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 *,
160 * initialize local objects for device. this method called after layer has
161 * been initialized (after LCFG_SETUP stage) and before it starts serving
165 int (*ldo_prepare)(const struct lu_env *,
166 struct lu_device *parent,
167 struct lu_device *dev);
172 * Object configuration, describing particulars of object being created. On
173 * server this is not used, as server objects are full identified by fid. On
174 * client configuration contains struct lustre_md.
176 struct lu_object_conf {
180 * Type of "printer" function used by lu_object_operations::loo_object_print()
183 * Printer function is needed to provide some flexibility in (semi-)debugging
184 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
186 typedef int (*lu_printer_t)(const struct lu_env *env,
187 void *cookie, const char *format, ...)
188 __attribute__ ((format (printf, 3, 4)));
191 * Operations specific for particular lu_object.
193 struct lu_object_operations {
196 * Allocate lower-layer parts of the object by calling
197 * lu_device_operations::ldo_object_alloc() of the corresponding
200 * This method is called once for each object inserted into object
201 * stack. It's responsibility of this method to insert lower-layer
202 * object(s) it create into appropriate places of object stack.
204 int (*loo_object_init)(const struct lu_env *env,
206 const struct lu_object_conf *conf);
208 * Called (in top-to-bottom order) during object allocation after all
209 * layers were allocated and initialized. Can be used to perform
210 * initialization depending on lower layers.
212 int (*loo_object_start)(const struct lu_env *env,
213 struct lu_object *o);
215 * Called before lu_object_operations::loo_object_free() to signal
216 * that object is being destroyed. Dual to
217 * lu_object_operations::loo_object_init().
219 void (*loo_object_delete)(const struct lu_env *env,
220 struct lu_object *o);
222 * Dual to lu_device_operations::ldo_object_alloc(). Called when
223 * object is removed from memory.
225 void (*loo_object_free)(const struct lu_env *env,
226 struct lu_object *o);
228 * Called when last active reference to the object is released (and
229 * object returns to the cache). This method is optional.
231 void (*loo_object_release)(const struct lu_env *env,
232 struct lu_object *o);
234 * Optional debugging helper. Print given object.
236 int (*loo_object_print)(const struct lu_env *env, void *cookie,
237 lu_printer_t p, const struct lu_object *o);
239 * Optional debugging method. Returns true iff method is internally
242 int (*loo_object_invariant)(const struct lu_object *o);
248 struct lu_device_type;
251 * Device: a layer in the server side abstraction stacking.
255 * reference count. This is incremented, in particular, on each object
256 * created at this layer.
258 * \todo XXX which means that atomic_t is probably too small.
262 * Pointer to device type. Never modified once set.
264 struct lu_device_type *ld_type;
266 * Operation vector for this device.
268 const struct lu_device_operations *ld_ops;
270 * Stack this device belongs to.
272 struct lu_site *ld_site;
273 struct proc_dir_entry *ld_proc_entry;
275 /** \todo XXX: temporary back pointer into obd. */
276 struct obd_device *ld_obd;
278 * A list of references to this object, for debugging.
280 struct lu_ref ld_reference;
283 struct lu_device_type_operations;
286 * Tag bits for device type. They are used to distinguish certain groups of
290 /** this is meta-data device */
291 LU_DEVICE_MD = (1 << 0),
292 /** this is data device */
293 LU_DEVICE_DT = (1 << 1),
294 /** data device in the client stack */
295 LU_DEVICE_CL = (1 << 2)
301 struct lu_device_type {
303 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
307 * Name of this class. Unique system-wide. Never modified once set.
311 * Operations for this type.
313 const struct lu_device_type_operations *ldt_ops;
315 * \todo XXX: temporary pointer to associated obd_type.
317 struct obd_type *ldt_obd_type;
319 * \todo XXX: temporary: context tags used by obd_*() calls.
323 * Number of existing device type instances.
325 unsigned ldt_device_nr;
327 * Linkage into a global list of all device types.
329 * \see lu_device_types.
331 struct list_head ldt_linkage;
335 * Operations on a device type.
337 struct lu_device_type_operations {
339 * Allocate new device.
341 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
342 struct lu_device_type *t,
343 struct lustre_cfg *lcfg);
345 * Free device. Dual to
346 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
347 * the next device in the stack.
