1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
39 #ifndef __LUSTRE_LU_OBJECT_H
40 #define __LUSTRE_LU_OBJECT_H
47 #include <libcfs/libcfs.h>
49 #include <lustre/lustre_idl.h>
54 struct proc_dir_entry;
59 * lu_* data-types represent server-side entities shared by data and meta-data
64 * -# support for layering.
66 * Server side object is split into layers, one per device in the
67 * corresponding device stack. Individual layer is represented by struct
68 * lu_object. Compound layered object --- by struct lu_object_header. Most
69 * interface functions take lu_object as an argument and operate on the
70 * whole compound object. This decision was made due to the following
73 * - it's envisaged that lu_object will be used much more often than
76 * - we want lower (non-top) layers to be able to initiate operations
77 * on the whole object.
79 * Generic code supports layering more complex than simple stacking, e.g.,
80 * it is possible that at some layer object "spawns" multiple sub-objects
83 * -# fid-based identification.
85 * Compound object is uniquely identified by its fid. Objects are indexed
86 * by their fids (hash table is used for index).
88 * -# caching and life-cycle management.
90 * Object's life-time is controlled by reference counting. When reference
91 * count drops to 0, object is returned to cache. Cached objects still
92 * retain their identity (i.e., fid), and can be recovered from cache.
94 * Objects are kept in the global LRU list, and lu_site_purge() function
95 * can be used to reclaim given number of unused objects from the tail of
98 * -# avoiding recursion.
100 * Generic code tries to replace recursion through layers by iterations
101 * where possible. Additionally to the end of reducing stack consumption,
102 * data, when practically possible, are allocated through lu_context_key
103 * interface rather than on stack.
110 struct lu_object_header;
115 * Operations common for data and meta-data devices.
117 struct lu_device_operations {
119 * Allocate object for the given device (without lower-layer
120 * parts). This is called by lu_object_operations::loo_object_init()
121 * from the parent layer, and should setup at least lu_object::lo_dev
122 * and lu_object::lo_ops fields of resulting lu_object.
124 * Object creation protocol.
126 * Due to design goal of avoiding recursion, object creation (see
127 * lu_object_alloc()) is somewhat involved:
129 * - first, lu_device_operations::ldo_object_alloc() method of the
130 * top-level device in the stack is called. It should allocate top
131 * level object (including lu_object_header), but without any
132 * lower-layer sub-object(s).
134 * - then lu_object_alloc() sets fid in the header of newly created
137 * - then lu_object_operations::loo_object_init() is called. It has
138 * to allocate lower-layer object(s). To do this,
139 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
140 * of the lower-layer device(s).
142 * - for all new objects allocated by
143 * lu_object_operations::loo_object_init() (and inserted into object
144 * stack), lu_object_operations::loo_object_init() is called again
145 * repeatedly, until no new objects are created.
147 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
148 * result->lo_ops != NULL);
150 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
151 const struct lu_object_header *h,
152 struct lu_device *d);
154 * process config specific for device.
156 int (*ldo_process_config)(const struct lu_env *env,
157 struct lu_device *, struct lustre_cfg *);
158 int (*ldo_recovery_complete)(const struct lu_env *,
162 * initialize local objects for device. this method called after layer has
163 * been initialized (after LCFG_SETUP stage) and before it starts serving
167 int (*ldo_prepare)(const struct lu_env *,
168 struct lu_device *parent,
169 struct lu_device *dev);
174 * For lu_object_conf flags
177 /* Currently, only used for client-side object initialization. */
182 * Object configuration, describing particulars of object being created. On
183 * server this is not used, as server objects are full identified by fid. On
184 * client configuration contains struct lustre_md.
186 struct lu_object_conf {
188 * Some hints for obj find and alloc.
190 loc_flags_t loc_flags;
194 * Type of "printer" function used by lu_object_operations::loo_object_print()
197 * Printer function is needed to provide some flexibility in (semi-)debugging
198 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
200 typedef int (*lu_printer_t)(const struct lu_env *env,
201 void *cookie, const char *format, ...)
