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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
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
41 #include <libcfs/libcfs.h>
42 #include <lustre/lustre_idl.h>
46 struct proc_dir_entry;
51 * lu_* data-types represent server-side entities shared by data and meta-data
56 * -# support for layering.
58 * Server side object is split into layers, one per device in the
59 * corresponding device stack. Individual layer is represented by struct
60 * lu_object. Compound layered object --- by struct lu_object_header. Most
61 * interface functions take lu_object as an argument and operate on the
62 * whole compound object. This decision was made due to the following
65 * - it's envisaged that lu_object will be used much more often than
68 * - we want lower (non-top) layers to be able to initiate operations
69 * on the whole object.
71 * Generic code supports layering more complex than simple stacking, e.g.,
72 * it is possible that at some layer object "spawns" multiple sub-objects
75 * -# fid-based identification.
77 * Compound object is uniquely identified by its fid. Objects are indexed
78 * by their fids (hash table is used for index).
80 * -# caching and life-cycle management.
82 * Object's life-time is controlled by reference counting. When reference
83 * count drops to 0, object is returned to cache. Cached objects still
84 * retain their identity (i.e., fid), and can be recovered from cache.
86 * Objects are kept in the global LRU list, and lu_site_purge() function
87 * can be used to reclaim given number of unused objects from the tail of
90 * -# avoiding recursion.
92 * Generic code tries to replace recursion through layers by iterations
93 * where possible. Additionally to the end of reducing stack consumption,
94 * data, when practically possible, are allocated through lu_context_key
95 * interface rather than on stack.
102 struct lu_object_header;
107 * Operations common for data and meta-data devices.
109 struct lu_device_operations {
111 * Allocate object for the given device (without lower-layer
112 * parts). This is called by lu_object_operations::loo_object_init()
113 * from the parent layer, and should setup at least lu_object::lo_dev
114 * and lu_object::lo_ops fields of resulting lu_object.
116 * Object creation protocol.
118 * Due to design goal of avoiding recursion, object creation (see
119 * lu_object_alloc()) is somewhat involved:
121 * - first, lu_device_operations::ldo_object_alloc() method of the
122 * top-level device in the stack is called. It should allocate top
123 * level object (including lu_object_header), but without any
124 * lower-layer sub-object(s).
126 * - then lu_object_alloc() sets fid in the header of newly created
129 * - then lu_object_operations::loo_object_init() is called. It has
130 * to allocate lower-layer object(s). To do this,
131 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
132 * of the lower-layer device(s).
134 * - for all new objects allocated by
135 * lu_object_operations::loo_object_init() (and inserted into object
136 * stack), lu_object_operations::loo_object_init() is called again
137 * repeatedly, until no new objects are created.
139 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
140 * result->lo_ops != NULL);
142 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
143 const struct lu_object_header *h,
144 struct lu_device *d);
146 * process config specific for device.
148 int (*ldo_process_config)(const struct lu_env *env,
149 struct lu_device *, struct lustre_cfg *);
150 int (*ldo_recovery_complete)(const struct lu_env *,
154 * initialize local objects for device. this method called after layer has
155 * been initialized (after LCFG_SETUP stage) and before it starts serving
159 int (*ldo_prepare)(const struct lu_env *,
160 struct lu_device *parent,
161 struct lu_device *dev);
166 * For lu_object_conf flags
169 /* This is a new object to be allocated, or the file
170 * corresponding to the object does not exists. */
171 LOC_F_NEW = 0x00000001,
175 * Object configuration, describing particulars of object being created. On
176 * server this is not used, as server objects are full identified by fid. On
177 * client configuration contains struct lustre_md.
179 struct lu_object_conf {
181 * Some hints for obj find and alloc.
183 loc_flags_t loc_flags;
187 * Type of "printer" function used by lu_object_operations::loo_object_print()
190 * Printer function is needed to provide some flexibility in (semi-)debugging
191 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
193 typedef int (*lu_printer_t)(const struct lu_env *env,
194 void *cookie, const char *format, ...)
195 __attribute__ ((format (printf, 3, 4)));
198 * Operations specific for particular lu_object.
200 struct lu_object_operations {
203 * Allocate lower-layer parts of the object by calling
204 * lu_device_operations::ldo_object_alloc() of the corresponding
207 * This method is called once for each object inserted into object
208 * stack. It's responsibility of this method to insert lower-layer
209 * object(s) it create into appropriate places of object stack.
