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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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
45 #include <lustre/lustre_idl.h>
47 #include <libcfs/libcfs.h>
52 struct proc_dir_entry;
57 * lu_* data-types represent server-side entities shared by data and meta-data
62 * -# support for layering.
64 * Server side object is split into layers, one per device in the
65 * corresponding device stack. Individual layer is represented by struct
66 * lu_object. Compound layered object --- by struct lu_object_header. Most
67 * interface functions take lu_object as an argument and operate on the
68 * whole compound object. This decision was made due to the following
71 * - it's envisaged that lu_object will be used much more often than
74 * - we want lower (non-top) layers to be able to initiate operations
75 * on the whole object.
77 * Generic code supports layering more complex than simple stacking, e.g.,
78 * it is possible that at some layer object "spawns" multiple sub-objects
81 * -# fid-based identification.
83 * Compound object is uniquely identified by its fid. Objects are indexed
84 * by their fids (hash table is used for index).
86 * -# caching and life-cycle management.
88 * Object's life-time is controlled by reference counting. When reference
89 * count drops to 0, object is returned to cache. Cached objects still
90 * retain their identity (i.e., fid), and can be recovered from cache.
92 * Objects are kept in the global LRU list, and lu_site_purge() function
93 * can be used to reclaim given number of unused objects from the tail of
96 * -# avoiding recursion.
98 * Generic code tries to replace recursion through layers by iterations
99 * where possible. Additionally to the end of reducing stack consumption,
100 * data, when practically possible, are allocated through lu_context_key
101 * interface rather than on stack.
108 struct lu_object_header;
113 * Operations common for data and meta-data devices.
115 struct lu_device_operations {
117 * Allocate object for the given device (without lower-layer
118 * parts). This is called by lu_object_operations::loo_object_init()
119 * from the parent layer, and should setup at least lu_object::lo_dev
120 * and lu_object::lo_ops fields of resulting lu_object.
122 * Object creation protocol.
124 * Due to design goal of avoiding recursion, object creation (see
125 * lu_object_alloc()) is somewhat involved:
127 * - first, lu_device_operations::ldo_object_alloc() method of the
128 * top-level device in the stack is called. It should allocate top
129 * level object (including lu_object_header), but without any
130 * lower-layer sub-object(s).
132 * - then lu_object_alloc() sets fid in the header of newly created
135 * - then lu_object_operations::loo_object_init() is called. It has
136 * to allocate lower-layer object(s). To do this,
137 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
138 * of the lower-layer device(s).
140 * - for all new objects allocated by
141 * lu_object_operations::loo_object_init() (and inserted into object
142 * stack), lu_object_operations::loo_object_init() is called again
143 * repeatedly, until no new objects are created.
145 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
146 * result->lo_ops != NULL);
148 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
149 const struct lu_object_header *h,
150 struct lu_device *d);
152 * process config specific for device.
154 int (*ldo_process_config)(const struct lu_env *env,
155 struct lu_device *, struct lustre_cfg *);
156 int (*ldo_recovery_complete)(const struct lu_env *,
162 * Object configuration, describing particulars of object being created. On
163 * server this is not used, as server objects are full identified by fid. On
164 * client configuration contains struct lustre_md.
166 struct lu_object_conf {
170 * Type of "printer" function used by lu_object_operations::loo_object_print()
173 * Printer function is needed to provide some flexibility in (semi-)debugging
174 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
176 typedef int (*lu_printer_t)(const struct lu_env *env,
177 void *cookie, const char *format, ...)
178 __attribute__ ((format (printf, 3, 4)));
181 * Operations specific for particular lu_object.
183 struct lu_object_operations {
186 * Allocate lower-layer parts of the object by calling
187 * lu_device_operations::ldo_object_alloc() of the corresponding
190 * This method is called once for each object inserted into object
191 * stack. It's responsibility of this method to insert lower-layer
192 * object(s) it create into appropriate places of object stack.
194 int (*loo_object_init)(const struct lu_env *env,
196 const struct lu_object_conf *conf);
198 * Called (in top-to-bottom order) during object allocation after all
199 * layers were allocated and initialized. Can be used to perform
200 * initialization depending on lower layers.