202 __attribute__ ((format (printf, 3, 4)));
205 * Operations specific for particular lu_object.
207 struct lu_object_operations {
210 * Allocate lower-layer parts of the object by calling
211 * lu_device_operations::ldo_object_alloc() of the corresponding
214 * This method is called once for each object inserted into object
215 * stack. It's responsibility of this method to insert lower-layer
216 * object(s) it create into appropriate places of object stack.
218 int (*loo_object_init)(const struct lu_env *env,
220 const struct lu_object_conf *conf);
222 * Called (in top-to-bottom order) during object allocation after all
223 * layers were allocated and initialized. Can be used to perform
224 * initialization depending on lower layers.
226 int (*loo_object_start)(const struct lu_env *env,
227 struct lu_object *o);
229 * Called before lu_object_operations::loo_object_free() to signal
230 * that object is being destroyed. Dual to
231 * lu_object_operations::loo_object_init().
233 void (*loo_object_delete)(const struct lu_env *env,
234 struct lu_object *o);
236 * Dual to lu_device_operations::ldo_object_alloc(). Called when
237 * object is removed from memory.
239 void (*loo_object_free)(const struct lu_env *env,
240 struct lu_object *o);
242 * Called when last active reference to the object is released (and
243 * object returns to the cache). This method is optional.
245 void (*loo_object_release)(const struct lu_env *env,
246 struct lu_object *o);
248 * Optional debugging helper. Print given object.
250 int (*loo_object_print)(const struct lu_env *env, void *cookie,
251 lu_printer_t p, const struct lu_object *o);
253 * Optional debugging method. Returns true iff method is internally
256 int (*loo_object_invariant)(const struct lu_object *o);
262 struct lu_device_type;
265 * Device: a layer in the server side abstraction stacking.
269 * reference count. This is incremented, in particular, on each object
270 * created at this layer.
272 * \todo XXX which means that atomic_t is probably too small.
276 * Pointer to device type. Never modified once set.
278 struct lu_device_type *ld_type;
280 * Operation vector for this device.
282 const struct lu_device_operations *ld_ops;
284 * Stack this device belongs to.
286 struct lu_site *ld_site;
287 struct proc_dir_entry *ld_proc_entry;
289 /** \todo XXX: temporary back pointer into obd. */
290 struct obd_device *ld_obd;
292 * A list of references to this object, for debugging.
294 struct lu_ref ld_reference;
296 * Link the device to the site.
298 cfs_list_t ld_linkage;
301 struct lu_device_type_operations;
304 * Tag bits for device type. They are used to distinguish certain groups of
308 /** this is meta-data device */
309 LU_DEVICE_MD = (1 << 0),
310 /** this is data device */
311 LU_DEVICE_DT = (1 << 1),
312 /** data device in the client stack */
313 LU_DEVICE_CL = (1 << 2)
319 struct lu_device_type {
321 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
325 * Name of this class. Unique system-wide. Never modified once set.
329 * Operations for this type.
331 const struct lu_device_type_operations *ldt_ops;
333 * \todo XXX: temporary pointer to associated obd_type.
335 struct obd_type *ldt_obd_type;
337 * \todo XXX: temporary: context tags used by obd_*() calls.
341 * Number of existing device type instances.
343 unsigned ldt_device_nr;
345 * Linkage into a global list of all device types.
347 * \see lu_device_types.
349 cfs_list_t ldt_linkage;
353 * Operations on a device type.
355 struct lu_device_type_operations {
357 * Allocate new device.
359 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
360 struct lu_device_type *t,
361 struct lustre_cfg *lcfg);
363 * Free device. Dual to
364 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
365 * the next device in the stack.
367 struct lu_device *(*ldto_device_free)(const struct lu_env *,
371 * Initialize the devices after allocation
373 int (*ldto_device_init)(const struct lu_env *env,
374 struct lu_device *, const char *,
377 * Finalize device. Dual to
378 * lu_device_type_operations::ldto_device_init(). Returns pointer to
379 * the next device in the stack.
381 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
384 * Initialize device type. This is called on module load.
