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, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef __LUSTRE_LU_OBJECT_H
38 #define __LUSTRE_LU_OBJECT_H
45 #include <libcfs/libcfs.h>
47 #include <lustre/lustre_idl.h>
52 struct proc_dir_entry;
57 * lu_* data-types represent server-side entities shared by data and meta-data
62 * -# support for layering.
64 * Server side object is split into layers, one per device in the
65 * corresponding device stack. Individual layer is represented by struct
66 * lu_object. Compound layered object --- by struct lu_object_header. Most
67 * interface functions take lu_object as an argument and operate on the
68 * whole compound object. This decision was made due to the following
71 * - it's envisaged that lu_object will be used much more often than
74 * - we want lower (non-top) layers to be able to initiate operations
75 * on the whole object.
77 * Generic code supports layering more complex than simple stacking, e.g.,
78 * it is possible that at some layer object "spawns" multiple sub-objects
81 * -# fid-based identification.
83 * Compound object is uniquely identified by its fid. Objects are indexed
84 * by their fids (hash table is used for index).
86 * -# caching and life-cycle management.
88 * Object's life-time is controlled by reference counting. When reference
89 * count drops to 0, object is returned to cache. Cached objects still
90 * retain their identity (i.e., fid), and can be recovered from cache.
92 * Objects are kept in the global LRU list, and lu_site_purge() function
93 * can be used to reclaim given number of unused objects from the tail of
96 * -# avoiding recursion.
98 * Generic code tries to replace recursion through layers by iterations
99 * where possible. Additionally to the end of reducing stack consumption,
100 * data, when practically possible, are allocated through lu_context_key
101 * interface rather than on stack.
108 struct lu_object_header;
113 * Operations common for data and meta-data devices.
115 struct lu_device_operations {
117 * Allocate object for the given device (without lower-layer
118 * parts). This is called by lu_object_operations::loo_object_init()
119 * from the parent layer, and should setup at least lu_object::lo_dev
120 * and lu_object::lo_ops fields of resulting lu_object.
122 * Object creation protocol.
124 * Due to design goal of avoiding recursion, object creation (see
125 * lu_object_alloc()) is somewhat involved:
127 * - first, lu_device_operations::ldo_object_alloc() method of the
128 * top-level device in the stack is called. It should allocate top
129 * level object (including lu_object_header), but without any
130 * lower-layer sub-object(s).
132 * - then lu_object_alloc() sets fid in the header of newly created
135 * - then lu_object_operations::loo_object_init() is called. It has
136 * to allocate lower-layer object(s). To do this,
137 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
138 * of the lower-layer device(s).
140 * - for all new objects allocated by
141 * lu_object_operations::loo_object_init() (and inserted into object
142 * stack), lu_object_operations::loo_object_init() is called again
143 * repeatedly, until no new objects are created.
145 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
146 * result->lo_ops != NULL);
148 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
149 const struct lu_object_header *h,
150 struct lu_device *d);
152 * process config specific for device.
154 int (*ldo_process_config)(const struct lu_env *env,
155 struct lu_device *, struct lustre_cfg *);
156 int (*ldo_recovery_complete)(const struct lu_env *,
160 * initialize local objects for device. this method called after layer has
161 * been initialized (after LCFG_SETUP stage) and before it starts serving
165 int (*ldo_prepare)(const struct lu_env *,
166 struct lu_device *parent,
167 struct lu_device *dev);
172 * For lu_object_conf flags
175 /* This is a new object to be allocated, or the file
176 * corresponding to the object does not exists. */
177 LOC_F_NEW = 0x00000001,
181 * Object configuration, describing particulars of object being created. On
182 * server this is not used, as server objects are full identified by fid. On
183 * client configuration contains struct lustre_md.
185 struct lu_object_conf {
187 * Some hints for obj find and alloc.
189 loc_flags_t loc_flags;
193 * Type of "printer" function used by lu_object_operations::loo_object_print()
196 * Printer function is needed to provide some flexibility in (semi-)debugging
197 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
199 typedef int (*lu_printer_t)(const struct lu_env *env,
200 void *cookie, const char *format, ...)
201 __attribute__ ((format (printf, 3, 4)));
204 * Operations specific for particular lu_object.