202 int (*loo_object_start)(const struct lu_env *env,
203 struct lu_object *o);
205 * Called before lu_object_operations::loo_object_free() to signal
206 * that object is being destroyed. Dual to
207 * lu_object_operations::loo_object_init().
209 void (*loo_object_delete)(const struct lu_env *env,
210 struct lu_object *o);
212 * Dual to lu_device_operations::ldo_object_alloc(). Called when
213 * object is removed from memory.
215 void (*loo_object_free)(const struct lu_env *env,
216 struct lu_object *o);
218 * Called when last active reference to the object is released (and
219 * object returns to the cache). This method is optional.
221 void (*loo_object_release)(const struct lu_env *env,
222 struct lu_object *o);
224 * Optional debugging helper. Print given object.
226 int (*loo_object_print)(const struct lu_env *env, void *cookie,
227 lu_printer_t p, const struct lu_object *o);
229 * Optional debugging method. Returns true iff method is internally
232 int (*loo_object_invariant)(const struct lu_object *o);
238 struct lu_device_type;
241 * Device: a layer in the server side abstraction stacking.
245 * reference count. This is incremented, in particular, on each object
246 * created at this layer.
248 * \todo XXX which means that atomic_t is probably too small.
252 * Pointer to device type. Never modified once set.
254 struct lu_device_type *ld_type;
256 * Operation vector for this device.
258 const struct lu_device_operations *ld_ops;
260 * Stack this device belongs to.
262 struct lu_site *ld_site;
263 struct proc_dir_entry *ld_proc_entry;
265 /** \todo XXX: temporary back pointer into obd. */
266 struct obd_device *ld_obd;
268 * A list of references to this object, for debugging.
270 struct lu_ref ld_reference;
273 struct lu_device_type_operations;
276 * Tag bits for device type. They are used to distinguish certain groups of
280 /** this is meta-data device */
281 LU_DEVICE_MD = (1 << 0),
282 /** this is data device */
283 LU_DEVICE_DT = (1 << 1),
284 /** data device in the client stack */
285 LU_DEVICE_CL = (1 << 2)
291 struct lu_device_type {
293 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
297 * Name of this class. Unique system-wide. Never modified once set.
301 * Operations for this type.
303 const struct lu_device_type_operations *ldt_ops;
305 * \todo XXX: temporary pointer to associated obd_type.
307 struct obd_type *ldt_obd_type;
309 * \todo XXX: temporary: context tags used by obd_*() calls.
313 * Number of existing device type instances.
315 unsigned ldt_device_nr;
317 * Linkage into a global list of all device types.
319 * \see lu_device_types.
321 struct list_head ldt_linkage;
325 * Operations on a device type.
327 struct lu_device_type_operations {
329 * Allocate new device.
331 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
332 struct lu_device_type *t,
333 struct lustre_cfg *lcfg);
335 * Free device. Dual to
336 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
337 * the next device in the stack.
339 struct lu_device *(*ldto_device_free)(const struct lu_env *,
343 * Initialize the devices after allocation
345 int (*ldto_device_init)(const struct lu_env *env,
346 struct lu_device *, const char *,
349 * Finalize device. Dual to
350 * lu_device_type_operations::ldto_device_init(). Returns pointer to
351 * the next device in the stack.
353 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
356 * Initialize device type. This is called on module load.
358 int (*ldto_init)(struct lu_device_type *t);
360 * Finalize device type. Dual to
361 * lu_device_type_operations::ldto_init(). Called on module unload.
363 void (*ldto_fini)(struct lu_device_type *t);
365 * Called when the first device is created.
367 void (*ldto_start)(struct lu_device_type *t);
369 * Called when number of devices drops to 0.
371 void (*ldto_stop)(struct lu_device_type *t);
375 * Flags for the object layers.
377 enum lu_object_flags {
379 * this flags is set if lu_object_operations::loo_object_init() has
380 * been called for this layer. Used by lu_object_alloc().
382 LU_OBJECT_ALLOCATED = (1 << 0)
386 * Common object attributes.