211 int (*loo_object_init)(const struct lu_env *env,
213 const struct lu_object_conf *conf);
215 * Called (in top-to-bottom order) during object allocation after all
216 * layers were allocated and initialized. Can be used to perform
217 * initialization depending on lower layers.
219 int (*loo_object_start)(const struct lu_env *env,
220 struct lu_object *o);
222 * Called before lu_object_operations::loo_object_free() to signal
223 * that object is being destroyed. Dual to
224 * lu_object_operations::loo_object_init().
226 void (*loo_object_delete)(const struct lu_env *env,
227 struct lu_object *o);
229 * Dual to lu_device_operations::ldo_object_alloc(). Called when
230 * object is removed from memory.
232 void (*loo_object_free)(const struct lu_env *env,
233 struct lu_object *o);
235 * Called when last active reference to the object is released (and
236 * object returns to the cache). This method is optional.
238 void (*loo_object_release)(const struct lu_env *env,
239 struct lu_object *o);
241 * Optional debugging helper. Print given object.
243 int (*loo_object_print)(const struct lu_env *env, void *cookie,
244 lu_printer_t p, const struct lu_object *o);
246 * Optional debugging method. Returns true iff method is internally
249 int (*loo_object_invariant)(const struct lu_object *o);
255 struct lu_device_type;
258 * Device: a layer in the server side abstraction stacking.
262 * reference count. This is incremented, in particular, on each object
263 * created at this layer.
265 * \todo XXX which means that atomic_t is probably too small.
269 * Pointer to device type. Never modified once set.
271 struct lu_device_type *ld_type;
273 * Operation vector for this device.
275 const struct lu_device_operations *ld_ops;
277 * Stack this device belongs to.
279 struct lu_site *ld_site;
280 struct proc_dir_entry *ld_proc_entry;
282 /** \todo XXX: temporary back pointer into obd. */
283 struct obd_device *ld_obd;
285 * A list of references to this object, for debugging.
287 struct lu_ref ld_reference;
289 * Link the device to the site.
291 cfs_list_t ld_linkage;
294 struct lu_device_type_operations;
297 * Tag bits for device type. They are used to distinguish certain groups of
301 /** this is meta-data device */
302 LU_DEVICE_MD = (1 << 0),
303 /** this is data device */
304 LU_DEVICE_DT = (1 << 1),
305 /** data device in the client stack */
306 LU_DEVICE_CL = (1 << 2)
312 struct lu_device_type {
314 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
318 * Name of this class. Unique system-wide. Never modified once set.
322 * Operations for this type.
324 const struct lu_device_type_operations *ldt_ops;
326 * \todo XXX: temporary pointer to associated obd_type.
328 struct obd_type *ldt_obd_type;
330 * \todo XXX: temporary: context tags used by obd_*() calls.
334 * Number of existing device type instances.
336 unsigned ldt_device_nr;
338 * Linkage into a global list of all device types.
340 * \see lu_device_types.
342 cfs_list_t ldt_linkage;
346 * Operations on a device type.
348 struct lu_device_type_operations {
350 * Allocate new device.
352 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
353 struct lu_device_type *t,
354 struct lustre_cfg *lcfg);
356 * Free device. Dual to
357 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
358 * the next device in the stack.
360 struct lu_device *(*ldto_device_free)(const struct lu_env *,
364 * Initialize the devices after allocation
366 int (*ldto_device_init)(const struct lu_env *env,
367 struct lu_device *, const char *,
370 * Finalize device. Dual to
371 * lu_device_type_operations::ldto_device_init(). Returns pointer to
372 * the next device in the stack.
374 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
377 * Initialize device type. This is called on module load.
379 int (*ldto_init)(struct lu_device_type *t);
381 * Finalize device type. Dual to
382 * lu_device_type_operations::ldto_init(). Called on module unload.
384 void (*ldto_fini)(struct lu_device_type *t);
386 * Called when the first device is created.
388 void (*ldto_start)(struct lu_device_type *t);
390 * Called when number of devices drops to 0.
392 void (*ldto_stop)(struct lu_device_type *t);
395 static inline int lu_device_is_md(const struct lu_device *d)
397 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
401 * Flags for the object layers.