206 struct lu_object_operations {
209 * Allocate lower-layer parts of the object by calling
210 * lu_device_operations::ldo_object_alloc() of the corresponding
213 * This method is called once for each object inserted into object
214 * stack. It's responsibility of this method to insert lower-layer
215 * object(s) it create into appropriate places of object stack.
217 int (*loo_object_init)(const struct lu_env *env,
219 const struct lu_object_conf *conf);
221 * Called (in top-to-bottom order) during object allocation after all
222 * layers were allocated and initialized. Can be used to perform
223 * initialization depending on lower layers.
225 int (*loo_object_start)(const struct lu_env *env,
226 struct lu_object *o);
228 * Called before lu_object_operations::loo_object_free() to signal
229 * that object is being destroyed. Dual to
230 * lu_object_operations::loo_object_init().
232 void (*loo_object_delete)(const struct lu_env *env,
233 struct lu_object *o);
235 * Dual to lu_device_operations::ldo_object_alloc(). Called when
236 * object is removed from memory.
238 void (*loo_object_free)(const struct lu_env *env,
239 struct lu_object *o);
241 * Called when last active reference to the object is released (and
242 * object returns to the cache). This method is optional.
244 void (*loo_object_release)(const struct lu_env *env,
245 struct lu_object *o);
247 * Optional debugging helper. Print given object.
249 int (*loo_object_print)(const struct lu_env *env, void *cookie,
250 lu_printer_t p, const struct lu_object *o);
252 * Optional debugging method. Returns true iff method is internally
255 int (*loo_object_invariant)(const struct lu_object *o);
261 struct lu_device_type;
264 * Device: a layer in the server side abstraction stacking.
268 * reference count. This is incremented, in particular, on each object
269 * created at this layer.
271 * \todo XXX which means that atomic_t is probably too small.
275 * Pointer to device type. Never modified once set.
277 struct lu_device_type *ld_type;
279 * Operation vector for this device.
281 const struct lu_device_operations *ld_ops;
283 * Stack this device belongs to.
285 struct lu_site *ld_site;
286 struct proc_dir_entry *ld_proc_entry;
288 /** \todo XXX: temporary back pointer into obd. */
289 struct obd_device *ld_obd;
291 * A list of references to this object, for debugging.
293 struct lu_ref ld_reference;
295 * Link the device to the site.
297 cfs_list_t ld_linkage;
300 struct lu_device_type_operations;
303 * Tag bits for device type. They are used to distinguish certain groups of
307 /** this is meta-data device */
308 LU_DEVICE_MD = (1 << 0),
309 /** this is data device */
310 LU_DEVICE_DT = (1 << 1),
311 /** data device in the client stack */
312 LU_DEVICE_CL = (1 << 2)
318 struct lu_device_type {
320 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
324 * Name of this class. Unique system-wide. Never modified once set.
328 * Operations for this type.
330 const struct lu_device_type_operations *ldt_ops;
332 * \todo XXX: temporary pointer to associated obd_type.
334 struct obd_type *ldt_obd_type;
336 * \todo XXX: temporary: context tags used by obd_*() calls.
340 * Number of existing device type instances.
342 unsigned ldt_device_nr;
344 * Linkage into a global list of all device types.
346 * \see lu_device_types.
348 cfs_list_t ldt_linkage;
352 * Operations on a device type.
354 struct lu_device_type_operations {
356 * Allocate new device.
358 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
359 struct lu_device_type *t,
360 struct lustre_cfg *lcfg);
362 * Free device. Dual to
363 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
364 * the next device in the stack.
366 struct lu_device *(*ldto_device_free)(const struct lu_env *,
370 * Initialize the devices after allocation
372 int (*ldto_device_init)(const struct lu_env *env,
373 struct lu_device *, const char *,
376 * Finalize device. Dual to
377 * lu_device_type_operations::ldto_device_init(). Returns pointer to
378 * the next device in the stack.
380 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
383 * Initialize device type. This is called on module load.
385 int (*ldto_init)(struct lu_device_type *t);
387 * Finalize device type. Dual to
388 * lu_device_type_operations::ldto_init(). Called on module unload.
390 void (*ldto_fini)(struct lu_device_type *t);
392 * Called when the first device is created.
394 void (*ldto_start)(struct lu_device_type *t);
396 * Called when number of devices drops to 0.
398 void (*ldto_stop)(struct lu_device_type *t);
402 * Flags for the object layers.