391 /** modification time in seconds since Epoch */
393 /** access time in seconds since Epoch */
395 /** change time in seconds since Epoch */
397 /** 512-byte blocks allocated to object */
399 /** permission bits and file type */
407 /** number of persistent references to this object */
409 /** blk bits of the object*/
411 /** blk size of the object*/
423 /** Bit-mask of valid attributes */
437 LA_BLKSIZE = 1 << 12,
441 * Layer in the layered object.
445 * Header for this object.
447 struct lu_object_header *lo_header;
449 * Device for this layer.
451 struct lu_device *lo_dev;
453 * Operations for this object.
455 const struct lu_object_operations *lo_ops;
457 * Linkage into list of all layers.
459 struct list_head lo_linkage;
461 * Depth. Top level layer depth is 0.
465 * Flags from enum lu_object_flags.
467 unsigned long lo_flags;
469 * Link to the device, for debugging.
471 struct lu_ref_link *lo_dev_ref;
474 enum lu_object_header_flags {
476 * Don't keep this object in cache. Object will be destroyed as soon
477 * as last reference to it is released. This flag cannot be cleared
480 LU_OBJECT_HEARD_BANSHEE = 0
483 enum lu_object_header_attr {
484 LOHA_EXISTS = 1 << 0,
485 LOHA_REMOTE = 1 << 1,
487 * UNIX file type is stored in S_IFMT bits.
489 LOHA_FT_START = 1 << 12, /**< S_IFIFO */
490 LOHA_FT_END = 1 << 15, /**< S_IFREG */
494 * "Compound" object, consisting of multiple layers.
496 * Compound object with given fid is unique with given lu_site.
498 * Note, that object does *not* necessary correspond to the real object in the
499 * persistent storage: object is an anchor for locking and method calling, so
500 * it is created for things like not-yet-existing child created by mkdir or
501 * create calls. lu_object_operations::loo_exists() can be used to check
502 * whether object is backed by persistent storage entity.
504 struct lu_object_header {
506 * Object flags from enum lu_object_header_flags. Set and checked
509 unsigned long loh_flags;
511 * Object reference count. Protected by lu_site::ls_guard.
515 * Fid, uniquely identifying this object.
517 struct lu_fid loh_fid;
519 * Common object attributes, cached for efficiency. From enum
520 * lu_object_header_attr.
524 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
526 struct hlist_node loh_hash;
528 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
530 struct list_head loh_lru;
532 * Linkage into list of layers. Never modified once set (except lately
533 * during object destruction). No locking is necessary.
535 struct list_head loh_layers;
537 * A list of references to this object, for debugging.
539 struct lu_ref loh_reference;
545 * lu_site is a "compartment" within which objects are unique, and LRU
546 * discipline is maintained.
548 * lu_site exists so that multiple layered stacks can co-exist in the same
551 * lu_site has the same relation to lu_device as lu_object_header to
560 * - lu_site::ls_hash hash table (and its linkages in objects);
562 * - lu_site::ls_lru list (and its linkages in objects);
564 * - 0/1 transitions of object lu_object_header::loh_ref
571 * Hash-table where objects are indexed by fid.
573 struct hlist_head *ls_hash;
575 * Bit-mask for hash-table size.
579 * Order of hash-table.
583 * Number of buckets in the hash-table.
588 * LRU list, updated on each access to object. Protected by
591 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
592 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
593 * of list_for_each_entry_safe_reverse()).
595 struct list_head ls_lru;
597 * Total number of objects in this site. Protected by
602 * Total number of objects in this site with reference counter greater
603 * than 0. Protected by lu_site::ls_guard.
608 * Top-level device for this stack.
610 struct lu_device *ls_top_dev;
613 * Wait-queue signaled when an object in this site is ultimately
614 * destroyed (lu_object_free()). It is used by lu_object_find() to
615 * wait before re-trying when object in the process of destruction is
616 * found in the hash table.
618 * If having a single wait-queue turns out to be a problem, a
619 * wait-queue per hash-table bucket can be easily implemented.
621 * \see htable_lookup().