403 enum lu_object_flags {
405 * this flags is set if lu_object_operations::loo_object_init() has
406 * been called for this layer. Used by lu_object_alloc().
408 LU_OBJECT_ALLOCATED = (1 << 0)
412 * Common object attributes.
417 /** modification time in seconds since Epoch */
419 /** access time in seconds since Epoch */
421 /** change time in seconds since Epoch */
423 /** 512-byte blocks allocated to object */
425 /** permission bits and file type */
433 /** number of persistent references to this object */
435 /** blk bits of the object*/
437 /** blk size of the object*/
449 /** Bit-mask of valid attributes */
463 LA_BLKSIZE = 1 << 12,
464 LA_KILL_SUID = 1 << 13,
465 LA_KILL_SGID = 1 << 14,
469 * Layer in the layered object.
473 * Header for this object.
475 struct lu_object_header *lo_header;
477 * Device for this layer.
479 struct lu_device *lo_dev;
481 * Operations for this object.
483 const struct lu_object_operations *lo_ops;
485 * Linkage into list of all layers.
487 cfs_list_t lo_linkage;
489 * Depth. Top level layer depth is 0.
493 * Flags from enum lu_object_flags.
497 * Link to the device, for debugging.
499 struct lu_ref_link lo_dev_ref;
502 enum lu_object_header_flags {
504 * Don't keep this object in cache. Object will be destroyed as soon
505 * as last reference to it is released. This flag cannot be cleared
508 LU_OBJECT_HEARD_BANSHEE = 0,
510 * Mark this object has already been taken out of cache.
512 LU_OBJECT_UNHASHED = 1,
515 enum lu_object_header_attr {
516 LOHA_EXISTS = 1 << 0,
517 LOHA_REMOTE = 1 << 1,
519 * UNIX file type is stored in S_IFMT bits.
521 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
522 LOHA_FT_END = 017 << 12, /**< S_IFMT */
526 * "Compound" object, consisting of multiple layers.
528 * Compound object with given fid is unique with given lu_site.
530 * Note, that object does *not* necessary correspond to the real object in the
531 * persistent storage: object is an anchor for locking and method calling, so
532 * it is created for things like not-yet-existing child created by mkdir or
533 * create calls. lu_object_operations::loo_exists() can be used to check
534 * whether object is backed by persistent storage entity.
536 struct lu_object_header {
538 * Object flags from enum lu_object_header_flags. Set and checked
541 unsigned long loh_flags;
543 * Object reference count. Protected by lu_site::ls_guard.
545 cfs_atomic_t loh_ref;
547 * Fid, uniquely identifying this object.
549 struct lu_fid loh_fid;
551 * Common object attributes, cached for efficiency. From enum
552 * lu_object_header_attr.
556 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
558 cfs_hlist_node_t loh_hash;
560 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
564 * Linkage into list of layers. Never modified once set (except lately
565 * during object destruction). No locking is necessary.
567 cfs_list_t loh_layers;
569 * A list of references to this object, for debugging.
571 struct lu_ref loh_reference;
576 struct lu_site_bkt_data {
578 * number of busy object on this bucket
582 * LRU list, updated on each access to object. Protected by
583 * bucket lock of lu_site::ls_obj_hash.
585 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
586 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
587 * of list_for_each_entry_safe_reverse()).
591 * Wait-queue signaled when an object in this site is ultimately
592 * destroyed (lu_object_free()). It is used by lu_object_find() to
593 * wait before re-trying when object in the process of destruction is
594 * found in the hash table.
596 * \see htable_lookup().
598 cfs_waitq_t lsb_marche_funebre;
606 LU_SS_CACHE_DEATH_RACE,
612 * lu_site is a "compartment" within which objects are unique, and LRU
613 * discipline is maintained.
615 * lu_site exists so that multiple layered stacks can co-exist in the same
618 * lu_site has the same relation to lu_device as lu_object_header to
625 cfs_hash_t *ls_obj_hash;
627 * index of bucket on hash table while purging
631 * Top-level device for this stack.
633 struct lu_device *ls_top_dev;
635 * Bottom-level device for this stack
637 struct lu_device *ls_bottom_dev;
639 * Linkage into global list of sites.