404 enum lu_object_flags {
406 * this flags is set if lu_object_operations::loo_object_init() has
407 * been called for this layer. Used by lu_object_alloc().
409 LU_OBJECT_ALLOCATED = (1 << 0)
413 * Common object attributes.
418 /** modification time in seconds since Epoch */
420 /** access time in seconds since Epoch */
422 /** change time in seconds since Epoch */
424 /** 512-byte blocks allocated to object */
426 /** permission bits and file type */
434 /** number of persistent references to this object */
436 /** blk bits of the object*/
438 /** blk size of the object*/
450 /** Bit-mask of valid attributes */
464 LA_BLKSIZE = 1 << 12,
465 LA_KILL_SUID = 1 << 13,
466 LA_KILL_SGID = 1 << 14,
470 * Layer in the layered object.
474 * Header for this object.
476 struct lu_object_header *lo_header;
478 * Device for this layer.
480 struct lu_device *lo_dev;
482 * Operations for this object.
484 const struct lu_object_operations *lo_ops;
486 * Linkage into list of all layers.
488 cfs_list_t lo_linkage;
490 * Depth. Top level layer depth is 0.
494 * Flags from enum lu_object_flags.
496 unsigned long lo_flags;
498 * Link to the device, for debugging.
500 struct lu_ref_link *lo_dev_ref;
503 enum lu_object_header_flags {
505 * Don't keep this object in cache. Object will be destroyed as soon
506 * as last reference to it is released. This flag cannot be cleared
509 LU_OBJECT_HEARD_BANSHEE = 0
512 enum lu_object_header_attr {
513 LOHA_EXISTS = 1 << 0,
514 LOHA_REMOTE = 1 << 1,
516 * UNIX file type is stored in S_IFMT bits.
518 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
519 LOHA_FT_END = 017 << 12, /**< S_IFMT */
523 * "Compound" object, consisting of multiple layers.
525 * Compound object with given fid is unique with given lu_site.
527 * Note, that object does *not* necessary correspond to the real object in the
528 * persistent storage: object is an anchor for locking and method calling, so
529 * it is created for things like not-yet-existing child created by mkdir or
530 * create calls. lu_object_operations::loo_exists() can be used to check
531 * whether object is backed by persistent storage entity.
533 struct lu_object_header {
535 * Object flags from enum lu_object_header_flags. Set and checked
538 unsigned long loh_flags;
540 * Object reference count. Protected by lu_site::ls_guard.
542 cfs_atomic_t loh_ref;
544 * Fid, uniquely identifying this object.
546 struct lu_fid loh_fid;
548 * Common object attributes, cached for efficiency. From enum
549 * lu_object_header_attr.
553 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
555 cfs_hlist_node_t loh_hash;
557 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
561 * Linkage into list of layers. Never modified once set (except lately
562 * during object destruction). No locking is necessary.
564 cfs_list_t loh_layers;
566 * A list of references to this object, for debugging.
568 struct lu_ref loh_reference;
573 struct lu_site_bkt_data {
575 * number of busy object on this bucket
579 * LRU list, updated on each access to object. Protected by
580 * bucket lock of lu_site::ls_obj_hash.
582 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
583 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
584 * of list_for_each_entry_safe_reverse()).
588 * Wait-queue signaled when an object in this site is ultimately
589 * destroyed (lu_object_free()). It is used by lu_object_find() to
590 * wait before re-trying when object in the process of destruction is
591 * found in the hash table.
593 * \see htable_lookup().
595 cfs_waitq_t lsb_marche_funebre;
603 LU_SS_CACHE_DEATH_RACE,
609 * lu_site is a "compartment" within which objects are unique, and LRU
610 * discipline is maintained.
612 * lu_site exists so that multiple layered stacks can co-exist in the same
615 * lu_site has the same relation to lu_device as lu_object_header to
622 cfs_hash_t *ls_obj_hash;
624 * index of bucket on hash table while purging
628 * Top-level device for this stack.
630 struct lu_device *ls_top_dev;
632 * Bottom-level device for this stack
634 struct lu_device *ls_bottom_dev;
636 * Linkage into global list of sites.
