4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef __LUSTRE_LU_OBJECT_H
34 #define __LUSTRE_LU_OBJECT_H
37 #include "../../include/linux/libcfs/libcfs.h"
38 #include "lustre/lustre_idl.h"
46 * lu_* data-types represent server-side entities shared by data and meta-data
51 * -# support for layering.
53 * Server side object is split into layers, one per device in the
54 * corresponding device stack. Individual layer is represented by struct
55 * lu_object. Compound layered object --- by struct lu_object_header. Most
56 * interface functions take lu_object as an argument and operate on the
57 * whole compound object. This decision was made due to the following
60 * - it's envisaged that lu_object will be used much more often than
63 * - we want lower (non-top) layers to be able to initiate operations
64 * on the whole object.
66 * Generic code supports layering more complex than simple stacking, e.g.,
67 * it is possible that at some layer object "spawns" multiple sub-objects
70 * -# fid-based identification.
72 * Compound object is uniquely identified by its fid. Objects are indexed
73 * by their fids (hash table is used for index).
75 * -# caching and life-cycle management.
77 * Object's life-time is controlled by reference counting. When reference
78 * count drops to 0, object is returned to cache. Cached objects still
79 * retain their identity (i.e., fid), and can be recovered from cache.
81 * Objects are kept in the global LRU list, and lu_site_purge() function
82 * can be used to reclaim given number of unused objects from the tail of
85 * -# avoiding recursion.
87 * Generic code tries to replace recursion through layers by iterations
88 * where possible. Additionally to the end of reducing stack consumption,
89 * data, when practically possible, are allocated through lu_context_key
90 * interface rather than on stack.
97 struct lu_object_header;
102 * Operations common for data and meta-data devices.
104 struct lu_device_operations {
106 * Allocate object for the given device (without lower-layer
107 * parts). This is called by lu_object_operations::loo_object_init()
108 * from the parent layer, and should setup at least lu_object::lo_dev
109 * and lu_object::lo_ops fields of resulting lu_object.
111 * Object creation protocol.
113 * Due to design goal of avoiding recursion, object creation (see
114 * lu_object_alloc()) is somewhat involved:
116 * - first, lu_device_operations::ldo_object_alloc() method of the
117 * top-level device in the stack is called. It should allocate top
118 * level object (including lu_object_header), but without any
119 * lower-layer sub-object(s).
121 * - then lu_object_alloc() sets fid in the header of newly created
124 * - then lu_object_operations::loo_object_init() is called. It has
125 * to allocate lower-layer object(s). To do this,
126 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
127 * of the lower-layer device(s).
129 * - for all new objects allocated by
130 * lu_object_operations::loo_object_init() (and inserted into object
131 * stack), lu_object_operations::loo_object_init() is called again
132 * repeatedly, until no new objects are created.
134 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
135 * result->lo_ops != NULL);
137 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
138 const struct lu_object_header *h,
139 struct lu_device *d);
141 * process config specific for device.
143 int (*ldo_process_config)(const struct lu_env *env,
144 struct lu_device *, struct lustre_cfg *);
145 int (*ldo_recovery_complete)(const struct lu_env *,
149 * initialize local objects for device. this method called after layer has
150 * been initialized (after LCFG_SETUP stage) and before it starts serving
154 int (*ldo_prepare)(const struct lu_env *,
155 struct lu_device *parent,
156 struct lu_device *dev);
161 * For lu_object_conf flags
164 /* This is a new object to be allocated, or the file
165 * corresponding to the object does not exists.
167 LOC_F_NEW = 0x00000001,
171 * Object configuration, describing particulars of object being created. On
172 * server this is not used, as server objects are full identified by fid. On
173 * client configuration contains struct lustre_md.
175 struct lu_object_conf {
177 * Some hints for obj find and alloc.
179 enum loc_flags loc_flags;
183 * Type of "printer" function used by lu_object_operations::loo_object_print()
186 * Printer function is needed to provide some flexibility in (semi-)debugging
187 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
189 typedef int (*lu_printer_t)(const struct lu_env *env,
190 void *cookie, const char *format, ...)
194 * Operations specific for particular lu_object.