349 struct lu_device *(*ldto_device_free)(const struct lu_env *,
353 * Initialize the devices after allocation
355 int (*ldto_device_init)(const struct lu_env *env,
356 struct lu_device *, const char *,
359 * Finalize device. Dual to
360 * lu_device_type_operations::ldto_device_init(). Returns pointer to
361 * the next device in the stack.
363 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
366 * Initialize device type. This is called on module load.
368 int (*ldto_init)(struct lu_device_type *t);
370 * Finalize device type. Dual to
371 * lu_device_type_operations::ldto_init(). Called on module unload.
373 void (*ldto_fini)(struct lu_device_type *t);
375 * Called when the first device is created.
377 void (*ldto_start)(struct lu_device_type *t);
379 * Called when number of devices drops to 0.
381 void (*ldto_stop)(struct lu_device_type *t);
385 * Flags for the object layers.
387 enum lu_object_flags {
389 * this flags is set if lu_object_operations::loo_object_init() has
390 * been called for this layer. Used by lu_object_alloc().
392 LU_OBJECT_ALLOCATED = (1 << 0)
396 * Common object attributes.
401 /** modification time in seconds since Epoch */
403 /** access time in seconds since Epoch */
405 /** change time in seconds since Epoch */
407 /** 512-byte blocks allocated to object */
409 /** permission bits and file type */
417 /** number of persistent references to this object */
419 /** blk bits of the object*/
421 /** blk size of the object*/
433 /** Bit-mask of valid attributes */
447 LA_BLKSIZE = 1 << 12,
451 * Layer in the layered object.
455 * Header for this object.
457 struct lu_object_header *lo_header;
459 * Device for this layer.
461 struct lu_device *lo_dev;
463 * Operations for this object.
465 const struct lu_object_operations *lo_ops;
467 * Linkage into list of all layers.
469 struct list_head lo_linkage;
471 * Depth. Top level layer depth is 0.
475 * Flags from enum lu_object_flags.
477 unsigned long lo_flags;
479 * Link to the device, for debugging.
481 struct lu_ref_link *lo_dev_ref;
484 enum lu_object_header_flags {
486 * Don't keep this object in cache. Object will be destroyed as soon
487 * as last reference to it is released. This flag cannot be cleared
490 LU_OBJECT_HEARD_BANSHEE = 0
493 enum lu_object_header_attr {
494 LOHA_EXISTS = 1 << 0,
495 LOHA_REMOTE = 1 << 1,
497 * UNIX file type is stored in S_IFMT bits.
499 LOHA_FT_START = 1 << 12, /**< S_IFIFO */
500 LOHA_FT_END = 1 << 15, /**< S_IFREG */
504 * "Compound" object, consisting of multiple layers.
506 * Compound object with given fid is unique with given lu_site.
508 * Note, that object does *not* necessary correspond to the real object in the
509 * persistent storage: object is an anchor for locking and method calling, so
510 * it is created for things like not-yet-existing child created by mkdir or
511 * create calls. lu_object_operations::loo_exists() can be used to check
512 * whether object is backed by persistent storage entity.
514 struct lu_object_header {
516 * Object flags from enum lu_object_header_flags. Set and checked
519 unsigned long loh_flags;
521 * Object reference count. Protected by lu_site::ls_guard.
525 * Fid, uniquely identifying this object.
527 struct lu_fid loh_fid;
529 * Common object attributes, cached for efficiency. From enum
530 * lu_object_header_attr.
534 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
536 struct hlist_node loh_hash;
538 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
540 struct list_head loh_lru;
542 * Linkage into list of layers. Never modified once set (except lately
543 * during object destruction). No locking is necessary.
545 struct list_head loh_layers;
547 * A list of references to this object, for debugging.
549 struct lu_ref loh_reference;
555 * lu_site is a "compartment" within which objects are unique, and LRU
556 * discipline is maintained.
558 * lu_site exists so that multiple layered stacks can co-exist in the same
561 * lu_site has the same relation to lu_device as lu_object_header to
570 * - lu_site::ls_hash hash table (and its linkages in objects);
572 * - lu_site::ls_lru list (and its linkages in objects);
574 * - 0/1 transitions of object lu_object_header::loh_ref
581 * Hash-table where objects are indexed by fid.
583 struct hlist_head *ls_hash;
585 * Bit-mask for hash-table size.
589 * Order of hash-table.
593 * Number of buckets in the hash-table.
598 * LRU list, updated on each access to object. Protected by
601 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
602 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
603 * of list_for_each_entry_safe_reverse()).