638 cfs_list_t ls_linkage;
640 * List for lu device for this site, protected
643 cfs_list_t ls_ld_linkage;
644 spinlock_t ls_ld_lock;
649 struct lprocfs_stats *ls_stats;
650 struct lprocfs_stats *ls_time_stats;
652 * XXX: a hack! fld has to find md_site via site, remove when possible
654 struct md_site *ld_md_site;
657 static inline struct lu_site_bkt_data *
658 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
662 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
663 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
667 * Constructors/destructors.
671 int lu_site_init (struct lu_site *s, struct lu_device *d);
672 void lu_site_fini (struct lu_site *s);
673 int lu_site_init_finish (struct lu_site *s);
674 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
675 void lu_device_get (struct lu_device *d);
676 void lu_device_put (struct lu_device *d);
677 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
678 void lu_device_fini (struct lu_device *d);
679 int lu_object_header_init(struct lu_object_header *h);
680 void lu_object_header_fini(struct lu_object_header *h);
681 int lu_object_init (struct lu_object *o,
682 struct lu_object_header *h, struct lu_device *d);
683 void lu_object_fini (struct lu_object *o);
684 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
685 void lu_object_add (struct lu_object *before, struct lu_object *o);
687 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
688 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
691 * Helpers to initialize and finalize device types.
694 int lu_device_type_init(struct lu_device_type *ldt);
695 void lu_device_type_fini(struct lu_device_type *ldt);
696 void lu_types_stop(void);
701 * Caching and reference counting.
706 * Acquire additional reference to the given object. This function is used to
707 * attain additional reference. To acquire initial reference use
710 static inline void lu_object_get(struct lu_object *o)
712 LASSERT(cfs_atomic_read(&o->lo_header->loh_ref) > 0);
713 cfs_atomic_inc(&o->lo_header->loh_ref);
717 * Return true of object will not be cached after last reference to it is
720 static inline int lu_object_is_dying(const struct lu_object_header *h)
722 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
725 void lu_object_put(const struct lu_env *env, struct lu_object *o);
726 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
728 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
730 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
731 lu_printer_t printer);
732 struct lu_object *lu_object_find(const struct lu_env *env,
733 struct lu_device *dev, const struct lu_fid *f,
734 const struct lu_object_conf *conf);
735 struct lu_object *lu_object_find_at(const struct lu_env *env,
736 struct lu_device *dev,
737 const struct lu_fid *f,
738 const struct lu_object_conf *conf);
739 struct lu_object *lu_object_find_slice(const struct lu_env *env,
740 struct lu_device *dev,
741 const struct lu_fid *f,
742 const struct lu_object_conf *conf);
751 * First (topmost) sub-object of given compound object
753 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
755 LASSERT(!cfs_list_empty(&h->loh_layers));
756 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
760 * Next sub-object in the layering
762 static inline struct lu_object *lu_object_next(const struct lu_object *o)
764 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
768 * Pointer to the fid of this object.
770 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
772 return &o->lo_header->loh_fid;
776 * return device operations vector for this object
778 static const inline struct lu_device_operations *
779 lu_object_ops(const struct lu_object *o)
781 return o->lo_dev->ld_ops;
785 * Given a compound object, find its slice, corresponding to the device type
788 struct lu_object *lu_object_locate(struct lu_object_header *h,
789 const struct lu_device_type *dtype);
792 * Printer function emitting messages through libcfs_debug_msg().
794 int lu_cdebug_printer(const struct lu_env *env,
795 void *cookie, const char *format, ...);
798 * Print object description followed by a user-supplied message.
800 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
802 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
804 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
805 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
806 CDEBUG(mask, format , ## __VA_ARGS__); \
811 * Print short object description followed by a user-supplied message.
813 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
815 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
817 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
818 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
819 (object)->lo_header); \
820 lu_cdebug_printer(env, &msgdata, "\n"); \
821 CDEBUG(mask, format , ## __VA_ARGS__); \
825 void lu_object_print (const struct lu_env *env, void *cookie,
826 lu_printer_t printer, const struct lu_object *o);
827 void lu_object_header_print(const struct lu_env *env, void *cookie,
828 lu_printer_t printer,
829 const struct lu_object_header *hdr);
832 * Check object consistency.
834 int lu_object_invariant(const struct lu_object *o);
838 * \retval 1 iff object \a o exists on stable storage,
839 * \retval 0 iff object \a o not exists on stable storage.
840 * \retval -1 iff object \a o is on remote server.