196 struct lu_object_operations {
198 * Allocate lower-layer parts of the object by calling
199 * lu_device_operations::ldo_object_alloc() of the corresponding
202 * This method is called once for each object inserted into object
203 * stack. It's responsibility of this method to insert lower-layer
204 * object(s) it create into appropriate places of object stack.
206 int (*loo_object_init)(const struct lu_env *env,
208 const struct lu_object_conf *conf);
210 * Called (in top-to-bottom order) during object allocation after all
211 * layers were allocated and initialized. Can be used to perform
212 * initialization depending on lower layers.
214 int (*loo_object_start)(const struct lu_env *env,
215 struct lu_object *o);
217 * Called before lu_object_operations::loo_object_free() to signal
218 * that object is being destroyed. Dual to
219 * lu_object_operations::loo_object_init().
221 void (*loo_object_delete)(const struct lu_env *env,
222 struct lu_object *o);
224 * Dual to lu_device_operations::ldo_object_alloc(). Called when
225 * object is removed from memory.
227 void (*loo_object_free)(const struct lu_env *env,
228 struct lu_object *o);
230 * Called when last active reference to the object is released (and
231 * object returns to the cache). This method is optional.
233 void (*loo_object_release)(const struct lu_env *env,
234 struct lu_object *o);
236 * Optional debugging helper. Print given object.
238 int (*loo_object_print)(const struct lu_env *env, void *cookie,
239 lu_printer_t p, const struct lu_object *o);
241 * Optional debugging method. Returns true iff method is internally
244 int (*loo_object_invariant)(const struct lu_object *o);
250 struct lu_device_type;
253 * Device: a layer in the server side abstraction stacking.
257 * reference count. This is incremented, in particular, on each object
258 * created at this layer.
260 * \todo XXX which means that atomic_t is probably too small.
264 * Pointer to device type. Never modified once set.
266 struct lu_device_type *ld_type;
268 * Operation vector for this device.
270 const struct lu_device_operations *ld_ops;
272 * Stack this device belongs to.
274 struct lu_site *ld_site;
276 /** \todo XXX: temporary back pointer into obd. */
277 struct obd_device *ld_obd;
279 * A list of references to this object, for debugging.
281 struct lu_ref ld_reference;
283 * Link the device to the site.
285 struct list_head ld_linkage;
288 struct lu_device_type_operations;
291 * Tag bits for device type. They are used to distinguish certain groups of
295 /** this is meta-data device */
296 LU_DEVICE_MD = (1 << 0),
297 /** this is data device */
298 LU_DEVICE_DT = (1 << 1),
299 /** data device in the client stack */
300 LU_DEVICE_CL = (1 << 2)
306 struct lu_device_type {
308 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
312 * Name of this class. Unique system-wide. Never modified once set.
316 * Operations for this type.
318 const struct lu_device_type_operations *ldt_ops;
320 * \todo XXX: temporary pointer to associated obd_type.
322 struct obd_type *ldt_obd_type;
324 * \todo XXX: temporary: context tags used by obd_*() calls.
328 * Number of existing device type instances.
330 atomic_t ldt_device_nr;
332 * Linkage into a global list of all device types.
334 * \see lu_device_types.
336 struct list_head ldt_linkage;
340 * Operations on a device type.
342 struct lu_device_type_operations {
344 * Allocate new device.
346 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
347 struct lu_device_type *t,
348 struct lustre_cfg *lcfg);
350 * Free device. Dual to
351 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
352 * the next device in the stack.
354 struct lu_device *(*ldto_device_free)(const struct lu_env *,
358 * Initialize the devices after allocation
360 int (*ldto_device_init)(const struct lu_env *env,
361 struct lu_device *, const char *,
364 * Finalize device. Dual to
365 * lu_device_type_operations::ldto_device_init(). Returns pointer to
366 * the next device in the stack.
368 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
371 * Initialize device type. This is called on module load.
373 int (*ldto_init)(struct lu_device_type *t);
375 * Finalize device type. Dual to
376 * lu_device_type_operations::ldto_init(). Called on module unload.
378 void (*ldto_fini)(struct lu_device_type *t);
380 * Called when the first device is created.
382 void (*ldto_start)(struct lu_device_type *t);
384 * Called when number of devices drops to 0.