386 int (*ldto_init)(struct lu_device_type *t);
388 * Finalize device type. Dual to
389 * lu_device_type_operations::ldto_init(). Called on module unload.
391 void (*ldto_fini)(struct lu_device_type *t);
393 * Called when the first device is created.
395 void (*ldto_start)(struct lu_device_type *t);
397 * Called when number of devices drops to 0.
399 void (*ldto_stop)(struct lu_device_type *t);
403 * Flags for the object layers.
405 enum lu_object_flags {
407 * this flags is set if lu_object_operations::loo_object_init() has
408 * been called for this layer. Used by lu_object_alloc().
410 LU_OBJECT_ALLOCATED = (1 << 0)
414 * Common object attributes.
419 /** modification time in seconds since Epoch */
421 /** access time in seconds since Epoch */
423 /** change time in seconds since Epoch */
425 /** 512-byte blocks allocated to object */
427 /** permission bits and file type */
435 /** number of persistent references to this object */
437 /** blk bits of the object*/
439 /** blk size of the object*/
451 /** Bit-mask of valid attributes */
465 LA_BLKSIZE = 1 << 12,
466 LA_KILL_SUID = 1 << 13,
467 LA_KILL_SGID = 1 << 14,
471 * Layer in the layered object.
475 * Header for this object.
477 struct lu_object_header *lo_header;
479 * Device for this layer.
481 struct lu_device *lo_dev;
483 * Operations for this object.
485 const struct lu_object_operations *lo_ops;
487 * Linkage into list of all layers.
489 cfs_list_t lo_linkage;
491 * Depth. Top level layer depth is 0.
495 * Flags from enum lu_object_flags.
497 unsigned long lo_flags;
499 * Link to the device, for debugging.
501 struct lu_ref_link *lo_dev_ref;
504 enum lu_object_header_flags {
506 * Don't keep this object in cache. Object will be destroyed as soon
507 * as last reference to it is released. This flag cannot be cleared
510 LU_OBJECT_HEARD_BANSHEE = 0
513 enum lu_object_header_attr {
514 LOHA_EXISTS = 1 << 0,
515 LOHA_REMOTE = 1 << 1,
517 * UNIX file type is stored in S_IFMT bits.
519 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
520 LOHA_FT_END = 017 << 12, /**< S_IFMT */
524 * "Compound" object, consisting of multiple layers.
526 * Compound object with given fid is unique with given lu_site.
528 * Note, that object does *not* necessary correspond to the real object in the
529 * persistent storage: object is an anchor for locking and method calling, so
530 * it is created for things like not-yet-existing child created by mkdir or
531 * create calls. lu_object_operations::loo_exists() can be used to check
532 * whether object is backed by persistent storage entity.
534 struct lu_object_header {
536 * Object flags from enum lu_object_header_flags. Set and checked
539 unsigned long loh_flags;
541 * Object reference count. Protected by lu_site::ls_guard.
543 cfs_atomic_t loh_ref;
545 * Fid, uniquely identifying this object.
547 struct lu_fid loh_fid;
549 * Common object attributes, cached for efficiency. From enum
550 * lu_object_header_attr.
554 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
556 cfs_hlist_node_t loh_hash;
558 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
562 * Linkage into list of layers. Never modified once set (except lately
563 * during object destruction). No locking is necessary.
565 cfs_list_t loh_layers;
567 * A list of references to this object, for debugging.
569 struct lu_ref loh_reference;
574 struct lu_site_bkt_data {
576 * number of busy object on this bucket
580 * LRU list, updated on each access to object. Protected by
581 * bucket lock of lu_site::ls_obj_hash.
583 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
584 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
585 * of list_for_each_entry_safe_reverse()).
589 * Wait-queue signaled when an object in this site is ultimately
590 * destroyed (lu_object_free()). It is used by lu_object_find() to
591 * wait before re-trying when object in the process of destruction is
592 * found in the hash table.
594 * \see htable_lookup().
