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
814 lu_object_header_print(env, &__info, lu_cdebug_printer, \
815 (object)->lo_header); \
816 lu_cdebug_printer(env, &__info, "\n"); \
817 CDEBUG(mask, format , ## __VA_ARGS__); \
821 void lu_object_print (const struct lu_env *env, void *cookie,
822 lu_printer_t printer, const struct lu_object *o);
823 void lu_object_header_print(const struct lu_env *env, void *cookie,
824 lu_printer_t printer,
825 const struct lu_object_header *hdr);
828 * Check object consistency.
830 int lu_object_invariant(const struct lu_object *o);
834 * \retval 1 iff object \a o exists on stable storage,
835 * \retval 0 iff object \a o not exists on stable storage.
836 * \retval -1 iff object \a o is on remote server.
838 static inline int lu_object_exists(const struct lu_object *o)
842 attr = o->lo_header->loh_attr;
843 if (attr & LOHA_REMOTE)
845 else if (attr & LOHA_EXISTS)
851 static inline int lu_object_assert_exists(const struct lu_object *o)
853 return lu_object_exists(o) != 0;
856 static inline int lu_object_assert_not_exists(const struct lu_object *o)
858 return lu_object_exists(o) <= 0;
862 * Attr of this object.
864 static inline __u32 lu_object_attr(const struct lu_object *o)
866 LASSERT(lu_object_exists(o) > 0);
867 return o->lo_header->loh_attr;
870 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
874 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
877 static inline void lu_object_ref_del(struct lu_object *o,
878 const char *scope, const void *source)
880 lu_ref_del(&o->lo_header->loh_reference, scope, source);
883 static inline void lu_object_ref_del_at(struct lu_object *o,
884 struct lu_ref_link *link,
885 const char *scope, const void *source)
887 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
890 /** input params, should be filled out by mdt */
894 /** count in bytes */
895 unsigned int rp_count;
896 /** number of pages */
897 unsigned int rp_npages;
898 /** requested attr */
900 /** pointers to pages */
901 struct page **rp_pages;
904 enum lu_xattr_flags {
905 LU_XATTR_REPLACE = (1 << 0),
906 LU_XATTR_CREATE = (1 << 1)
914 /** For lu_context health-checks */
915 enum lu_context_state {
923 * lu_context. Execution context for lu_object methods. Currently associated
926 * All lu_object methods, except device and device type methods (called during
927 * system initialization and shutdown) are executed "within" some
928 * lu_context. This means, that pointer to some "current" lu_context is passed
929 * as an argument to all methods.
931 * All service ptlrpc threads create lu_context as part of their
932 * initialization. It is possible to create "stand-alone" context for other
933 * execution environments (like system calls).
935 * lu_object methods mainly use lu_context through lu_context_key interface
936 * that allows each layer to associate arbitrary pieces of data with each
937 * context (see pthread_key_create(3) for similar interface).
939 * On a client, lu_context is bound to a thread, see cl_env_get().
941 * \see lu_context_key
945 * lu_context is used on the client side too. Yet we don't want to
946 * allocate values of server-side keys for the client contexts and
949 * To achieve this, set of tags in introduced. Contexts and keys are
950 * marked with tags. Key value are created only for context whose set
951 * of tags has non-empty intersection with one for key. Tags are taken
952 * from enum lu_context_tag.
956 * Pointer to the home service thread. NULL for other execution
959 struct ptlrpc_thread *lc_thread;
961 * Pointer to an array with key values. Internal implementation
965 enum lu_context_state lc_state;
967 * Linkage into a list of all remembered contexts. Only
968 * `non-transient' contexts, i.e., ones created for service threads
971 cfs_list_t lc_remember;
973 * Version counter used to skip calls to lu_context_refill() when no
974 * keys were registered.
984 * lu_context_key interface. Similar to pthread_key.
987 enum lu_context_tag {
989 * Thread on md server
991 LCT_MD_THREAD = 1 << 0,
993 * Thread on dt server
995 LCT_DT_THREAD = 1 << 1,
997 * Context for transaction handle
999 LCT_TX_HANDLE = 1 << 2,
1003 LCT_CL_THREAD = 1 << 3,
1005 * A per-request session on a server, and a per-system-call session on
1008 LCT_SESSION = 1 << 4,
1011 * Set when at least one of keys, having values in this context has
1012 * non-NULL lu_context_key::lct_exit() method. This is used to
1013 * optimize lu_context_exit() call.