386 void (*ldto_stop)(struct lu_device_type *t);
389 static inline int lu_device_is_md(const struct lu_device *d)
391 return ergo(d, d->ld_type->ldt_tags & LU_DEVICE_MD);
395 * Common object attributes.
400 /** modification time in seconds since Epoch */
402 /** access time in seconds since Epoch */
404 /** change time in seconds since Epoch */
406 /** 512-byte blocks allocated to object */
408 /** permission bits and file type */
416 /** number of persistent references to this object */
418 /** blk bits of the object*/
420 /** blk size of the object*/
432 /** Bit-mask of valid attributes */
446 LA_BLKSIZE = 1 << 12,
447 LA_KILL_SUID = 1 << 13,
448 LA_KILL_SGID = 1 << 14,
452 * Layer in the layered object.
456 * Header for this object.
458 struct lu_object_header *lo_header;
460 * Device for this layer.
462 struct lu_device *lo_dev;
464 * Operations for this object.
466 const struct lu_object_operations *lo_ops;
468 * Linkage into list of all layers.
470 struct list_head lo_linkage;
472 * Link to the device, for debugging.
474 struct lu_ref_link lo_dev_ref;
477 enum lu_object_header_flags {
479 * Don't keep this object in cache. Object will be destroyed as soon
480 * as last reference to it is released. This flag cannot be cleared
483 LU_OBJECT_HEARD_BANSHEE = 0,
485 * Mark this object has already been taken out of cache.
487 LU_OBJECT_UNHASHED = 1,
490 enum lu_object_header_attr {
491 LOHA_EXISTS = 1 << 0,
492 LOHA_REMOTE = 1 << 1,
494 * UNIX file type is stored in S_IFMT bits.
496 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
497 LOHA_FT_END = 017 << 12, /**< S_IFMT */
501 * "Compound" object, consisting of multiple layers.
503 * Compound object with given fid is unique with given lu_site.
505 * Note, that object does *not* necessary correspond to the real object in the
506 * persistent storage: object is an anchor for locking and method calling, so
507 * it is created for things like not-yet-existing child created by mkdir or
508 * create calls. lu_object_operations::loo_exists() can be used to check
509 * whether object is backed by persistent storage entity.
511 struct lu_object_header {
513 * Fid, uniquely identifying this object.
515 struct lu_fid loh_fid;
517 * Object flags from enum lu_object_header_flags. Set and checked
520 unsigned long loh_flags;
522 * Object reference count. Protected by lu_site::ls_guard.
526 * Common object attributes, cached for efficiency. From enum
527 * lu_object_header_attr.
531 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
533 struct hlist_node loh_hash;
535 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
537 struct list_head loh_lru;
539 * Linkage into list of layers. Never modified once set (except lately
540 * during object destruction). No locking is necessary.
542 struct list_head loh_layers;
544 * A list of references to this object, for debugging.
546 struct lu_ref loh_reference;
551 struct lu_site_bkt_data {
553 * number of object in this bucket on the lsb_lru list.
557 * LRU list, updated on each access to object. Protected by
558 * bucket lock of lu_site::ls_obj_hash.
560 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
561 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
562 * of list_for_each_entry_safe_reverse()).
564 struct list_head lsb_lru;
566 * Wait-queue signaled when an object in this site is ultimately
567 * destroyed (lu_object_free()). It is used by lu_object_find() to
568 * wait before re-trying when object in the process of destruction is
569 * found in the hash table.
571 * \see htable_lookup().
573 wait_queue_head_t lsb_marche_funebre;
581 LU_SS_CACHE_DEATH_RACE,
583 LU_SS_LRU_LEN, /* # of objects in lsb_lru lists */
588 * lu_site is a "compartment" within which objects are unique, and LRU
589 * discipline is maintained.
591 * lu_site exists so that multiple layered stacks can co-exist in the same
594 * lu_site has the same relation to lu_device as lu_object_header to
601 struct cfs_hash *ls_obj_hash;
603 * index of bucket on hash table while purging
605 unsigned int ls_purge_start;
607 * Top-level device for this stack.
609 struct lu_device *ls_top_dev;
611 * Bottom-level device for this stack
613 struct lu_device *ls_bottom_dev;
615 * Linkage into global list of sites.
617 struct list_head ls_linkage;
619 * List for lu device for this site, protected
622 struct list_head ls_ld_linkage;
623 spinlock_t ls_ld_lock;
626 * Lock to serialize site purge.