596 cfs_waitq_t lsb_marche_funebre;
604 LU_SS_CACHE_DEATH_RACE,
610 * lu_site is a "compartment" within which objects are unique, and LRU
611 * discipline is maintained.
613 * lu_site exists so that multiple layered stacks can co-exist in the same
616 * lu_site has the same relation to lu_device as lu_object_header to
623 cfs_hash_t *ls_obj_hash;
625 * index of bucket on hash table while purging
629 * Top-level device for this stack.
631 struct lu_device *ls_top_dev;
633 * Linkage into global list of sites.
635 cfs_list_t ls_linkage;
637 * List for lu device for this site, protected
640 cfs_list_t ls_ld_linkage;
641 cfs_spinlock_t ls_ld_lock;
646 struct lprocfs_stats *ls_stats;
647 struct lprocfs_stats *ls_time_stats;
650 static inline struct lu_site_bkt_data *
651 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
655 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
656 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
660 * Constructors/destructors.
664 int lu_site_init (struct lu_site *s, struct lu_device *d);
665 void lu_site_fini (struct lu_site *s);
666 int lu_site_init_finish (struct lu_site *s);
667 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
668 void lu_device_get (struct lu_device *d);
669 void lu_device_put (struct lu_device *d);
670 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
671 void lu_device_fini (struct lu_device *d);
672 int lu_object_header_init(struct lu_object_header *h);
673 void lu_object_header_fini(struct lu_object_header *h);
674 int lu_object_init (struct lu_object *o,
675 struct lu_object_header *h, struct lu_device *d);
676 void lu_object_fini (struct lu_object *o);
677 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
678 void lu_object_add (struct lu_object *before, struct lu_object *o);
681 * Helpers to initialize and finalize device types.
684 int lu_device_type_init(struct lu_device_type *ldt);
685 void lu_device_type_fini(struct lu_device_type *ldt);
686 void lu_types_stop(void);
691 * Caching and reference counting.
696 * Acquire additional reference to the given object. This function is used to
697 * attain additional reference. To acquire initial reference use
700 static inline void lu_object_get(struct lu_object *o)
702 LASSERT(cfs_atomic_read(&o->lo_header->loh_ref) > 0);
703 cfs_atomic_inc(&o->lo_header->loh_ref);
707 * Return true of object will not be cached after last reference to it is
710 static inline int lu_object_is_dying(const struct lu_object_header *h)
712 return cfs_test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
715 void lu_object_put(const struct lu_env *env, struct lu_object *o);
717 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
719 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
720 lu_printer_t printer);
721 struct lu_object *lu_object_find(const struct lu_env *env,
722 struct lu_device *dev, const struct lu_fid *f,
723 const struct lu_object_conf *conf);
724 struct lu_object *lu_object_find_at(const struct lu_env *env,
725 struct lu_device *dev,
726 const struct lu_fid *f,
727 const struct lu_object_conf *conf);
728 struct lu_object *lu_object_find_slice(const struct lu_env *env,
729 struct lu_device *dev,
730 const struct lu_fid *f,
731 const struct lu_object_conf *conf);
740 * First (topmost) sub-object of given compound object
742 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
744 LASSERT(!cfs_list_empty(&h->loh_layers));
745 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
749 * Next sub-object in the layering
751 static inline struct lu_object *lu_object_next(const struct lu_object *o)
753 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
757 * Pointer to the fid of this object.
759 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
761 return &o->lo_header->loh_fid;
765 * return device operations vector for this object
767 static const inline struct lu_device_operations *
768 lu_object_ops(const struct lu_object *o)
770 return o->lo_dev->ld_ops;
774 * Given a compound object, find its slice, corresponding to the device type
777 struct lu_object *lu_object_locate(struct lu_object_header *h,
778 const struct lu_device_type *dtype);
781 * Printer function emitting messages through libcfs_debug_msg().
783 int lu_cdebug_printer(const struct lu_env *env,
784 void *cookie, const char *format, ...);
787 * Print object description followed by a user-supplied message.
789 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
791 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
793 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
794 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
795 CDEBUG(mask, format , ## __VA_ARGS__); \
800 * Print short object description followed by a user-supplied message.
