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;
297 * Link the device to the site.
299 cfs_list_t ld_linkage;
302 struct lu_device_type_operations;
305 * Tag bits for device type. They are used to distinguish certain groups of
309 /** this is meta-data device */
310 LU_DEVICE_MD = (1 << 0),
311 /** this is data device */
312 LU_DEVICE_DT = (1 << 1),
313 /** data device in the client stack */
314 LU_DEVICE_CL = (1 << 2)
320 struct lu_device_type {
322 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
326 * Name of this class. Unique system-wide. Never modified once set.
330 * Operations for this type.
332 const struct lu_device_type_operations *ldt_ops;
334 * \todo XXX: temporary pointer to associated obd_type.
336 struct obd_type *ldt_obd_type;
338 * \todo XXX: temporary: context tags used by obd_*() calls.
342 * Number of existing device type instances.
344 unsigned ldt_device_nr;
346 * Linkage into a global list of all device types.
348 * \see lu_device_types.
350 cfs_list_t ldt_linkage;
354 * Operations on a device type.
356 struct lu_device_type_operations {
358 * Allocate new device.
360 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
361 struct lu_device_type *t,
362 struct lustre_cfg *lcfg);
364 * Free device. Dual to
365 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
366 * the next device in the stack.
368 struct lu_device *(*ldto_device_free)(const struct lu_env *,
372 * Initialize the devices after allocation
374 int (*ldto_device_init)(const struct lu_env *env,
375 struct lu_device *, const char *,
378 * Finalize device. Dual to
379 * lu_device_type_operations::ldto_device_init(). Returns pointer to
380 * the next device in the stack.
382 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
385 * Initialize device type. This is called on module load.
387 int (*ldto_init)(struct lu_device_type *t);
389 * Finalize device type. Dual to
390 * lu_device_type_operations::ldto_init(). Called on module unload.
392 void (*ldto_fini)(struct lu_device_type *t);
394 * Called when the first device is created.
396 void (*ldto_start)(struct lu_device_type *t);
398 * Called when number of devices drops to 0.
400 void (*ldto_stop)(struct lu_device_type *t);
404 * Flags for the object layers.
406 enum lu_object_flags {
408 * this flags is set if lu_object_operations::loo_object_init() has
409 * been called for this layer. Used by lu_object_alloc().
411 LU_OBJECT_ALLOCATED = (1 << 0)
415 * Common object attributes.
420 /** modification time in seconds since Epoch */
422 /** access time in seconds since Epoch */
424 /** change time in seconds since Epoch */
426 /** 512-byte blocks allocated to object */
428 /** permission bits and file type */
436 /** number of persistent references to this object */
438 /** blk bits of the object*/
440 /** blk size of the object*/
452 /** Bit-mask of valid attributes */
466 LA_BLKSIZE = 1 << 12,
467 LA_KILL_SUID = 1 << 13,
468 LA_KILL_SGID = 1 << 14,
472 * Layer in the layered object.
476 * Header for this object.
478 struct lu_object_header *lo_header;
480 * Device for this layer.
482 struct lu_device *lo_dev;
484 * Operations for this object.
486 const struct lu_object_operations *lo_ops;
488 * Linkage into list of all layers.
490 cfs_list_t lo_linkage;
492 * Depth. Top level layer depth is 0.
496 * Flags from enum lu_object_flags.
498 unsigned long lo_flags;
500 * Link to the device, for debugging.
502 struct lu_ref_link *lo_dev_ref;
505 enum lu_object_header_flags {
507 * Don't keep this object in cache. Object will be destroyed as soon
508 * as last reference to it is released. This flag cannot be cleared
511 LU_OBJECT_HEARD_BANSHEE = 0
514 enum lu_object_header_attr {
515 LOHA_EXISTS = 1 << 0,
516 LOHA_REMOTE = 1 << 1,
518 * UNIX file type is stored in S_IFMT bits.