605 struct list_head ls_lru;
607 * Total number of objects in this site. Protected by
612 * Total number of objects in this site with reference counter greater
613 * than 0. Protected by lu_site::ls_guard.
618 * Top-level device for this stack.
620 struct lu_device *ls_top_dev;
623 * Wait-queue signaled when an object in this site is ultimately
624 * destroyed (lu_object_free()). It is used by lu_object_find() to
625 * wait before re-trying when object in the process of destruction is
626 * found in the hash table.
628 * If having a single wait-queue turns out to be a problem, a
629 * wait-queue per hash-table bucket can be easily implemented.
631 * \see htable_lookup().
633 cfs_waitq_t ls_marche_funebre;
635 /** statistical counters. Protected by nothing, races are accepted. */
641 * Number of hash-table entry checks made.
643 * ->s_cache_check / (->s_cache_miss + ->s_cache_hit)
645 * is an average number of hash slots inspected during single
649 /** Races with cache insertions. */
652 * Races with object destruction.
654 * \see lu_site::ls_marche_funebre.
656 __u32 s_cache_death_race;
661 * Linkage into global list of sites.
663 struct list_head ls_linkage;
664 struct lprocfs_stats *ls_time_stats;
668 * Constructors/destructors.
672 int lu_site_init (struct lu_site *s, struct lu_device *d);
673 void lu_site_fini (struct lu_site *s);
674 int lu_site_init_finish (struct lu_site *s);
675 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
676 void lu_device_get (struct lu_device *d);
677 void lu_device_put (struct lu_device *d);
678 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
679 void lu_device_fini (struct lu_device *d);
680 int lu_object_header_init(struct lu_object_header *h);
681 void lu_object_header_fini(struct lu_object_header *h);
682 int lu_object_init (struct lu_object *o,
683 struct lu_object_header *h, struct lu_device *d);
684 void lu_object_fini (struct lu_object *o);
685 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
686 void lu_object_add (struct lu_object *before, struct lu_object *o);
689 * Helpers to initialize and finalize device types.
692 int lu_device_type_init(struct lu_device_type *ldt);
693 void lu_device_type_fini(struct lu_device_type *ldt);
694 void lu_types_stop(void);
699 * Caching and reference counting.
704 * Acquire additional reference to the given object. This function is used to
705 * attain additional reference. To acquire initial reference use
708 static inline void lu_object_get(struct lu_object *o)
710 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
711 atomic_inc(&o->lo_header->loh_ref);
715 * Return true of object will not be cached after last reference to it is
718 static inline int lu_object_is_dying(const struct lu_object_header *h)
720 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
723 void lu_object_put(const struct lu_env *env, struct lu_object *o);
725 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
727 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
728 lu_printer_t printer);
729 struct lu_object *lu_object_find(const struct lu_env *env,
730 struct lu_device *dev, const struct lu_fid *f,
731 const struct lu_object_conf *conf);
732 struct lu_object *lu_object_find_at(const struct lu_env *env,
733 struct lu_device *dev,
734 const struct lu_fid *f,
735 const struct lu_object_conf *conf);
736 struct lu_object *lu_object_find_slice(const struct lu_env *env,
737 struct lu_device *dev,
738 const struct lu_fid *f,
739 const struct lu_object_conf *conf);
748 * First (topmost) sub-object of given compound object
750 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
752 LASSERT(!list_empty(&h->loh_layers));
753 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
757 * Next sub-object in the layering
759 static inline struct lu_object *lu_object_next(const struct lu_object *o)
761 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
765 * Pointer to the fid of this object.
767 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
769 return &o->lo_header->loh_fid;
773 * return device operations vector for this object
775 static const inline struct lu_device_operations *
776 lu_object_ops(const struct lu_object *o)
778 return o->lo_dev->ld_ops;
782 * Given a compound object, find its slice, corresponding to the device type
785 struct lu_object *lu_object_locate(struct lu_object_header *h,
786 const struct lu_device_type *dtype);
788 struct lu_cdebug_print_info {
791 const char *lpi_file;
797 * Printer function emitting messages through libcfs_debug_msg().
799 int lu_cdebug_printer(const struct lu_env *env,
800 void *cookie, const char *format, ...);
802 #define DECLARE_LU_CDEBUG_PRINT_INFO(var, mask) \
803 struct lu_cdebug_print_info var = { \
804 .lpi_subsys = DEBUG_SUBSYSTEM, \
805 .lpi_mask = (mask), \
806 .lpi_file = __FILE__, \
807 .lpi_fn = __FUNCTION__, \
808 .lpi_line = __LINE__ \
812 * Print object description followed by a user-supplied message.