641 cfs_list_t ls_linkage;
643 * List for lu device for this site, protected
646 cfs_list_t ls_ld_linkage;
647 spinlock_t ls_ld_lock;
652 struct lprocfs_stats *ls_stats;
654 * XXX: a hack! fld has to find md_site via site, remove when possible
656 struct seq_server_site *ld_seq_site;
659 static inline struct lu_site_bkt_data *
660 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
664 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
665 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
668 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
670 return s->ld_seq_site;
674 * Constructors/destructors.
678 int lu_site_init (struct lu_site *s, struct lu_device *d);
679 void lu_site_fini (struct lu_site *s);
680 int lu_site_init_finish (struct lu_site *s);
681 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
682 void lu_device_get (struct lu_device *d);
683 void lu_device_put (struct lu_device *d);
684 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
685 void lu_device_fini (struct lu_device *d);
686 int lu_object_header_init(struct lu_object_header *h);
687 void lu_object_header_fini(struct lu_object_header *h);
688 int lu_object_init (struct lu_object *o,
689 struct lu_object_header *h, struct lu_device *d);
690 void lu_object_fini (struct lu_object *o);
691 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
692 void lu_object_add (struct lu_object *before, struct lu_object *o);
694 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
695 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
698 * Helpers to initialize and finalize device types.
701 int lu_device_type_init(struct lu_device_type *ldt);
702 void lu_device_type_fini(struct lu_device_type *ldt);
703 void lu_types_stop(void);
708 * Caching and reference counting.
713 * Acquire additional reference to the given object. This function is used to
714 * attain additional reference. To acquire initial reference use
717 static inline void lu_object_get(struct lu_object *o)
719 LASSERT(cfs_atomic_read(&o->lo_header->loh_ref) > 0);
720 cfs_atomic_inc(&o->lo_header->loh_ref);
724 * Return true of object will not be cached after last reference to it is
727 static inline int lu_object_is_dying(const struct lu_object_header *h)
729 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
732 void lu_object_put(const struct lu_env *env, struct lu_object *o);
733 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
734 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
736 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
738 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
739 lu_printer_t printer);
740 struct lu_object *lu_object_find(const struct lu_env *env,
741 struct lu_device *dev, const struct lu_fid *f,
742 const struct lu_object_conf *conf);
743 struct lu_object *lu_object_find_at(const struct lu_env *env,
744 struct lu_device *dev,
745 const struct lu_fid *f,
746 const struct lu_object_conf *conf);
747 void lu_object_purge(const struct lu_env *env, struct lu_device *dev,
748 const struct lu_fid *f);
749 struct lu_object *lu_object_find_slice(const struct lu_env *env,
750 struct lu_device *dev,
751 const struct lu_fid *f,
752 const struct lu_object_conf *conf);
761 * First (topmost) sub-object of given compound object
763 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
765 LASSERT(!cfs_list_empty(&h->loh_layers));
766 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
770 * Next sub-object in the layering
772 static inline struct lu_object *lu_object_next(const struct lu_object *o)
774 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
778 * Pointer to the fid of this object.
780 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
782 return &o->lo_header->loh_fid;
786 * return device operations vector for this object
788 static const inline struct lu_device_operations *
789 lu_object_ops(const struct lu_object *o)
791 return o->lo_dev->ld_ops;
795 * Given a compound object, find its slice, corresponding to the device type
798 struct lu_object *lu_object_locate(struct lu_object_header *h,
799 const struct lu_device_type *dtype);
802 * Printer function emitting messages through libcfs_debug_msg().
804 int lu_cdebug_printer(const struct lu_env *env,
805 void *cookie, const char *format, ...);
808 * Print object description followed by a user-supplied message.
810 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
812 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
814 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
815 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
816 CDEBUG(mask, format , ## __VA_ARGS__); \
821 * Print short object description followed by a user-supplied message.
823 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
825 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
827 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
828 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
829 (object)->lo_header); \
830 lu_cdebug_printer(env, &msgdata, "\n"); \
831 CDEBUG(mask, format , ## __VA_ARGS__); \
835 void lu_object_print (const struct lu_env *env, void *cookie,
836 lu_printer_t printer, const struct lu_object *o);
837 void lu_object_header_print(const struct lu_env *env, void *cookie,
838 lu_printer_t printer,
839 const struct lu_object_header *hdr);
842 * Check object consistency.
844 int lu_object_invariant(const struct lu_object *o);
848 * Check whether object exists, no matter on local or remote storage.