842 static inline int lu_object_exists(const struct lu_object *o)
846 attr = o->lo_header->loh_attr;
847 if (attr & LOHA_REMOTE)
849 else if (attr & LOHA_EXISTS)
855 static inline int lu_object_assert_exists(const struct lu_object *o)
857 return lu_object_exists(o) != 0;
860 static inline int lu_object_assert_not_exists(const struct lu_object *o)
862 return lu_object_exists(o) <= 0;
866 * Attr of this object.
868 static inline __u32 lu_object_attr(const struct lu_object *o)
870 LASSERT(lu_object_exists(o) > 0);
871 return o->lo_header->loh_attr;
874 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
878 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
881 static inline void lu_object_ref_del(struct lu_object *o,
882 const char *scope, const void *source)
884 lu_ref_del(&o->lo_header->loh_reference, scope, source);
887 static inline void lu_object_ref_del_at(struct lu_object *o,
888 struct lu_ref_link *link,
889 const char *scope, const void *source)
891 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
894 /** input params, should be filled out by mdt */
898 /** count in bytes */
899 unsigned int rp_count;
900 /** number of pages */
901 unsigned int rp_npages;
902 /** requested attr */
904 /** pointers to pages */
905 struct page **rp_pages;
908 enum lu_xattr_flags {
909 LU_XATTR_REPLACE = (1 << 0),
910 LU_XATTR_CREATE = (1 << 1)
918 /** For lu_context health-checks */
919 enum lu_context_state {
927 * lu_context. Execution context for lu_object methods. Currently associated
930 * All lu_object methods, except device and device type methods (called during
931 * system initialization and shutdown) are executed "within" some
932 * lu_context. This means, that pointer to some "current" lu_context is passed
933 * as an argument to all methods.
935 * All service ptlrpc threads create lu_context as part of their
936 * initialization. It is possible to create "stand-alone" context for other
937 * execution environments (like system calls).
939 * lu_object methods mainly use lu_context through lu_context_key interface
940 * that allows each layer to associate arbitrary pieces of data with each
941 * context (see pthread_key_create(3) for similar interface).
943 * On a client, lu_context is bound to a thread, see cl_env_get().
945 * \see lu_context_key
949 * lu_context is used on the client side too. Yet we don't want to
950 * allocate values of server-side keys for the client contexts and
953 * To achieve this, set of tags in introduced. Contexts and keys are
954 * marked with tags. Key value are created only for context whose set
955 * of tags has non-empty intersection with one for key. Tags are taken
956 * from enum lu_context_tag.
959 enum lu_context_state lc_state;
961 * Pointer to the home service thread. NULL for other execution
964 struct ptlrpc_thread *lc_thread;
966 * Pointer to an array with key values. Internal implementation
971 * Linkage into a list of all remembered contexts. Only
972 * `non-transient' contexts, i.e., ones created for service threads
975 cfs_list_t lc_remember;
977 * Version counter used to skip calls to lu_context_refill() when no
978 * keys were registered.
988 * lu_context_key interface. Similar to pthread_key.
991 enum lu_context_tag {
993 * Thread on md server
995 LCT_MD_THREAD = 1 << 0,
997 * Thread on dt server
999 LCT_DT_THREAD = 1 << 1,
1001 * Context for transaction handle
1003 LCT_TX_HANDLE = 1 << 2,
1007 LCT_CL_THREAD = 1 << 3,
1009 * A per-request session on a server, and a per-system-call session on
1012 LCT_SESSION = 1 << 4,
1014 * A per-request data on OSP device
1016 LCT_OSP_THREAD = 1 << 5,
1020 LCT_MG_THREAD = 1 << 6,
1022 * Context for local operations
1026 * Set when at least one of keys, having values in this context has
1027 * non-NULL lu_context_key::lct_exit() method. This is used to
1028 * optimize lu_context_exit() call.
1030 LCT_HAS_EXIT = 1 << 28,
1032 * Don't add references for modules creating key values in that context.
1033 * This is only for contexts used internally by lu_object framework.
1035 LCT_NOREF = 1 << 29,
1037 * Key is being prepared for retiring, don't create new values for it.
1039 LCT_QUIESCENT = 1 << 30,
1041 * Context should be remembered.
1043 LCT_REMEMBER = 1 << 31,
1045 * Contexts usable in cache shrinker thread.