814 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
816 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
818 if (cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
819 lu_object_print(env, &__info, lu_cdebug_printer, object); \
820 CDEBUG(mask, format , ## __VA_ARGS__); \
825 * Print short object description followed by a user-supplied message.
827 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
829 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
831 if (cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
832 lu_object_header_print(env, &__info, lu_cdebug_printer, \
833 (object)->lo_header); \
834 lu_cdebug_printer(env, &__info, "\n"); \
835 CDEBUG(mask, format , ## __VA_ARGS__); \
839 void lu_object_print (const struct lu_env *env, void *cookie,
840 lu_printer_t printer, const struct lu_object *o);
841 void lu_object_header_print(const struct lu_env *env, void *cookie,
842 lu_printer_t printer,
843 const struct lu_object_header *hdr);
846 * Check object consistency.
848 int lu_object_invariant(const struct lu_object *o);
850 void lu_stack_fini(const struct lu_env *env, struct lu_device *top);
853 * Returns 1 iff object @o exists on the stable storage,
854 * returns -1 iff object @o is on remote server.
856 static inline int lu_object_exists(const struct lu_object *o)
860 attr = o->lo_header->loh_attr;
861 if (attr & LOHA_REMOTE)
863 else if (attr & LOHA_EXISTS)
869 static inline int lu_object_assert_exists(const struct lu_object *o)
871 return lu_object_exists(o) != 0;
874 static inline int lu_object_assert_not_exists(const struct lu_object *o)
876 return lu_object_exists(o) <= 0;
880 * Attr of this object.
882 static inline __u32 lu_object_attr(const struct lu_object *o)
884 LASSERT(lu_object_exists(o) > 0);
885 return o->lo_header->loh_attr;
888 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
892 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
895 static inline void lu_object_ref_del(struct lu_object *o,
896 const char *scope, const void *source)
898 lu_ref_del(&o->lo_header->loh_reference, scope, source);
901 static inline void lu_object_ref_del_at(struct lu_object *o,
902 struct lu_ref_link *link,
903 const char *scope, const void *source)
905 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
909 /* input params, should be filled out by mdt */
910 __u64 rp_hash; /* hash */
911 int rp_count; /* count in bytes */
912 int rp_npages; /* number of pages */
913 struct page **rp_pages; /* pointers to pages */
916 enum lu_xattr_flags {
917 LU_XATTR_REPLACE = (1 << 0),
918 LU_XATTR_CREATE = (1 << 1)
926 /** For lu_context health-checks */
927 enum lu_context_state {
935 * lu_context. Execution context for lu_object methods. Currently associated
938 * All lu_object methods, except device and device type methods (called during
939 * system initialization and shutdown) are executed "within" some
940 * lu_context. This means, that pointer to some "current" lu_context is passed
941 * as an argument to all methods.
943 * All service ptlrpc threads create lu_context as part of their
944 * initialization. It is possible to create "stand-alone" context for other
945 * execution environments (like system calls).
947 * lu_object methods mainly use lu_context through lu_context_key interface
948 * that allows each layer to associate arbitrary pieces of data with each
949 * context (see pthread_key_create(3) for similar interface).
951 * On a client, lu_context is bound to a thread, see cl_env_get().
953 * \see lu_context_key
957 * lu_context is used on the client side too. Yet we don't want to
958 * allocate values of server-side keys for the client contexts and
961 * To achieve this, set of tags in introduced. Contexts and keys are
962 * marked with tags. Key value are created only for context whose set
963 * of tags has non-empty intersection with one for key. Tags are taken
964 * from enum lu_context_tag.
968 * Pointer to the home service thread. NULL for other execution
971 struct ptlrpc_thread *lc_thread;
973 * Pointer to an array with key values. Internal implementation
977 enum lu_context_state lc_state;
979 * Linkage into a list of all remembered contexts. Only
980 * `non-transient' contexts, i.e., ones created for service threads
983 struct list_head lc_remember;
985 * Version counter used to skip calls to lu_context_refill() when no
986 * keys were registered.
996 * lu_context_key interface. Similar to pthread_key.