1015 LCT_HAS_EXIT = 1 << 28,
1017 * Don't add references for modules creating key values in that context.
1018 * This is only for contexts used internally by lu_object framework.
1020 LCT_NOREF = 1 << 29,
1022 * Key is being prepared for retiring, don't create new values for it.
1024 LCT_QUIESCENT = 1 << 30,
1026 * Context should be remembered.
1028 LCT_REMEMBER = 1 << 31,
1030 * Contexts usable in cache shrinker thread.
1032 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1036 * Key. Represents per-context value slot.
1038 * Keys are usually registered when module owning the key is initialized, and
1039 * de-registered when module is unloaded. Once key is registered, all new
1040 * contexts with matching tags, will get key value. "Old" contexts, already
1041 * initialized at the time of key registration, can be forced to get key value
1042 * by calling lu_context_refill().
1044 * Every key value is counted in lu_context_key::lct_used and acquires a
1045 * reference on an owning module. This means, that all key values have to be
1046 * destroyed before module can be unloaded. This is usually achieved by
1047 * stopping threads started by the module, that created contexts in their
1048 * entry functions. Situation is complicated by the threads shared by multiple
1049 * modules, like ptlrpcd daemon on a client. To work around this problem,
1050 * contexts, created in such threads, are `remembered' (see
1051 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1052 * for unloading it does the following:
1054 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1055 * preventing new key values from being allocated in the new contexts,
1058 * - scans a list of remembered contexts, destroying values of module
1059 * keys, thus releasing references to the module.
1061 * This is done by lu_context_key_quiesce(). If module is re-activated
1062 * before key has been de-registered, lu_context_key_revive() call clears
1063 * `quiescent' marker.
1065 * lu_context code doesn't provide any internal synchronization for these
1066 * activities---it's assumed that startup (including threads start-up) and
1067 * shutdown are serialized by some external means.
1071 struct lu_context_key {
1073 * Set of tags for which values of this key are to be instantiated.
1077 * Value constructor. This is called when new value is created for a
1078 * context. Returns pointer to new value of error pointer.
1080 void *(*lct_init)(const struct lu_context *ctx,
1081 struct lu_context_key *key);
1083 * Value destructor. Called when context with previously allocated
1084 * value of this slot is destroyed. \a data is a value that was returned
1085 * by a matching call to lu_context_key::lct_init().
1087 void (*lct_fini)(const struct lu_context *ctx,
1088 struct lu_context_key *key, void *data);
1090 * Optional method called on lu_context_exit() for all allocated
1091 * keys. Can be used by debugging code checking that locks are
1094 void (*lct_exit)(const struct lu_context *ctx,
1095 struct lu_context_key *key, void *data);
1097 * Internal implementation detail: index within lu_context::lc_value[]
1098 * reserved for this key.
1102 * Internal implementation detail: number of values created for this
1105 cfs_atomic_t lct_used;
1107 * Internal implementation detail: module for this key.
1109 cfs_module_t *lct_owner;
1111 * References to this key. For debugging.