849 * Note: LOHA_EXISTS will be set once some one created the object,
850 * and it does not needs to be committed to storage.
852 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
855 * Check whether object on the remote storage.
857 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
859 static inline int lu_object_assert_exists(const struct lu_object *o)
861 return lu_object_exists(o);
864 static inline int lu_object_assert_not_exists(const struct lu_object *o)
866 return !lu_object_exists(o);
870 * Attr of this object.
872 static inline __u32 lu_object_attr(const struct lu_object *o)
874 LASSERT(lu_object_exists(o) != 0);
875 return o->lo_header->loh_attr;
878 static inline void lu_object_ref_add(struct lu_object *o,
882 lu_ref_add(&o->lo_header->loh_reference, scope, source);
885 static inline void lu_object_ref_add_at(struct lu_object *o,
886 struct lu_ref_link *link,
890 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
893 static inline void lu_object_ref_del(struct lu_object *o,
894 const char *scope, const void *source)
896 lu_ref_del(&o->lo_header->loh_reference, scope, source);
899 static inline void lu_object_ref_del_at(struct lu_object *o,
900 struct lu_ref_link *link,
901 const char *scope, const void *source)
903 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
906 /** input params, should be filled out by mdt */
910 /** count in bytes */
911 unsigned int rp_count;
912 /** number of pages */
913 unsigned int rp_npages;
914 /** requested attr */
916 /** pointers to pages */
917 struct page **rp_pages;
920 enum lu_xattr_flags {
921 LU_XATTR_REPLACE = (1 << 0),
922 LU_XATTR_CREATE = (1 << 1)
930 /** For lu_context health-checks */
931 enum lu_context_state {
939 * lu_context. Execution context for lu_object methods. Currently associated
942 * All lu_object methods, except device and device type methods (called during
943 * system initialization and shutdown) are executed "within" some
944 * lu_context. This means, that pointer to some "current" lu_context is passed
945 * as an argument to all methods.
947 * All service ptlrpc threads create lu_context as part of their
948 * initialization. It is possible to create "stand-alone" context for other
949 * execution environments (like system calls).
951 * lu_object methods mainly use lu_context through lu_context_key interface
952 * that allows each layer to associate arbitrary pieces of data with each
953 * context (see pthread_key_create(3) for similar interface).
955 * On a client, lu_context is bound to a thread, see cl_env_get().
957 * \see lu_context_key
961 * lu_context is used on the client side too. Yet we don't want to
962 * allocate values of server-side keys for the client contexts and
965 * To achieve this, set of tags in introduced. Contexts and keys are
966 * marked with tags. Key value are created only for context whose set
967 * of tags has non-empty intersection with one for key. Tags are taken
968 * from enum lu_context_tag.
971 enum lu_context_state lc_state;
973 * Pointer to the home service thread. NULL for other execution
976 struct ptlrpc_thread *lc_thread;
978 * Pointer to an array with key values. Internal implementation
983 * Linkage into a list of all remembered contexts. Only
984 * `non-transient' contexts, i.e., ones created for service threads
987 cfs_list_t lc_remember;
989 * Version counter used to skip calls to lu_context_refill() when no
990 * keys were registered.
1000 * lu_context_key interface. Similar to pthread_key.
1003 enum lu_context_tag {
1005 * Thread on md server
1007 LCT_MD_THREAD = 1 << 0,
1009 * Thread on dt server
1011 LCT_DT_THREAD = 1 << 1,
1013 * Context for transaction handle
1015 LCT_TX_HANDLE = 1 << 2,
1019 LCT_CL_THREAD = 1 << 3,
1021 * A per-request session on a server, and a per-system-call session on
1024 LCT_SESSION = 1 << 4,
1026 * A per-request data on OSP device
1028 LCT_OSP_THREAD = 1 << 5,
1032 LCT_MG_THREAD = 1 << 6,
1034 * Context for local operations
1038 * Set when at least one of keys, having values in this context has
1039 * non-NULL lu_context_key::lct_exit() method. This is used to
1040 * optimize lu_context_exit() call.
1042 LCT_HAS_EXIT = 1 << 28,
1044 * Don't add references for modules creating key values in that context.
1045 * This is only for contexts used internally by lu_object framework.
1047 LCT_NOREF = 1 << 29,
1049 * Key is being prepared for retiring, don't create new values for it.