623 cfs_waitq_t ls_marche_funebre;
625 /** statistical counters. Protected by nothing, races are accepted. */
631 * Number of hash-table entry checks made.
633 * ->s_cache_check / (->s_cache_miss + ->s_cache_hit)
635 * is an average number of hash slots inspected during single
639 /** Races with cache insertions. */
642 * Races with object destruction.
644 * \see lu_site::ls_marche_funebre.
646 __u32 s_cache_death_race;
651 * Linkage into global list of sites.
653 struct list_head ls_linkage;
654 struct lprocfs_stats *ls_time_stats;
658 * Constructors/destructors.
662 int lu_site_init (struct lu_site *s, struct lu_device *d);
663 void lu_site_fini (struct lu_site *s);
664 int lu_site_init_finish (struct lu_site *s);
665 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
666 void lu_device_get (struct lu_device *d);
667 void lu_device_put (struct lu_device *d);
668 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
669 void lu_device_fini (struct lu_device *d);
670 int lu_object_header_init(struct lu_object_header *h);
671 void lu_object_header_fini(struct lu_object_header *h);
672 int lu_object_init (struct lu_object *o,
673 struct lu_object_header *h, struct lu_device *d);
674 void lu_object_fini (struct lu_object *o);
675 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
676 void lu_object_add (struct lu_object *before, struct lu_object *o);
679 * Helpers to initialize and finalize device types.
682 int lu_device_type_init(struct lu_device_type *ldt);
683 void lu_device_type_fini(struct lu_device_type *ldt);
684 void lu_types_stop(void);
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);
715 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
717 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
718 lu_printer_t printer);
719 struct lu_object *lu_object_find(const struct lu_env *env,
720 struct lu_device *dev, const struct lu_fid *f,
721 const struct lu_object_conf *conf);
722 struct lu_object *lu_object_find_at(const struct lu_env *env,
723 struct lu_device *dev,
724 const struct lu_fid *f,
725 const struct lu_object_conf *conf);
726 struct lu_object *lu_object_find_slice(const struct lu_env *env,
727 struct lu_device *dev,
728 const struct lu_fid *f,
729 const struct lu_object_conf *conf);
738 * First (topmost) sub-object of given compound object
740 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
742 LASSERT(!list_empty(&h->loh_layers));
743 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
747 * Next sub-object in the layering
749 static inline struct lu_object *lu_object_next(const struct lu_object *o)
751 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
755 * Pointer to the fid of this object.
757 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
759 return &o->lo_header->loh_fid;
763 * return device operations vector for this object
765 static const inline struct lu_device_operations *
766 lu_object_ops(const struct lu_object *o)
768 return o->lo_dev->ld_ops;
772 * Given a compound object, find its slice, corresponding to the device type
775 struct lu_object *lu_object_locate(struct lu_object_header *h,
776 const struct lu_device_type *dtype);
778 struct lu_cdebug_print_info {
781 const char *lpi_file;
787 * Printer function emitting messages through libcfs_debug_msg().
789 int lu_cdebug_printer(const struct lu_env *env,
790 void *cookie, const char *format, ...);
792 #define DECLARE_LU_CDEBUG_PRINT_INFO(var, mask) \
793 struct lu_cdebug_print_info var = { \
794 .lpi_subsys = DEBUG_SUBSYSTEM, \
795 .lpi_mask = (mask), \
796 .lpi_file = __FILE__, \
797 .lpi_fn = __FUNCTION__, \
798 .lpi_line = __LINE__ \
802 * Print object description followed by a user-supplied message.
804 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
806 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
808 if (cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
809 lu_object_print(env, &__info, lu_cdebug_printer, object); \
810 CDEBUG(mask, format , ## __VA_ARGS__); \
815 * Print short object description followed by a user-supplied message.
817 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
819 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
821 if (cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
822 lu_object_header_print(env, &__info, lu_cdebug_printer, \
823 (object)->lo_header); \
824 lu_cdebug_printer(env, &__info, "\n"); \
825 CDEBUG(mask, format , ## __VA_ARGS__); \
829 void lu_object_print (const struct lu_env *env, void *cookie,
830 lu_printer_t printer, const struct lu_object *o);
831 void lu_object_header_print(const struct lu_env *env, void *cookie,
832 lu_printer_t printer,
833 const struct lu_object_header *hdr);
836 * Check object consistency.