1047 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1051 * Key. Represents per-context value slot.
1053 * Keys are usually registered when module owning the key is initialized, and
1054 * de-registered when module is unloaded. Once key is registered, all new
1055 * contexts with matching tags, will get key value. "Old" contexts, already
1056 * initialized at the time of key registration, can be forced to get key value
1057 * by calling lu_context_refill().
1059 * Every key value is counted in lu_context_key::lct_used and acquires a
1060 * reference on an owning module. This means, that all key values have to be
1061 * destroyed before module can be unloaded. This is usually achieved by
1062 * stopping threads started by the module, that created contexts in their
1063 * entry functions. Situation is complicated by the threads shared by multiple
1064 * modules, like ptlrpcd daemon on a client. To work around this problem,
1065 * contexts, created in such threads, are `remembered' (see
1066 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1067 * for unloading it does the following:
1069 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1070 * preventing new key values from being allocated in the new contexts,
1073 * - scans a list of remembered contexts, destroying values of module
1074 * keys, thus releasing references to the module.
1076 * This is done by lu_context_key_quiesce(). If module is re-activated
1077 * before key has been de-registered, lu_context_key_revive() call clears
1078 * `quiescent' marker.
1080 * lu_context code doesn't provide any internal synchronization for these
1081 * activities---it's assumed that startup (including threads start-up) and
1082 * shutdown are serialized by some external means.
1086 struct lu_context_key {
1088 * Set of tags for which values of this key are to be instantiated.
1092 * Value constructor. This is called when new value is created for a
1093 * context. Returns pointer to new value of error pointer.
1095 void *(*lct_init)(const struct lu_context *ctx,
1096 struct lu_context_key *key);
1098 * Value destructor. Called when context with previously allocated
1099 * value of this slot is destroyed. \a data is a value that was returned
1100 * by a matching call to lu_context_key::lct_init().
1102 void (*lct_fini)(const struct lu_context *ctx,
1103 struct lu_context_key *key, void *data);
1105 * Optional method called on lu_context_exit() for all allocated
1106 * keys. Can be used by debugging code checking that locks are
1109 void (*lct_exit)(const struct lu_context *ctx,
1110 struct lu_context_key *key, void *data);
1112 * Internal implementation detail: index within lu_context::lc_value[]
1113 * reserved for this key.
1117 * Internal implementation detail: number of values created for this
1120 cfs_atomic_t lct_used;
1122 * Internal implementation detail: module for this key.
1124 cfs_module_t *lct_owner;
1126 * References to this key. For debugging.
1128 struct lu_ref lct_reference;
1131 #define LU_KEY_INIT(mod, type) \
1132 static void* mod##_key_init(const struct lu_context *ctx, \
1133 struct lu_context_key *key) \
1137 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1139 OBD_ALLOC_PTR(value); \
1140 if (value == NULL) \
1141 value = ERR_PTR(-ENOMEM); \
1145 struct __##mod##__dummy_init {;} /* semicolon catcher */
1147 #define LU_KEY_FINI(mod, type) \
1148 static void mod##_key_fini(const struct lu_context *ctx, \
1149 struct lu_context_key *key, void* data) \
1151 type *info = data; \
1153 OBD_FREE_PTR(info); \
1155 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1157 #define LU_KEY_INIT_FINI(mod, type) \
1158 LU_KEY_INIT(mod,type); \
1159 LU_KEY_FINI(mod,type)
1161 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1162 struct lu_context_key mod##_thread_key = { \
1164 .lct_init = mod##_key_init, \
1165 .lct_fini = mod##_key_fini \
1168 #define LU_CONTEXT_KEY_INIT(key) \
1170 (key)->lct_owner = THIS_MODULE; \
1173 int lu_context_key_register(struct lu_context_key *key);
1174 void lu_context_key_degister(struct lu_context_key *key);
1175 void *lu_context_key_get (const struct lu_context *ctx,
1176 const struct lu_context_key *key);
1177 void lu_context_key_quiesce (struct lu_context_key *key);
1178 void lu_context_key_revive (struct lu_context_key *key);