520 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
521 LOHA_FT_END = 017 << 12, /**< S_IFMT */
525 * "Compound" object, consisting of multiple layers.
527 * Compound object with given fid is unique with given lu_site.
529 * Note, that object does *not* necessary correspond to the real object in the
530 * persistent storage: object is an anchor for locking and method calling, so
531 * it is created for things like not-yet-existing child created by mkdir or
532 * create calls. lu_object_operations::loo_exists() can be used to check
533 * whether object is backed by persistent storage entity.
535 struct lu_object_header {
537 * Object flags from enum lu_object_header_flags. Set and checked
540 unsigned long loh_flags;
542 * Object reference count. Protected by lu_site::ls_guard.
544 cfs_atomic_t loh_ref;
546 * Fid, uniquely identifying this object.
548 struct lu_fid loh_fid;
550 * Common object attributes, cached for efficiency. From enum
551 * lu_object_header_attr.
555 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
557 cfs_hlist_node_t loh_hash;
559 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
563 * Linkage into list of layers. Never modified once set (except lately
564 * during object destruction). No locking is necessary.
566 cfs_list_t loh_layers;
568 * A list of references to this object, for debugging.
570 struct lu_ref loh_reference;
575 struct lu_site_bkt_data {
577 * number of busy object on this bucket
581 * LRU list, updated on each access to object. Protected by
582 * bucket lock of lu_site::ls_obj_hash.
584 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
585 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
586 * of list_for_each_entry_safe_reverse()).
590 * Wait-queue signaled when an object in this site is ultimately
591 * destroyed (lu_object_free()). It is used by lu_object_find() to
592 * wait before re-trying when object in the process of destruction is
593 * found in the hash table.
595 * \see htable_lookup().
597 cfs_waitq_t lsb_marche_funebre;
605 LU_SS_CACHE_DEATH_RACE,
611 * lu_site is a "compartment" within which objects are unique, and LRU
612 * discipline is maintained.
614 * lu_site exists so that multiple layered stacks can co-exist in the same
617 * lu_site has the same relation to lu_device as lu_object_header to
624 cfs_hash_t *ls_obj_hash;
626 * index of bucket on hash table while purging
630 * Top-level device for this stack.
632 struct lu_device *ls_top_dev;
634 * Linkage into global list of sites.
636 cfs_list_t ls_linkage;
638 * List for lu device for this site, protected
641 cfs_list_t ls_ld_linkage;
642 cfs_spinlock_t ls_ld_lock;
647 struct lprocfs_stats *ls_stats;
648 struct lprocfs_stats *ls_time_stats;
651 static inline struct lu_site_bkt_data *
652 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
656 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
657 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
661 * Constructors/destructors.
665 int lu_site_init (struct lu_site *s, struct lu_device *d);
666 void lu_site_fini (struct lu_site *s);
667 int lu_site_init_finish (struct lu_site *s);
668 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
669 void lu_device_get (struct lu_device *d);
670 void lu_device_put (struct lu_device *d);
671 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
672 void lu_device_fini (struct lu_device *d);
673 int lu_object_header_init(struct lu_object_header *h);
674 void lu_object_header_fini(struct lu_object_header *h);
675 int lu_object_init (struct lu_object *o,
676 struct lu_object_header *h, struct lu_device *d);
677 void lu_object_fini (struct lu_object *o);
678 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
679 void lu_object_add (struct lu_object *before, struct lu_object *o);
682 * Helpers to initialize and finalize device types.
685 int lu_device_type_init(struct lu_device_type *ldt);
686 void lu_device_type_fini(struct lu_device_type *ldt);
687 void lu_types_stop(void);
692 * Caching and reference counting.