999 enum lu_context_tag {
1001 * Thread on md server
1003 LCT_MD_THREAD = 1 << 0,
1005 * Thread on dt server
1007 LCT_DT_THREAD = 1 << 1,
1009 * Context for transaction handle
1011 LCT_TX_HANDLE = 1 << 2,
1015 LCT_CL_THREAD = 1 << 3,
1017 * A per-request session on a server, and a per-system-call session on
1020 LCT_SESSION = 1 << 4,
1023 * Set when at least one of keys, having values in this context has
1024 * non-NULL lu_context_key::lct_exit() method. This is used to
1025 * optimize lu_context_exit() call.
1027 LCT_HAS_EXIT = 1 << 28,
1029 * Don't add references for modules creating key values in that context.
1030 * This is only for contexts used internally by lu_object framework.
1032 LCT_NOREF = 1 << 29,
1034 * Key is being prepared for retiring, don't create new values for it.
1036 LCT_QUIESCENT = 1 << 30,
1038 * Context should be remembered.
1040 LCT_REMEMBER = 1 << 31,
1042 * Contexts usable in cache shrinker thread.
1044 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1048 * Key. Represents per-context value slot.
1050 * Keys are usually registered when module owning the key is initialized, and
1051 * de-registered when module is unloaded. Once key is registered, all new
1052 * contexts with matching tags, will get key value. "Old" contexts, already
1053 * initialized at the time of key registration, can be forced to get key value
1054 * by calling lu_context_refill().
1056 * Every key value is counted in lu_context_key::lct_used and acquires a
1057 * reference on an owning module. This means, that all key values have to be
1058 * destroyed before module can be unloaded. This is usually achieved by
1059 * stopping threads started by the module, that created contexts in their
1060 * entry functions. Situation is complicated by the threads shared by multiple
1061 * modules, like ptlrpcd daemon on a client. To work around this problem,
1062 * contexts, created in such threads, are `remembered' (see
1063 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1064 * for unloading it does the following:
1066 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1067 * preventing new key values from being allocated in the new contexts,
1070 * - scans a list of remembered contexts, destroying values of module
1071 * keys, thus releasing references to the module.
1073 * This is done by lu_context_key_quiesce(). If module is re-activated
1074 * before key has been de-registered, lu_context_key_revive() call clears
1075 * `quiescent' marker.
1077 * lu_context code doesn't provide any internal synchronization for these
1078 * activities---it's assumed that startup (including threads start-up) and
1079 * shutdown are serialized by some external means.
1083 struct lu_context_key {
1085 * Set of tags for which values of this key are to be instantiated.
1089 * Value constructor. This is called when new value is created for a
1090 * context. Returns pointer to new value of error pointer.
1092 void *(*lct_init)(const struct lu_context *ctx,
1093 struct lu_context_key *key);
1095 * Value destructor. Called when context with previously allocated
1096 * value of this slot is destroyed. \a data is a value that was returned
1097 * by a matching call to lu_context_key::lct_init().
1099 void (*lct_fini)(const struct lu_context *ctx,
1100 struct lu_context_key *key, void *data);
1102 * Optional method called on lu_context_exit() for all allocated
1103 * keys. Can be used by debugging code checking that locks are
1106 void (*lct_exit)(const struct lu_context *ctx,
1107 struct lu_context_key *key, void *data);
1109 * Internal implementation detail: index within lu_context::lc_value[]
1110 * reserved for this key.
1114 * Internal implementation detail: number of values created for this
1119 * Internal implementation detail: module for this key.
1121 struct module *lct_owner;
1123 * References to this key. For debugging.