1182 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1186 #define LU_KEY_INIT_GENERIC(mod) \
1187 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1189 struct lu_context_key *key = k; \
1192 va_start(args, k); \
1194 LU_CONTEXT_KEY_INIT(key); \
1195 key = va_arg(args, struct lu_context_key *); \
1196 } while (key != NULL); \
1200 #define LU_TYPE_INIT(mod, ...) \
1201 LU_KEY_INIT_GENERIC(mod) \
1202 static int mod##_type_init(struct lu_device_type *t) \
1204 mod##_key_init_generic(__VA_ARGS__, NULL); \
1205 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1207 struct __##mod##_dummy_type_init {;}
1209 #define LU_TYPE_FINI(mod, ...) \
1210 static void mod##_type_fini(struct lu_device_type *t) \
1212 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1214 struct __##mod##_dummy_type_fini {;}
1216 #define LU_TYPE_START(mod, ...) \
1217 static void mod##_type_start(struct lu_device_type *t) \
1219 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1221 struct __##mod##_dummy_type_start {;}
1223 #define LU_TYPE_STOP(mod, ...) \
1224 static void mod##_type_stop(struct lu_device_type *t) \
1226 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1228 struct __##mod##_dummy_type_stop {;}
1232 #define LU_TYPE_INIT_FINI(mod, ...) \
1233 LU_TYPE_INIT(mod, __VA_ARGS__); \
1234 LU_TYPE_FINI(mod, __VA_ARGS__); \
1235 LU_TYPE_START(mod, __VA_ARGS__); \
1236 LU_TYPE_STOP(mod, __VA_ARGS__)
1238 int lu_context_init (struct lu_context *ctx, __u32 tags);
1239 void lu_context_fini (struct lu_context *ctx);
1240 void lu_context_enter (struct lu_context *ctx);
1241 void lu_context_exit (struct lu_context *ctx);
1242 int lu_context_refill(struct lu_context *ctx);
1245 * Helper functions to operate on multiple keys. These are used by the default
1246 * device type operations, defined by LU_TYPE_INIT_FINI().
1249 int lu_context_key_register_many(struct lu_context_key *k, ...);
1250 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1251 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1252 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1255 * update/clear ctx/ses tags.
1257 void lu_context_tags_update(__u32 tags);
1258 void lu_context_tags_clear(__u32 tags);
1259 void lu_session_tags_update(__u32 tags);
1260 void lu_session_tags_clear(__u32 tags);
1267 * "Local" context, used to store data instead of stack.
1269 struct lu_context le_ctx;
1271 * "Session" context for per-request data.
1273 struct lu_context *le_ses;
1276 int lu_env_init (struct lu_env *env, __u32 tags);
1277 void lu_env_fini (struct lu_env *env);
1278 int lu_env_refill(struct lu_env *env);
1279 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1281 /** @} lu_context */
1293 __u32 uc_suppgids[2];
1296 cfs_group_info_t *uc_ginfo;
1297 struct md_identity *uc_identity;
1299 struct lu_ucred *lu_ucred(const struct lu_env *env);
1300 struct lu_ucred *lu_ucred_check(const struct lu_env *env);
1301 struct lu_ucred *lu_ucred_assert(const struct lu_env *env);
1304 * Output site statistical counters into a buffer. Suitable for
1305 * ll_rd_*()-style functions.
1307 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1310 * Common name structure to be passed around for various name related methods.
1313 const char *ln_name;
1318 * Common buffer structure to be passed around for various xattr_{s,g}et()
1327 extern struct lu_buf LU_BUF_NULL;
1329 #define DLUBUF "(%p %zu)"
1330 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1332 * One-time initializers, called at obdclass module initialization, not
1337 * Initialization of global lu_* data.
1339 int lu_global_init(void);
1342 * Dual to lu_global_init().
1344 void lu_global_fini(void);
1347 LU_TIME_FIND_LOOKUP,
1349 LU_TIME_FIND_INSERT,
1353 extern const char *lu_time_names[LU_TIME_NR];
1355 struct lu_kmem_descr {
1356 cfs_mem_cache_t **ckd_cache;
1357 const char *ckd_name;
1358 const size_t ckd_size;
1361 int lu_kmem_init(struct lu_kmem_descr *caches);
1362 void lu_kmem_fini(struct lu_kmem_descr *caches);
1364 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1365 const struct lu_fid *fid);
1366 struct lu_object *lu_object_anon(const struct lu_env *env,
1367 struct lu_device *dev,
1368 const struct lu_object_conf *conf);
1371 #endif /* __LUSTRE_LU_OBJECT_H */