697 * Acquire additional reference to the given object. This function is used to
698 * attain additional reference. To acquire initial reference use
701 static inline void lu_object_get(struct lu_object *o)
703 LASSERT(cfs_atomic_read(&o->lo_header->loh_ref) > 0);
704 cfs_atomic_inc(&o->lo_header->loh_ref);
708 * Return true of object will not be cached after last reference to it is
711 static inline int lu_object_is_dying(const struct lu_object_header *h)
713 return cfs_test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
716 void lu_object_put(const struct lu_env *env, struct lu_object *o);
718 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
720 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
721 lu_printer_t printer);
722 struct lu_object *lu_object_find(const struct lu_env *env,
723 struct lu_device *dev, const struct lu_fid *f,
724 const struct lu_object_conf *conf);
725 struct lu_object *lu_object_find_at(const struct lu_env *env,
726 struct lu_device *dev,
727 const struct lu_fid *f,
728 const struct lu_object_conf *conf);
729 struct lu_object *lu_object_find_slice(const struct lu_env *env,
730 struct lu_device *dev,
731 const struct lu_fid *f,
732 const struct lu_object_conf *conf);
741 * First (topmost) sub-object of given compound object
743 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
745 LASSERT(!cfs_list_empty(&h->loh_layers));
746 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
750 * Next sub-object in the layering
752 static inline struct lu_object *lu_object_next(const struct lu_object *o)
754 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
758 * Pointer to the fid of this object.
760 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
762 return &o->lo_header->loh_fid;
766 * return device operations vector for this object
768 static const inline struct lu_device_operations *
769 lu_object_ops(const struct lu_object *o)
771 return o->lo_dev->ld_ops;
775 * Given a compound object, find its slice, corresponding to the device type
778 struct lu_object *lu_object_locate(struct lu_object_header *h,
779 const struct lu_device_type *dtype);
781 struct lu_cdebug_print_info {
784 const char *lpi_file;
790 * Printer function emitting messages through libcfs_debug_msg().
792 int lu_cdebug_printer(const struct lu_env *env,
793 void *cookie, const char *format, ...);
795 #define DECLARE_LU_CDEBUG_PRINT_INFO(var, mask) \
796 struct lu_cdebug_print_info var = { \
797 .lpi_subsys = DEBUG_SUBSYSTEM, \
798 .lpi_mask = (mask), \
799 .lpi_file = __FILE__, \
800 .lpi_fn = __FUNCTION__, \
801 .lpi_line = __LINE__ \
805 * Print object description followed by a user-supplied message.
807 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
809 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
811 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
812 lu_object_print(env, &__info, lu_cdebug_printer, object); \
813 CDEBUG(mask, format , ## __VA_ARGS__); \
818 * Print short object description followed by a user-supplied message.
820 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
822 static DECLARE_LU_CDEBUG_PRINT_INFO(__info, mask); \
824 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
825 lu_object_header_print(env, &__info, lu_cdebug_printer, \
826 (object)->lo_header); \
827 lu_cdebug_printer(env, &__info, "\n"); \
828 CDEBUG(mask, format , ## __VA_ARGS__); \
832 void lu_object_print (const struct lu_env *env, void *cookie,
833 lu_printer_t printer, const struct lu_object *o);
834 void lu_object_header_print(const struct lu_env *env, void *cookie,
835 lu_printer_t printer,
836 const struct lu_object_header *hdr);
839 * Check object consistency.
841 int lu_object_invariant(const struct lu_object *o);
845 * \retval 1 iff object \a o exists on stable storage,
846 * \retval 0 iff object \a o not exists on stable storage.
847 * \retval -1 iff object \a o is on remote server.
849 static inline int lu_object_exists(const struct lu_object *o)
853 attr = o->lo_header->loh_attr;
854 if (attr & LOHA_REMOTE)
856 else if (attr & LOHA_EXISTS)
862 static inline int lu_object_assert_exists(const struct lu_object *o)
864 return lu_object_exists(o) != 0;
867 static inline int lu_object_assert_not_exists(const struct lu_object *o)
869 return lu_object_exists(o) <= 0;
873 * Attr of this object.