838 int lu_object_invariant(const struct lu_object *o);
840 void lu_stack_fini(const struct lu_env *env, struct lu_device *top);
843 * Returns 1 iff object @o exists on the stable storage,
844 * returns -1 iff object @o is on remote server.
846 static inline int lu_object_exists(const struct lu_object *o)
850 attr = o->lo_header->loh_attr;
851 if (attr & LOHA_REMOTE)
853 else if (attr & LOHA_EXISTS)
859 static inline int lu_object_assert_exists(const struct lu_object *o)
861 return lu_object_exists(o) != 0;
864 static inline int lu_object_assert_not_exists(const struct lu_object *o)
866 return lu_object_exists(o) <= 0;
870 * Attr of this object.
872 static inline __u32 lu_object_attr(const struct lu_object *o)
874 LASSERT(lu_object_exists(o) > 0);
875 return o->lo_header->loh_attr;
878 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
882 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
885 static inline void lu_object_ref_del(struct lu_object *o,
886 const char *scope, const void *source)
888 lu_ref_del(&o->lo_header->loh_reference, scope, source);
891 static inline void lu_object_ref_del_at(struct lu_object *o,
892 struct lu_ref_link *link,
893 const char *scope, const void *source)
895 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
899 /* input params, should be filled out by mdt */
900 __u64 rp_hash; /* hash */
901 int rp_count; /* count in bytes */
902 int rp_npages; /* number of pages */
903 struct page **rp_pages; /* pointers to pages */
906 enum lu_xattr_flags {
907 LU_XATTR_REPLACE = (1 << 0),
908 LU_XATTR_CREATE = (1 << 1)
916 /** For lu_context health-checks */
917 enum lu_context_state {
925 * lu_context. Execution context for lu_object methods. Currently associated
928 * All lu_object methods, except device and device type methods (called during
929 * system initialization and shutdown) are executed "within" some
930 * lu_context. This means, that pointer to some "current" lu_context is passed
931 * as an argument to all methods.
933 * All service ptlrpc threads create lu_context as part of their
934 * initialization. It is possible to create "stand-alone" context for other
935 * execution environments (like system calls).
937 * lu_object methods mainly use lu_context through lu_context_key interface
938 * that allows each layer to associate arbitrary pieces of data with each
939 * context (see pthread_key_create(3) for similar interface).
941 * On a client, lu_context is bound to a thread, see cl_env_get().
943 * \see lu_context_key
947 * lu_context is used on the client side too. Yet we don't want to
948 * allocate values of server-side keys for the client contexts and
951 * To achieve this, set of tags in introduced. Contexts and keys are
952 * marked with tags. Key value are created only for context whose set
953 * of tags has non-empty intersection with one for key. Tags are taken
954 * from enum lu_context_tag.
958 * Pointer to the home service thread. NULL for other execution
961 struct ptlrpc_thread *lc_thread;
963 * Pointer to an array with key values. Internal implementation
967 enum lu_context_state lc_state;
969 * Linkage into a list of all remembered contexts. Only
970 * `non-transient' contexts, i.e., ones created for service threads
973 struct list_head lc_remember;
975 * Version counter used to skip calls to lu_context_refill() when no
976 * keys were registered.
986 * lu_context_key interface. Similar to pthread_key.
989 enum lu_context_tag {
991 * Thread on md server
993 LCT_MD_THREAD = 1 << 0,
995 * Thread on dt server
997 LCT_DT_THREAD = 1 << 1,
999 * Context for transaction handle
1001 LCT_TX_HANDLE = 1 << 2,
1005 LCT_CL_THREAD = 1 << 3,
1007 * A per-request session on a server, and a per-system-call session on
1010 LCT_SESSION = 1 << 4,
1013 * Set when at least one of keys, having values in this context has
1014 * non-NULL lu_context_key::lct_exit() method. This is used to
1015 * optimize lu_context_exit() call.