1051 LCT_QUIESCENT = 1 << 30,
1053 * Context should be remembered.
1055 LCT_REMEMBER = 1 << 31,
1057 * Contexts usable in cache shrinker thread.
1059 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1063 * Key. Represents per-context value slot.
1065 * Keys are usually registered when module owning the key is initialized, and
1066 * de-registered when module is unloaded. Once key is registered, all new
1067 * contexts with matching tags, will get key value. "Old" contexts, already
1068 * initialized at the time of key registration, can be forced to get key value
1069 * by calling lu_context_refill().
1071 * Every key value is counted in lu_context_key::lct_used and acquires a
1072 * reference on an owning module. This means, that all key values have to be
1073 * destroyed before module can be unloaded. This is usually achieved by
1074 * stopping threads started by the module, that created contexts in their
1075 * entry functions. Situation is complicated by the threads shared by multiple
1076 * modules, like ptlrpcd daemon on a client. To work around this problem,
1077 * contexts, created in such threads, are `remembered' (see
1078 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1079 * for unloading it does the following:
1081 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1082 * preventing new key values from being allocated in the new contexts,
1085 * - scans a list of remembered contexts, destroying values of module
1086 * keys, thus releasing references to the module.
1088 * This is done by lu_context_key_quiesce(). If module is re-activated
1089 * before key has been de-registered, lu_context_key_revive() call clears
1090 * `quiescent' marker.
1092 * lu_context code doesn't provide any internal synchronization for these
1093 * activities---it's assumed that startup (including threads start-up) and
1094 * shutdown are serialized by some external means.
1098 struct lu_context_key {
1100 * Set of tags for which values of this key are to be instantiated.
1104 * Value constructor. This is called when new value is created for a
1105 * context. Returns pointer to new value of error pointer.
1107 void *(*lct_init)(const struct lu_context *ctx,
1108 struct lu_context_key *key);
1110 * Value destructor. Called when context with previously allocated
1111 * value of this slot is destroyed. \a data is a value that was returned
1112 * by a matching call to lu_context_key::lct_init().
1114 void (*lct_fini)(const struct lu_context *ctx,
1115 struct lu_context_key *key, void *data);
1117 * Optional method called on lu_context_exit() for all allocated
1118 * keys. Can be used by debugging code checking that locks are
1121 void (*lct_exit)(const struct lu_context *ctx,
1122 struct lu_context_key *key, void *data);
1124 * Internal implementation detail: index within lu_context::lc_value[]
1125 * reserved for this key.
1129 * Internal implementation detail: number of values created for this
1132 cfs_atomic_t lct_used;
1134 * Internal implementation detail: module for this key.
1136 cfs_module_t *lct_owner;
1138 * References to this key. For debugging.
1140 struct lu_ref lct_reference;
1143 #define LU_KEY_INIT(mod, type) \
1144 static void* mod##_key_init(const struct lu_context *ctx, \
1145 struct lu_context_key *key) \
1149 CLASSERT(PAGE_CACHE_SIZE >= sizeof (*value)); \
1151 OBD_ALLOC_PTR(value); \
1152 if (value == NULL) \
1153 value = ERR_PTR(-ENOMEM); \
1157 struct __##mod##__dummy_init {;} /* semicolon catcher */
1159 #define LU_KEY_FINI(mod, type) \
1160 static void mod##_key_fini(const struct lu_context *ctx, \
1161 struct lu_context_key *key, void* data) \
1163 type *info = data; \
1165 OBD_FREE_PTR(info); \
1167 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1169 #define LU_KEY_INIT_FINI(mod, type) \
1170 LU_KEY_INIT(mod,type); \
1171 LU_KEY_FINI(mod,type)
1173 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1174 struct lu_context_key mod##_thread_key = { \
1176 .lct_init = mod##_key_init, \
1177 .lct_fini = mod##_key_fini \
1180 #define LU_CONTEXT_KEY_INIT(key) \
1182 (key)->lct_owner = THIS_MODULE; \
1185 int lu_context_key_register(struct lu_context_key *key);
1186 void lu_context_key_degister(struct lu_context_key *key);
1187 void *lu_context_key_get (const struct lu_context *ctx,