875 static inline __u32 lu_object_attr(const struct lu_object *o)
877 LASSERT(lu_object_exists(o) > 0);
878 return o->lo_header->loh_attr;
881 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
885 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
888 static inline void lu_object_ref_del(struct lu_object *o,
889 const char *scope, const void *source)
891 lu_ref_del(&o->lo_header->loh_reference, scope, source);
894 static inline void lu_object_ref_del_at(struct lu_object *o,
895 struct lu_ref_link *link,
896 const char *scope, const void *source)
898 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
901 /** input params, should be filled out by mdt */
905 /** count in bytes */
906 unsigned int rp_count;
907 /** number of pages */
908 unsigned int rp_npages;
909 /** requested attr */
911 /** pointers to pages */
912 struct page **rp_pages;
915 enum lu_xattr_flags {
916 LU_XATTR_REPLACE = (1 << 0),
917 LU_XATTR_CREATE = (1 << 1)
925 /** For lu_context health-checks */
926 enum lu_context_state {
934 * lu_context. Execution context for lu_object methods. Currently associated
937 * All lu_object methods, except device and device type methods (called during
938 * system initialization and shutdown) are executed "within" some
939 * lu_context. This means, that pointer to some "current" lu_context is passed
940 * as an argument to all methods.
942 * All service ptlrpc threads create lu_context as part of their
943 * initialization. It is possible to create "stand-alone" context for other
944 * execution environments (like system calls).
946 * lu_object methods mainly use lu_context through lu_context_key interface
947 * that allows each layer to associate arbitrary pieces of data with each
948 * context (see pthread_key_create(3) for similar interface).
950 * On a client, lu_context is bound to a thread, see cl_env_get().
952 * \see lu_context_key
956 * lu_context is used on the client side too. Yet we don't want to
957 * allocate values of server-side keys for the client contexts and
960 * To achieve this, set of tags in introduced. Contexts and keys are
961 * marked with tags. Key value are created only for context whose set
962 * of tags has non-empty intersection with one for key. Tags are taken
963 * from enum lu_context_tag.
967 * Pointer to the home service thread. NULL for other execution
970 struct ptlrpc_thread *lc_thread;
972 * Pointer to an array with key values. Internal implementation
976 enum lu_context_state lc_state;
978 * Linkage into a list of all remembered contexts. Only
979 * `non-transient' contexts, i.e., ones created for service threads
982 cfs_list_t lc_remember;
984 * Version counter used to skip calls to lu_context_refill() when no
985 * keys were registered.
995 * lu_context_key interface. Similar to pthread_key.
998 enum lu_context_tag {
1000 * Thread on md server
1002 LCT_MD_THREAD = 1 << 0,
1004 * Thread on dt server
1006 LCT_DT_THREAD = 1 << 1,
1008 * Context for transaction handle
1010 LCT_TX_HANDLE = 1 << 2,
1014 LCT_CL_THREAD = 1 << 3,
1016 * A per-request session on a server, and a per-system-call session on
1019 LCT_SESSION = 1 << 4,
1022 * Set when at least one of keys, having values in this context has
1023 * non-NULL lu_context_key::lct_exit() method. This is used to
1024 * optimize lu_context_exit() call.
1026 LCT_HAS_EXIT = 1 << 28,
1028 * Don't add references for modules creating key values in that context.
1029 * This is only for contexts used internally by lu_object framework.
1031 LCT_NOREF = 1 << 29,
1033 * Key is being prepared for retiring, don't create new values for it.
1035 LCT_QUIESCENT = 1 << 30,
1037 * Context should be remembered.
1039 LCT_REMEMBER = 1 << 31,
1041 * Contexts usable in cache shrinker thread.
1043 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1047 * Key. Represents per-context value slot.
1049 * Keys are usually registered when module owning the key is initialized, and
1050 * de-registered when module is unloaded. Once key is registered, all new
1051 * contexts with matching tags, will get key value. "Old" contexts, already
1052 * initialized at the time of key registration, can be forced to get key value
1053 * by calling lu_context_refill().