1017 LCT_HAS_EXIT = 1 << 28,
1019 * Don't add references for modules creating key values in that context.
1020 * This is only for contexts used internally by lu_object framework.
1022 LCT_NOREF = 1 << 29,
1024 * Key is being prepared for retiring, don't create new values for it.
1026 LCT_QUIESCENT = 1 << 30,
1028 * Context should be remembered.
1030 LCT_REMEMBER = 1 << 31,
1032 * Contexts usable in cache shrinker thread.
1034 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1038 * Key. Represents per-context value slot.
1040 * Keys are usually registered when module owning the key is initialized, and
1041 * de-registered when module is unloaded. Once key is registered, all new
1042 * contexts with matching tags, will get key value. "Old" contexts, already
1043 * initialized at the time of key registration, can be forced to get key value
1044 * by calling lu_context_refill().
1046 * Every key value is counted in lu_context_key::lct_used and acquires a
1047 * reference on an owning module. This means, that all key values have to be
1048 * destroyed before module can be unloaded. This is usually achieved by
1049 * stopping threads started by the module, that created contexts in their
1050 * entry functions. Situation is complicated by the threads shared by multiple
1051 * modules, like ptlrpcd daemon on a client. To work around this problem,
1052 * contexts, created in such threads, are `remembered' (see
1053 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1054 * for unloading it does the following:
1056 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1057 * preventing new key values from being allocated in the new contexts,
1060 * - scans a list of remembered contexts, destroying values of module
1061 * keys, thus releasing references to the module.
1063 * This is done by lu_context_key_quiesce(). If module is re-activated
1064 * before key has been de-registered, lu_context_key_revive() call clears
1065 * `quiescent' marker.
1067 * lu_context code doesn't provide any internal synchronization for these
1068 * activities---it's assumed that startup (including threads start-up) and
1069 * shutdown are serialized by some external means.
1073 struct lu_context_key {
1075 * Set of tags for which values of this key are to be instantiated.
1079 * Value constructor. This is called when new value is created for a
1080 * context. Returns pointer to new value of error pointer.
1082 void *(*lct_init)(const struct lu_context *ctx,
1083 struct lu_context_key *key);
1085 * Value destructor. Called when context with previously allocated
1086 * value of this slot is destroyed. \a data is a value that was returned
1087 * by a matching call to lu_context_key::lct_init().
1089 void (*lct_fini)(const struct lu_context *ctx,
1090 struct lu_context_key *key, void *data);
1092 * Optional method called on lu_context_exit() for all allocated
1093 * keys. Can be used by debugging code checking that locks are
1096 void (*lct_exit)(const struct lu_context *ctx,
1097 struct lu_context_key *key, void *data);
1099 * Internal implementation detail: index within lu_context::lc_value[]
1100 * reserved for this key.
1104 * Internal implementation detail: number of values created for this
1109 * Internal implementation detail: module for this key.
1111 struct module *lct_owner;
1113 * References to this key. For debugging.