628 struct mutex ls_purge_mutex;
633 struct lprocfs_stats *ls_stats;
635 * XXX: a hack! fld has to find md_site via site, remove when possible
637 struct seq_server_site *ld_seq_site;
640 static inline struct lu_site_bkt_data *
641 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
643 struct cfs_hash_bd bd;
645 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
646 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
649 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
651 return s->ld_seq_site;
655 * Constructors/destructors.
659 int lu_site_init(struct lu_site *s, struct lu_device *d);
660 void lu_site_fini(struct lu_site *s);
661 int lu_site_init_finish(struct lu_site *s);
662 void lu_stack_fini(const struct lu_env *env, struct lu_device *top);
663 void lu_device_get(struct lu_device *d);
664 void lu_device_put(struct lu_device *d);
665 int lu_device_init(struct lu_device *d, struct lu_device_type *t);
666 void lu_device_fini(struct lu_device *d);
667 int lu_object_header_init(struct lu_object_header *h);
668 void lu_object_header_fini(struct lu_object_header *h);
669 int lu_object_init(struct lu_object *o,
670 struct lu_object_header *h, struct lu_device *d);
671 void lu_object_fini(struct lu_object *o);
672 void lu_object_add_top(struct lu_object_header *h, struct lu_object *o);
673 void lu_object_add(struct lu_object *before, struct lu_object *o);
676 * Helpers to initialize and finalize device types.
679 int lu_device_type_init(struct lu_device_type *ldt);
680 void lu_device_type_fini(struct lu_device_type *ldt);
685 * Caching and reference counting.
690 * Acquire additional reference to the given object. This function is used to
691 * attain additional reference. To acquire initial reference use
694 static inline void lu_object_get(struct lu_object *o)
696 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
697 atomic_inc(&o->lo_header->loh_ref);
701 * Return true of object will not be cached after last reference to it is
704 static inline int lu_object_is_dying(const struct lu_object_header *h)
706 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
709 void lu_object_put(const struct lu_env *env, struct lu_object *o);
710 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
712 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
714 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
715 lu_printer_t printer);
716 struct lu_object *lu_object_find_at(const struct lu_env *env,
717 struct lu_device *dev,
718 const struct lu_fid *f,
719 const struct lu_object_conf *conf);
720 struct lu_object *lu_object_find_slice(const struct lu_env *env,
721 struct lu_device *dev,
722 const struct lu_fid *f,
723 const struct lu_object_conf *conf);
732 * First (topmost) sub-object of given compound object
734 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
736 LASSERT(!list_empty(&h->loh_layers));
737 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
741 * Next sub-object in the layering
743 static inline struct lu_object *lu_object_next(const struct lu_object *o)
745 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
749 * Pointer to the fid of this object.
751 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
753 return &o->lo_header->loh_fid;
757 * return device operations vector for this object
759 static inline const struct lu_device_operations *
760 lu_object_ops(const struct lu_object *o)
762 return o->lo_dev->ld_ops;
766 * Given a compound object, find its slice, corresponding to the device type
769 struct lu_object *lu_object_locate(struct lu_object_header *h,
770 const struct lu_device_type *dtype);
773 * Printer function emitting messages through libcfs_debug_msg().
775 int lu_cdebug_printer(const struct lu_env *env,
776 void *cookie, const char *format, ...);
779 * Print object description followed by a user-supplied message.
781 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
783 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
784 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
785 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
786 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
791 * Print short object description followed by a user-supplied message.
793 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
795 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
796 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
797 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
798 (object)->lo_header); \
799 lu_cdebug_printer(env, &msgdata, "\n"); \
800 CDEBUG(mask, format, ## __VA_ARGS__); \
804 void lu_object_print (const struct lu_env *env, void *cookie,
805 lu_printer_t printer, const struct lu_object *o);
806 void lu_object_header_print(const struct lu_env *env, void *cookie,
807 lu_printer_t printer,
808 const struct lu_object_header *hdr);
811 * Check object consistency.
813 int lu_object_invariant(const struct lu_object *o);
816 * Check whether object exists, no matter on local or remote storage.
817 * Note: LOHA_EXISTS will be set once some one created the object,
818 * and it does not needs to be committed to storage.