1125 struct lu_ref lct_reference;
1128 #define LU_KEY_INIT(mod, type) \
1129 static void* mod##_key_init(const struct lu_context *ctx, \
1130 struct lu_context_key *key) \
1134 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1136 OBD_ALLOC_PTR(value); \
1137 if (value == NULL) \
1138 value = ERR_PTR(-ENOMEM); \
1142 struct __##mod##__dummy_init {;} /* semicolon catcher */
1144 #define LU_KEY_FINI(mod, type) \
1145 static void mod##_key_fini(const struct lu_context *ctx, \
1146 struct lu_context_key *key, void* data) \
1148 type *info = data; \
1150 OBD_FREE_PTR(info); \
1152 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1154 #define LU_KEY_INIT_FINI(mod, type) \
1155 LU_KEY_INIT(mod,type); \
1156 LU_KEY_FINI(mod,type)
1158 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1159 struct lu_context_key mod##_thread_key = { \
1161 .lct_init = mod##_key_init, \
1162 .lct_fini = mod##_key_fini \
1165 #define LU_CONTEXT_KEY_INIT(key) \
1167 (key)->lct_owner = THIS_MODULE; \
1170 int lu_context_key_register(struct lu_context_key *key);
1171 void lu_context_key_degister(struct lu_context_key *key);
1172 void *lu_context_key_get (const struct lu_context *ctx,
1173 const struct lu_context_key *key);
1174 void lu_context_key_quiesce (struct lu_context_key *key);
1175 void lu_context_key_revive (struct lu_context_key *key);
1179 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1183 #define LU_KEY_INIT_GENERIC(mod) \
1184 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1186 struct lu_context_key *key = k; \
1189 va_start(args, k); \
1191 LU_CONTEXT_KEY_INIT(key); \
1192 key = va_arg(args, struct lu_context_key *); \
1193 } while (key != NULL); \
1197 #define LU_TYPE_INIT(mod, ...) \
1198 LU_KEY_INIT_GENERIC(mod) \
1199 static int mod##_type_init(struct lu_device_type *t) \
1201 mod##_key_init_generic(__VA_ARGS__, NULL); \
1202 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1204 struct __##mod##_dummy_type_init {;}
1206 #define LU_TYPE_FINI(mod, ...) \
1207 static void mod##_type_fini(struct lu_device_type *t) \
1209 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1211 struct __##mod##_dummy_type_fini {;}
1213 #define LU_TYPE_START(mod, ...) \
1214 static void mod##_type_start(struct lu_device_type *t) \
1216 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1218 struct __##mod##_dummy_type_start {;}
1220 #define LU_TYPE_STOP(mod, ...) \
1221 static void mod##_type_stop(struct lu_device_type *t) \
1223 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1225 struct __##mod##_dummy_type_stop {;}
1229 #define LU_TYPE_INIT_FINI(mod, ...) \
1230 LU_TYPE_INIT(mod, __VA_ARGS__); \
1231 LU_TYPE_FINI(mod, __VA_ARGS__); \
1232 LU_TYPE_START(mod, __VA_ARGS__); \
1233 LU_TYPE_STOP(mod, __VA_ARGS__)
1235 int lu_context_init (struct lu_context *ctx, __u32 tags);
1236 void lu_context_fini (struct lu_context *ctx);
1237 void lu_context_enter (struct lu_context *ctx);
1238 void lu_context_exit (struct lu_context *ctx);
1239 int lu_context_refill(struct lu_context *ctx);
1242 * Helper functions to operate on multiple keys. These are used by the default
1243 * device type operations, defined by LU_TYPE_INIT_FINI().
1246 int lu_context_key_register_many(struct lu_context_key *k, ...);
1247 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1248 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1249 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1256 * "Local" context, used to store data instead of stack.
1258 struct lu_context le_ctx;
1260 * "Session" context for per-request data.
1262 struct lu_context *le_ses;
1265 int lu_env_init (struct lu_env *env, __u32 tags);
1266 void lu_env_fini (struct lu_env *env);
1267 int lu_env_refill(struct lu_env *env);
1269 /** @} lu_context */
1272 * Output site statistical counters into a buffer. Suitable for
1273 * ll_rd_*()-style functions.
1275 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1278 * Common name structure to be passed around for various name related methods.
1281 const char *ln_name;
1286 * Common buffer structure to be passed around for various xattr_{s,g}et()
1296 extern struct lu_buf LU_BUF_NULL;
1298 #define DLUBUF "(%p %z)"
1299 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1301 * One-time initializers, called at obdclass module initialization, not
1306 * Initialization of global lu_* data.
1308 int lu_global_init(void);
1311 * Dual to lu_global_init().
1313 void lu_global_fini(void);
1316 LU_TIME_FIND_LOOKUP,
1318 LU_TIME_FIND_INSERT,
1322 extern const char *lu_time_names[LU_TIME_NR];
1324 struct lu_kmem_descr {
1325 cfs_mem_cache_t **ckd_cache;
1326 const char *ckd_name;
1327 const size_t ckd_size;
1330 int lu_kmem_init(struct lu_kmem_descr *caches);
1331 void lu_kmem_fini(struct lu_kmem_descr *caches);
1334 #endif /* __LUSTRE_LU_OBJECT_H */