1055 * Every key value is counted in lu_context_key::lct_used and acquires a
1056 * reference on an owning module. This means, that all key values have to be
1057 * destroyed before module can be unloaded. This is usually achieved by
1058 * stopping threads started by the module, that created contexts in their
1059 * entry functions. Situation is complicated by the threads shared by multiple
1060 * modules, like ptlrpcd daemon on a client. To work around this problem,
1061 * contexts, created in such threads, are `remembered' (see
1062 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1063 * for unloading it does the following:
1065 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1066 * preventing new key values from being allocated in the new contexts,
1069 * - scans a list of remembered contexts, destroying values of module
1070 * keys, thus releasing references to the module.
1072 * This is done by lu_context_key_quiesce(). If module is re-activated
1073 * before key has been de-registered, lu_context_key_revive() call clears
1074 * `quiescent' marker.
1076 * lu_context code doesn't provide any internal synchronization for these
1077 * activities---it's assumed that startup (including threads start-up) and
1078 * shutdown are serialized by some external means.
1082 struct lu_context_key {
1084 * Set of tags for which values of this key are to be instantiated.
1088 * Value constructor. This is called when new value is created for a
1089 * context. Returns pointer to new value of error pointer.
1091 void *(*lct_init)(const struct lu_context *ctx,
1092 struct lu_context_key *key);
1094 * Value destructor. Called when context with previously allocated
1095 * value of this slot is destroyed. \a data is a value that was returned
1096 * by a matching call to lu_context_key::lct_init().
1098 void (*lct_fini)(const struct lu_context *ctx,
1099 struct lu_context_key *key, void *data);
1101 * Optional method called on lu_context_exit() for all allocated
1102 * keys. Can be used by debugging code checking that locks are
1105 void (*lct_exit)(const struct lu_context *ctx,
1106 struct lu_context_key *key, void *data);
1108 * Internal implementation detail: index within lu_context::lc_value[]
1109 * reserved for this key.
1113 * Internal implementation detail: number of values created for this
1116 cfs_atomic_t lct_used;
1118 * Internal implementation detail: module for this key.
1120 cfs_module_t *lct_owner;
1122 * References to this key. For debugging.
1124 struct lu_ref lct_reference;
1127 #define LU_KEY_INIT(mod, type) \
1128 static void* mod##_key_init(const struct lu_context *ctx, \
1129 struct lu_context_key *key) \
1133 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1135 OBD_ALLOC_PTR(value); \
1136 if (value == NULL) \
1137 value = ERR_PTR(-ENOMEM); \
1141 struct __##mod##__dummy_init {;} /* semicolon catcher */
1143 #define LU_KEY_FINI(mod, type) \
1144 static void mod##_key_fini(const struct lu_context *ctx, \
1145 struct lu_context_key *key, void* data) \
1147 type *info = data; \
1149 OBD_FREE_PTR(info); \
1151 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1153 #define LU_KEY_INIT_FINI(mod, type) \
1154 LU_KEY_INIT(mod,type); \
1155 LU_KEY_FINI(mod,type)
1157 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1158 struct lu_context_key mod##_thread_key = { \
1160 .lct_init = mod##_key_init, \
1161 .lct_fini = mod##_key_fini \
1164 #define LU_CONTEXT_KEY_INIT(key) \
1166 (key)->lct_owner = THIS_MODULE; \
1169 int lu_context_key_register(struct lu_context_key *key);
1170 void lu_context_key_degister(struct lu_context_key *key);
1171 void *lu_context_key_get (const struct lu_context *ctx,
1172 const struct lu_context_key *key);
1173 void lu_context_key_quiesce (struct lu_context_key *key);