820 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
823 * Check whether object on the remote storage.
825 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
827 static inline int lu_object_assert_exists(const struct lu_object *o)
829 return lu_object_exists(o);
832 static inline int lu_object_assert_not_exists(const struct lu_object *o)
834 return !lu_object_exists(o);
838 * Attr of this object.
840 static inline __u32 lu_object_attr(const struct lu_object *o)
842 LASSERT(lu_object_exists(o) != 0);
843 return o->lo_header->loh_attr;
846 static inline void lu_object_ref_add(struct lu_object *o,
850 lu_ref_add(&o->lo_header->loh_reference, scope, source);
853 static inline void lu_object_ref_add_at(struct lu_object *o,
854 struct lu_ref_link *link,
858 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
861 static inline void lu_object_ref_del(struct lu_object *o,
862 const char *scope, const void *source)
864 lu_ref_del(&o->lo_header->loh_reference, scope, source);
867 static inline void lu_object_ref_del_at(struct lu_object *o,
868 struct lu_ref_link *link,
869 const char *scope, const void *source)
871 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
874 /** input params, should be filled out by mdt */
878 /** count in bytes */
879 unsigned int rp_count;
880 /** number of pages */
881 unsigned int rp_npages;
882 /** requested attr */
884 /** pointers to pages */
885 struct page **rp_pages;
888 enum lu_xattr_flags {
889 LU_XATTR_REPLACE = (1 << 0),
890 LU_XATTR_CREATE = (1 << 1)
899 /** For lu_context health-checks */
900 enum lu_context_state {
908 * lu_context. Execution context for lu_object methods. Currently associated
911 * All lu_object methods, except device and device type methods (called during
912 * system initialization and shutdown) are executed "within" some
913 * lu_context. This means, that pointer to some "current" lu_context is passed
914 * as an argument to all methods.
916 * All service ptlrpc threads create lu_context as part of their
917 * initialization. It is possible to create "stand-alone" context for other
918 * execution environments (like system calls).
920 * lu_object methods mainly use lu_context through lu_context_key interface
921 * that allows each layer to associate arbitrary pieces of data with each
922 * context (see pthread_key_create(3) for similar interface).
924 * On a client, lu_context is bound to a thread, see cl_env_get().
926 * \see lu_context_key
930 * lu_context is used on the client side too. Yet we don't want to
931 * allocate values of server-side keys for the client contexts and
934 * To achieve this, set of tags in introduced. Contexts and keys are
935 * marked with tags. Key value are created only for context whose set
936 * of tags has non-empty intersection with one for key. Tags are taken
937 * from enum lu_context_tag.
940 enum lu_context_state lc_state;
942 * Pointer to the home service thread. NULL for other execution
945 struct ptlrpc_thread *lc_thread;
947 * Pointer to an array with key values. Internal implementation
952 * Linkage into a list of all remembered contexts. Only
953 * `non-transient' contexts, i.e., ones created for service threads
956 struct list_head lc_remember;
958 * Version counter used to skip calls to lu_context_refill() when no
959 * keys were registered.
969 * lu_context_key interface. Similar to pthread_key.
972 enum lu_context_tag {
974 * Thread on md server
976 LCT_MD_THREAD = 1 << 0,
978 * Thread on dt server
980 LCT_DT_THREAD = 1 << 1,
982 * Context for transaction handle
984 LCT_TX_HANDLE = 1 << 2,
988 LCT_CL_THREAD = 1 << 3,
990 * A per-request session on a server, and a per-system-call session on
993 LCT_SESSION = 1 << 4,
995 * A per-request data on OSP device
997 LCT_OSP_THREAD = 1 << 5,
1001 LCT_MG_THREAD = 1 << 6,
1003 * Context for local operations
1007 * session for server thread
1009 LCT_SERVER_SESSION = BIT(8),
1011 * Set when at least one of keys, having values in this context has
1012 * non-NULL lu_context_key::lct_exit() method. This is used to
1013 * optimize lu_context_exit() call.
1015 LCT_HAS_EXIT = 1 << 28,
1017 * Don't add references for modules creating key values in that context.
1018 * This is only for contexts used internally by lu_object framework.
1020 LCT_NOREF = 1 << 29,
1022 * Key is being prepared for retiring, don't create new values for it.