1115 struct lu_ref lct_reference;
1118 #define LU_KEY_INIT(mod, type) \
1119 static void* mod##_key_init(const struct lu_context *ctx, \
1120 struct lu_context_key *key) \
1124 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1126 OBD_ALLOC_PTR(value); \
1127 if (value == NULL) \
1128 value = ERR_PTR(-ENOMEM); \
1132 struct __##mod##__dummy_init {;} /* semicolon catcher */
1134 #define LU_KEY_FINI(mod, type) \
1135 static void mod##_key_fini(const struct lu_context *ctx, \
1136 struct lu_context_key *key, void* data) \
1138 type *info = data; \
1140 OBD_FREE_PTR(info); \
1142 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1144 #define LU_KEY_INIT_FINI(mod, type) \
1145 LU_KEY_INIT(mod,type); \
1146 LU_KEY_FINI(mod,type)
1148 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1149 struct lu_context_key mod##_thread_key = { \
1151 .lct_init = mod##_key_init, \
1152 .lct_fini = mod##_key_fini \
1155 #define LU_CONTEXT_KEY_INIT(key) \
1157 (key)->lct_owner = THIS_MODULE; \
1160 int lu_context_key_register(struct lu_context_key *key);
1161 void lu_context_key_degister(struct lu_context_key *key);
1162 void *lu_context_key_get (const struct lu_context *ctx,
1163 const struct lu_context_key *key);
1164 void lu_context_key_quiesce (struct lu_context_key *key);
1165 void lu_context_key_revive (struct lu_context_key *key);
1169 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1173 #define LU_KEY_INIT_GENERIC(mod) \
1174 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1176 struct lu_context_key *key = k; \
1179 va_start(args, k); \
1181 LU_CONTEXT_KEY_INIT(key); \
1182 key = va_arg(args, struct lu_context_key *); \
1183 } while (key != NULL); \
1187 #define LU_TYPE_INIT(mod, ...) \
1188 LU_KEY_INIT_GENERIC(mod) \
1189 static int mod##_type_init(struct lu_device_type *t) \
1191 mod##_key_init_generic(__VA_ARGS__, NULL); \
1192 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1194 struct __##mod##_dummy_type_init {;}
1196 #define LU_TYPE_FINI(mod, ...) \
1197 static void mod##_type_fini(struct lu_device_type *t) \
1199 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1201 struct __##mod##_dummy_type_fini {;}
1203 #define LU_TYPE_START(mod, ...) \
1204 static void mod##_type_start(struct lu_device_type *t) \
1206 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1208 struct __##mod##_dummy_type_start {;}
1210 #define LU_TYPE_STOP(mod, ...) \
1211 static void mod##_type_stop(struct lu_device_type *t) \
1213 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1215 struct __##mod##_dummy_type_stop {;}
1219 #define LU_TYPE_INIT_FINI(mod, ...) \
1220 LU_TYPE_INIT(mod, __VA_ARGS__); \
1221 LU_TYPE_FINI(mod, __VA_ARGS__); \
1222 LU_TYPE_START(mod, __VA_ARGS__); \
1223 LU_TYPE_STOP(mod, __VA_ARGS__)
1225 int lu_context_init (struct lu_context *ctx, __u32 tags);
1226 void lu_context_fini (struct lu_context *ctx);
1227 void lu_context_enter (struct lu_context *ctx);
1228 void lu_context_exit (struct lu_context *ctx);
1229 int lu_context_refill(struct lu_context *ctx);
1232 * Helper functions to operate on multiple keys. These are used by the default
1233 * device type operations, defined by LU_TYPE_INIT_FINI().
1236 int lu_context_key_register_many(struct lu_context_key *k, ...);
1237 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1238 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1239 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1246 * "Local" context, used to store data instead of stack.
1248 struct lu_context le_ctx;
1250 * "Session" context for per-request data.
1252 struct lu_context *le_ses;
1255 int lu_env_init (struct lu_env *env, __u32 tags);
1256 void lu_env_fini (struct lu_env *env);
1257 int lu_env_refill(struct lu_env *env);
1259 /** @} lu_context */
1262 * Output site statistical counters into a buffer. Suitable for
1263 * ll_rd_*()-style functions.
1265 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1268 * Common name structure to be passed around for various name related methods.
1276 * Common buffer structure to be passed around for various xattr_{s,g}et()
1285 extern struct lu_buf LU_BUF_NULL;
1287 #define DLUBUF "(%p %z)"
1288 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1290 * One-time initializers, called at obdclass module initialization, not
1295 * Initialization of global lu_* data.
1297 int lu_global_init(void);
1300 * Dual to lu_global_init().
1302 void lu_global_fini(void);
1305 LU_TIME_FIND_LOOKUP,
1307 LU_TIME_FIND_INSERT,
1311 extern const char *lu_time_names[LU_TIME_NR];
1313 struct lu_kmem_descr {
1314 cfs_mem_cache_t **ckd_cache;
1315 const char *ckd_name;
1316 const size_t ckd_size;
1319 int lu_kmem_init(struct lu_kmem_descr *caches);
1320 void lu_kmem_fini(struct lu_kmem_descr *caches);
1324 #endif /* __LUSTRE_LU_OBJECT_H */