1174 void lu_context_key_revive (struct lu_context_key *key);
1178 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1182 #define LU_KEY_INIT_GENERIC(mod) \
1183 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1185 struct lu_context_key *key = k; \
1188 va_start(args, k); \
1190 LU_CONTEXT_KEY_INIT(key); \
1191 key = va_arg(args, struct lu_context_key *); \
1192 } while (key != NULL); \
1196 #define LU_TYPE_INIT(mod, ...) \
1197 LU_KEY_INIT_GENERIC(mod) \
1198 static int mod##_type_init(struct lu_device_type *t) \
1200 mod##_key_init_generic(__VA_ARGS__, NULL); \
1201 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1203 struct __##mod##_dummy_type_init {;}
1205 #define LU_TYPE_FINI(mod, ...) \
1206 static void mod##_type_fini(struct lu_device_type *t) \
1208 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1210 struct __##mod##_dummy_type_fini {;}
1212 #define LU_TYPE_START(mod, ...) \
1213 static void mod##_type_start(struct lu_device_type *t) \
1215 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1217 struct __##mod##_dummy_type_start {;}
1219 #define LU_TYPE_STOP(mod, ...) \
1220 static void mod##_type_stop(struct lu_device_type *t) \
1222 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1224 struct __##mod##_dummy_type_stop {;}
1228 #define LU_TYPE_INIT_FINI(mod, ...) \
1229 LU_TYPE_INIT(mod, __VA_ARGS__); \
1230 LU_TYPE_FINI(mod, __VA_ARGS__); \
1231 LU_TYPE_START(mod, __VA_ARGS__); \
1232 LU_TYPE_STOP(mod, __VA_ARGS__)
1234 int lu_context_init (struct lu_context *ctx, __u32 tags);
1235 void lu_context_fini (struct lu_context *ctx);
1236 void lu_context_enter (struct lu_context *ctx);
1237 void lu_context_exit (struct lu_context *ctx);
1238 int lu_context_refill(struct lu_context *ctx);
1241 * Helper functions to operate on multiple keys. These are used by the default
1242 * device type operations, defined by LU_TYPE_INIT_FINI().
1245 int lu_context_key_register_many(struct lu_context_key *k, ...);
1246 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1247 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1248 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1251 * update/clear ctx/ses tags.
1253 void lu_context_tags_update(__u32 tags);
1254 void lu_context_tags_clear(__u32 tags);
1255 void lu_session_tags_update(__u32 tags);
1256 void lu_session_tags_clear(__u32 tags);
1263 * "Local" context, used to store data instead of stack.
1265 struct lu_context le_ctx;
1267 * "Session" context for per-request data.
1269 struct lu_context *le_ses;
1272 int lu_env_init (struct lu_env *env, __u32 tags);
1273 void lu_env_fini (struct lu_env *env);
1274 int lu_env_refill(struct lu_env *env);
1275 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1277 /** @} lu_context */
1280 * Output site statistical counters into a buffer. Suitable for
1281 * ll_rd_*()-style functions.
1283 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1286 * Common name structure to be passed around for various name related methods.
1289 const char *ln_name;
1294 * Common buffer structure to be passed around for various xattr_{s,g}et()
1303 extern struct lu_buf LU_BUF_NULL;
1305 #define DLUBUF "(%p %z)"
1306 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1308 * One-time initializers, called at obdclass module initialization, not
1313 * Initialization of global lu_* data.
1315 int lu_global_init(void);
1318 * Dual to lu_global_init().
1320 void lu_global_fini(void);
1323 LU_TIME_FIND_LOOKUP,
1325 LU_TIME_FIND_INSERT,
1329 extern const char *lu_time_names[LU_TIME_NR];
1331 struct lu_kmem_descr {
1332 cfs_mem_cache_t **ckd_cache;
1333 const char *ckd_name;
1334 const size_t ckd_size;
1337 int lu_kmem_init(struct lu_kmem_descr *caches);
1338 void lu_kmem_fini(struct lu_kmem_descr *caches);
1341 #endif /* __LUSTRE_LU_OBJECT_H */