1024 LCT_QUIESCENT = 1 << 30,
1026 * Context should be remembered.
1028 LCT_REMEMBER = 1 << 31,
1030 * Contexts usable in cache shrinker thread.
1032 LCT_SHRINKER = LCT_MD_THREAD | LCT_DT_THREAD | LCT_CL_THREAD |
1037 * Key. Represents per-context value slot.
1039 * Keys are usually registered when module owning the key is initialized, and
1040 * de-registered when module is unloaded. Once key is registered, all new
1041 * contexts with matching tags, will get key value. "Old" contexts, already
1042 * initialized at the time of key registration, can be forced to get key value
1043 * by calling lu_context_refill().
1045 * Every key value is counted in lu_context_key::lct_used and acquires a
1046 * reference on an owning module. This means, that all key values have to be
1047 * destroyed before module can be unloaded. This is usually achieved by
1048 * stopping threads started by the module, that created contexts in their
1049 * entry functions. Situation is complicated by the threads shared by multiple
1050 * modules, like ptlrpcd daemon on a client. To work around this problem,
1051 * contexts, created in such threads, are `remembered' (see
1052 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1053 * for unloading it does the following:
1055 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1056 * preventing new key values from being allocated in the new contexts,
1059 * - scans a list of remembered contexts, destroying values of module
1060 * keys, thus releasing references to the module.
1062 * This is done by lu_context_key_quiesce(). If module is re-activated
1063 * before key has been de-registered, lu_context_key_revive() call clears
1064 * `quiescent' marker.
1066 * lu_context code doesn't provide any internal synchronization for these
1067 * activities---it's assumed that startup (including threads start-up) and
1068 * shutdown are serialized by some external means.
1072 struct lu_context_key {
1074 * Set of tags for which values of this key are to be instantiated.
1078 * Value constructor. This is called when new value is created for a
1079 * context. Returns pointer to new value of error pointer.
1081 void *(*lct_init)(const struct lu_context *ctx,
1082 struct lu_context_key *key);
1084 * Value destructor. Called when context with previously allocated
1085 * value of this slot is destroyed. \a data is a value that was returned
1086 * by a matching call to lu_context_key::lct_init().
1088 void (*lct_fini)(const struct lu_context *ctx,
1089 struct lu_context_key *key, void *data);
1091 * Optional method called on lu_context_exit() for all allocated
1092 * keys. Can be used by debugging code checking that locks are
1095 void (*lct_exit)(const struct lu_context *ctx,
1096 struct lu_context_key *key, void *data);
1098 * Internal implementation detail: index within lu_context::lc_value[]
1099 * reserved for this key.
1103 * Internal implementation detail: number of values created for this
1108 * Internal implementation detail: module for this key.
1110 struct module *lct_owner;
1112 * References to this key. For debugging.
1114 struct lu_ref lct_reference;
1117 #define LU_KEY_INIT(mod, type) \
1118 static void *mod##_key_init(const struct lu_context *ctx, \
1119 struct lu_context_key *key) \
1123 CLASSERT(PAGE_SIZE >= sizeof(*value)); \
1125 value = kzalloc(sizeof(*value), GFP_NOFS); \
1127 value = ERR_PTR(-ENOMEM); \
1131 struct __##mod##__dummy_init {; } /* semicolon catcher */
1133 #define LU_KEY_FINI(mod, type) \
1134 static void mod##_key_fini(const struct lu_context *ctx, \
1135 struct lu_context_key *key, void *data) \
1137 type *info = data; \
1141 struct __##mod##__dummy_fini {; } /* semicolon catcher */
1143 #define LU_KEY_INIT_FINI(mod, type) \
1144 LU_KEY_INIT(mod, type); \
1145 LU_KEY_FINI(mod, type)
1147 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1148 struct lu_context_key mod##_thread_key = { \
1150 .lct_init = mod##_key_init, \
1151 .lct_fini = mod##_key_fini \
1154 #define LU_CONTEXT_KEY_INIT(key) \
1156 (key)->lct_owner = THIS_MODULE; \
1159 int lu_context_key_register(struct lu_context_key *key);
1160 void lu_context_key_degister(struct lu_context_key *key);
1161 void *lu_context_key_get(const struct lu_context *ctx,