1113 struct lu_ref lct_reference;
1116 #define LU_KEY_INIT(mod, type) \
1117 static void* mod##_key_init(const struct lu_context *ctx, \
1118 struct lu_context_key *key) \
1122 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1124 OBD_ALLOC_PTR(value); \
1125 if (value == NULL) \
1126 value = ERR_PTR(-ENOMEM); \
1130 struct __##mod##__dummy_init {;} /* semicolon catcher */
1132 #define LU_KEY_FINI(mod, type) \
1133 static void mod##_key_fini(const struct lu_context *ctx, \
1134 struct lu_context_key *key, void* data) \
1136 type *info = data; \
1138 OBD_FREE_PTR(info); \
1140 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1142 #define LU_KEY_INIT_FINI(mod, type) \
1143 LU_KEY_INIT(mod,type); \
1144 LU_KEY_FINI(mod,type)
1146 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1147 struct lu_context_key mod##_thread_key = { \
1149 .lct_init = mod##_key_init, \
1150 .lct_fini = mod##_key_fini \
1153 #define LU_CONTEXT_KEY_INIT(key) \
1155 (key)->lct_owner = THIS_MODULE; \
1158 int lu_context_key_register(struct lu_context_key *key);
1159 void lu_context_key_degister(struct lu_context_key *key);
1160 void *lu_context_key_get (const struct lu_context *ctx,
1161 const struct lu_context_key *key);
1162 void lu_context_key_quiesce (struct lu_context_key *key);
1163 void lu_context_key_revive (struct lu_context_key *key);
1167 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1171 #define LU_KEY_INIT_GENERIC(mod) \
1172 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1174 struct lu_context_key *key = k; \
1177 va_start(args, k); \
1179 LU_CONTEXT_KEY_INIT(key); \
1180 key = va_arg(args, struct lu_context_key *); \
1181 } while (key != NULL); \
1185 #define LU_TYPE_INIT(mod, ...) \
1186 LU_KEY_INIT_GENERIC(mod) \
1187 static int mod##_type_init(struct lu_device_type *t) \
1189 mod##_key_init_generic(__VA_ARGS__, NULL); \
1190 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1192 struct __##mod##_dummy_type_init {;}
1194 #define LU_TYPE_FINI(mod, ...) \
1195 static void mod##_type_fini(struct lu_device_type *t) \
1197 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1199 struct __##mod##_dummy_type_fini {;}
1201 #define LU_TYPE_START(mod, ...) \
1202 static void mod##_type_start(struct lu_device_type *t) \
1204 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1206 struct __##mod##_dummy_type_start {;}
1208 #define LU_TYPE_STOP(mod, ...) \
1209 static void mod##_type_stop(struct lu_device_type *t) \
1211 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1213 struct __##mod##_dummy_type_stop {;}
1217 #define LU_TYPE_INIT_FINI(mod, ...) \
1218 LU_TYPE_INIT(mod, __VA_ARGS__); \
1219 LU_TYPE_FINI(mod, __VA_ARGS__); \
1220 LU_TYPE_START(mod, __VA_ARGS__); \
1221 LU_TYPE_STOP(mod, __VA_ARGS__)
1223 int lu_context_init (struct lu_context *ctx, __u32 tags);
1224 void lu_context_fini (struct lu_context *ctx);
1225 void lu_context_enter (struct lu_context *ctx);
1226 void lu_context_exit (struct lu_context *ctx);
1227 int lu_context_refill(struct lu_context *ctx);
1230 * Helper functions to operate on multiple keys. These are used by the default
1231 * device type operations, defined by LU_TYPE_INIT_FINI().
1234 int lu_context_key_register_many(struct lu_context_key *k, ...);
1235 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1236 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1237 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1244 * "Local" context, used to store data instead of stack.
1246 struct lu_context le_ctx;
1248 * "Session" context for per-request data.
1250 struct lu_context *le_ses;
1253 int lu_env_init (struct lu_env *env, __u32 tags);
1254 void lu_env_fini (struct lu_env *env);
1255 int lu_env_refill(struct lu_env *env);
1257 /** @} lu_context */
1260 * Output site statistical counters into a buffer. Suitable for
1261 * ll_rd_*()-style functions.
1263 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1266 * Common name structure to be passed around for various name related methods.
1269 const char *ln_name;
1274 * Common buffer structure to be passed around for various xattr_{s,g}et()
1283 extern struct lu_buf LU_BUF_NULL;
1285 #define DLUBUF "(%p %z)"
1286 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1288 * One-time initializers, called at obdclass module initialization, not
1293 * Initialization of global lu_* data.
1295 int lu_global_init(void);
1298 * Dual to lu_global_init().
1300 void lu_global_fini(void);
1303 LU_TIME_FIND_LOOKUP,
1305 LU_TIME_FIND_INSERT,
1309 extern const char *lu_time_names[LU_TIME_NR];
1311 struct lu_kmem_descr {
1312 cfs_mem_cache_t **ckd_cache;
1313 const char *ckd_name;
1314 const size_t ckd_size;
1317 int lu_kmem_init(struct lu_kmem_descr *caches);
1318 void lu_kmem_fini(struct lu_kmem_descr *caches);
1321 #endif /* __LUSTRE_LU_OBJECT_H */