1188 const struct lu_context_key *key);
1189 void lu_context_key_quiesce (struct lu_context_key *key);
1190 void lu_context_key_revive (struct lu_context_key *key);
1194 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1198 #define LU_KEY_INIT_GENERIC(mod) \
1199 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1201 struct lu_context_key *key = k; \
1204 va_start(args, k); \
1206 LU_CONTEXT_KEY_INIT(key); \
1207 key = va_arg(args, struct lu_context_key *); \
1208 } while (key != NULL); \
1212 #define LU_TYPE_INIT(mod, ...) \
1213 LU_KEY_INIT_GENERIC(mod) \
1214 static int mod##_type_init(struct lu_device_type *t) \
1216 mod##_key_init_generic(__VA_ARGS__, NULL); \
1217 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1219 struct __##mod##_dummy_type_init {;}
1221 #define LU_TYPE_FINI(mod, ...) \
1222 static void mod##_type_fini(struct lu_device_type *t) \
1224 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1226 struct __##mod##_dummy_type_fini {;}
1228 #define LU_TYPE_START(mod, ...) \
1229 static void mod##_type_start(struct lu_device_type *t) \
1231 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1233 struct __##mod##_dummy_type_start {;}
1235 #define LU_TYPE_STOP(mod, ...) \
1236 static void mod##_type_stop(struct lu_device_type *t) \
1238 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1240 struct __##mod##_dummy_type_stop {;}
1244 #define LU_TYPE_INIT_FINI(mod, ...) \
1245 LU_TYPE_INIT(mod, __VA_ARGS__); \
1246 LU_TYPE_FINI(mod, __VA_ARGS__); \
1247 LU_TYPE_START(mod, __VA_ARGS__); \
1248 LU_TYPE_STOP(mod, __VA_ARGS__)
1250 int lu_context_init (struct lu_context *ctx, __u32 tags);
1251 void lu_context_fini (struct lu_context *ctx);
1252 void lu_context_enter (struct lu_context *ctx);
1253 void lu_context_exit (struct lu_context *ctx);
1254 int lu_context_refill(struct lu_context *ctx);
1257 * Helper functions to operate on multiple keys. These are used by the default
1258 * device type operations, defined by LU_TYPE_INIT_FINI().
1261 int lu_context_key_register_many(struct lu_context_key *k, ...);
1262 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1263 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1264 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1267 * update/clear ctx/ses tags.
1269 void lu_context_tags_update(__u32 tags);
1270 void lu_context_tags_clear(__u32 tags);
1271 void lu_session_tags_update(__u32 tags);
1272 void lu_session_tags_clear(__u32 tags);
1279 * "Local" context, used to store data instead of stack.
1281 struct lu_context le_ctx;
1283 * "Session" context for per-request data.
1285 struct lu_context *le_ses;
1288 int lu_env_init (struct lu_env *env, __u32 tags);
1289 void lu_env_fini (struct lu_env *env);
1290 int lu_env_refill(struct lu_env *env);
1291 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1293 /** @} lu_context */
1296 * Output site statistical counters into a buffer. Suitable for
1297 * ll_rd_*()-style functions.
1299 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1302 * Common name structure to be passed around for various name related methods.
1305 const char *ln_name;
1310 * Common buffer structure to be passed around for various xattr_{s,g}et()
1318 #define DLUBUF "(%p %zu)"
1319 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1321 * One-time initializers, called at obdclass module initialization, not
1326 * Initialization of global lu_* data.
1328 int lu_global_init(void);
1331 * Dual to lu_global_init().
1333 void lu_global_fini(void);
1335 struct lu_kmem_descr {
1336 struct kmem_cache **ckd_cache;
1337 const char *ckd_name;
1338 const size_t ckd_size;
1341 int lu_kmem_init(struct lu_kmem_descr *caches);
1342 void lu_kmem_fini(struct lu_kmem_descr *caches);
1344 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1345 const struct lu_fid *fid);
1346 struct lu_object *lu_object_anon(const struct lu_env *env,
1347 struct lu_device *dev,
1348 const struct lu_object_conf *conf);
1351 extern struct lu_buf LU_BUF_NULL;
1353 void lu_buf_free(struct lu_buf *buf);
1354 void lu_buf_alloc(struct lu_buf *buf, int size);
1355 void lu_buf_realloc(struct lu_buf *buf, int size);
1357 int lu_buf_check_and_grow(struct lu_buf *buf, int len);
1358 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len);
1361 #endif /* __LUSTRE_LU_OBJECT_H */