1162 const struct lu_context_key *key);
1163 void lu_context_key_quiesce(struct lu_context_key *key);
1164 void lu_context_key_revive(struct lu_context_key *key);
1167 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1171 #define LU_KEY_INIT_GENERIC(mod) \
1172 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1174 struct lu_context_key *key = k; \
1177 va_start(args, k); \
1179 LU_CONTEXT_KEY_INIT(key); \
1180 key = va_arg(args, struct lu_context_key *); \
1185 #define LU_TYPE_INIT(mod, ...) \
1186 LU_KEY_INIT_GENERIC(mod) \
1187 static int mod##_type_init(struct lu_device_type *t) \
1189 mod##_key_init_generic(__VA_ARGS__, NULL); \
1190 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1192 struct __##mod##_dummy_type_init {; }
1194 #define LU_TYPE_FINI(mod, ...) \
1195 static void mod##_type_fini(struct lu_device_type *t) \
1197 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1199 struct __##mod##_dummy_type_fini {; }
1201 #define LU_TYPE_START(mod, ...) \
1202 static void mod##_type_start(struct lu_device_type *t) \
1204 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1206 struct __##mod##_dummy_type_start {; }
1208 #define LU_TYPE_STOP(mod, ...) \
1209 static void mod##_type_stop(struct lu_device_type *t) \
1211 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1213 struct __##mod##_dummy_type_stop {; }
1215 #define LU_TYPE_INIT_FINI(mod, ...) \
1216 LU_TYPE_INIT(mod, __VA_ARGS__); \
1217 LU_TYPE_FINI(mod, __VA_ARGS__); \
1218 LU_TYPE_START(mod, __VA_ARGS__); \
1219 LU_TYPE_STOP(mod, __VA_ARGS__)
1221 int lu_context_init(struct lu_context *ctx, __u32 tags);
1222 void lu_context_fini(struct lu_context *ctx);
1223 void lu_context_enter(struct lu_context *ctx);
1224 void lu_context_exit(struct lu_context *ctx);
1225 int lu_context_refill(struct lu_context *ctx);
1228 * Helper functions to operate on multiple keys. These are used by the default
1229 * device type operations, defined by LU_TYPE_INIT_FINI().
1232 int lu_context_key_register_many(struct lu_context_key *k, ...);
1233 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1234 void lu_context_key_revive_many(struct lu_context_key *k, ...);
1235 void lu_context_key_quiesce_many(struct lu_context_key *k, ...);
1242 * "Local" context, used to store data instead of stack.
1244 struct lu_context le_ctx;
1246 * "Session" context for per-request data.
1248 struct lu_context *le_ses;
1251 int lu_env_init(struct lu_env *env, __u32 tags);
1252 void lu_env_fini(struct lu_env *env);
1253 int lu_env_refill(struct lu_env *env);
1255 /** @} lu_context */
1258 * Output site statistical counters into a buffer. Suitable for
1259 * ll_rd_*()-style functions.
1261 int lu_site_stats_print(const struct lu_site *s, struct seq_file *m);
1264 * Common name structure to be passed around for various name related methods.
1267 const char *ln_name;
1272 * Validate names (path components)
1274 * To be valid \a name must be non-empty, '\0' terminated of length \a
1275 * name_len, and not contain '/'. The maximum length of a name (before
1276 * say -ENAMETOOLONG will be returned) is really controlled by llite
1277 * and the server. We only check for something insane coming from bad
1278 * integer handling here.
1280 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1282 return name && name_len > 0 && name_len < INT_MAX &&
1283 name[name_len] == '\0' && strlen(name) == name_len &&
1284 !memchr(name, '/', name_len);
1288 * Common buffer structure to be passed around for various xattr_{s,g}et()
1296 #define DLUBUF "(%p %zu)"
1297 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1299 * One-time initializers, called at obdclass module initialization, not
1304 * Initialization of global lu_* data.
1306 int lu_global_init(void);
1309 * Dual to lu_global_init().
1311 void lu_global_fini(void);
1313 struct lu_kmem_descr {
1314 struct kmem_cache **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);
1322 void lu_buf_free(struct lu_buf *buf);
1323 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1324 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1326 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1327 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1330 #endif /* __LUSTRE_LU_OBJECT_H */