802 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
804 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
806 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
807 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
808 (object)->lo_header); \
809 lu_cdebug_printer(env, &msgdata, "\n"); \
810 CDEBUG(mask, format , ## __VA_ARGS__); \
814 void lu_object_print (const struct lu_env *env, void *cookie,
815 lu_printer_t printer, const struct lu_object *o);
816 void lu_object_header_print(const struct lu_env *env, void *cookie,
817 lu_printer_t printer,
818 const struct lu_object_header *hdr);
821 * Check object consistency.
823 int lu_object_invariant(const struct lu_object *o);
827 * \retval 1 iff object \a o exists on stable storage,
828 * \retval 0 iff object \a o not exists on stable storage.
829 * \retval -1 iff object \a o is on remote server.
831 static inline int lu_object_exists(const struct lu_object *o)
835 attr = o->lo_header->loh_attr;
836 if (attr & LOHA_REMOTE)
838 else if (attr & LOHA_EXISTS)
844 static inline int lu_object_assert_exists(const struct lu_object *o)
846 return lu_object_exists(o) != 0;
849 static inline int lu_object_assert_not_exists(const struct lu_object *o)
851 return lu_object_exists(o) <= 0;
855 * Attr of this object.
857 static inline __u32 lu_object_attr(const struct lu_object *o)
859 LASSERT(lu_object_exists(o) > 0);
860 return o->lo_header->loh_attr;
863 static inline struct lu_ref_link *lu_object_ref_add(struct lu_object *o,
867 return lu_ref_add(&o->lo_header->loh_reference, scope, source);
870 static inline void lu_object_ref_del(struct lu_object *o,
871 const char *scope, const void *source)
873 lu_ref_del(&o->lo_header->loh_reference, scope, source);
876 static inline void lu_object_ref_del_at(struct lu_object *o,
877 struct lu_ref_link *link,
878 const char *scope, const void *source)
880 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
883 /** input params, should be filled out by mdt */
887 /** count in bytes */
888 unsigned int rp_count;
889 /** number of pages */
890 unsigned int rp_npages;
891 /** requested attr */
893 /** pointers to pages */
894 struct page **rp_pages;
897 enum lu_xattr_flags {
898 LU_XATTR_REPLACE = (1 << 0),
899 LU_XATTR_CREATE = (1 << 1)
907 /** For lu_context health-checks */
908 enum lu_context_state {
916 * lu_context. Execution context for lu_object methods. Currently associated
919 * All lu_object methods, except device and device type methods (called during
920 * system initialization and shutdown) are executed "within" some
921 * lu_context. This means, that pointer to some "current" lu_context is passed
922 * as an argument to all methods.
924 * All service ptlrpc threads create lu_context as part of their
925 * initialization. It is possible to create "stand-alone" context for other
926 * execution environments (like system calls).
928 * lu_object methods mainly use lu_context through lu_context_key interface
929 * that allows each layer to associate arbitrary pieces of data with each
930 * context (see pthread_key_create(3) for similar interface).
932 * On a client, lu_context is bound to a thread, see cl_env_get().
934 * \see lu_context_key
938 * lu_context is used on the client side too. Yet we don't want to
939 * allocate values of server-side keys for the client contexts and
942 * To achieve this, set of tags in introduced. Contexts and keys are
943 * marked with tags. Key value are created only for context whose set
944 * of tags has non-empty intersection with one for key. Tags are taken
945 * from enum lu_context_tag.
949 * Pointer to the home service thread. NULL for other execution
952 struct ptlrpc_thread *lc_thread;
954 * Pointer to an array with key values. Internal implementation
958 enum lu_context_state lc_state;
960 * Linkage into a list of all remembered contexts. Only
961 * `non-transient' contexts, i.e., ones created for service threads
964 cfs_list_t lc_remember;
966 * Version counter used to skip calls to lu_context_refill() when no
967 * keys were registered.
977 * lu_context_key interface. Similar to pthread_key.
980 enum lu_context_tag {
982 * Thread on md server
984 LCT_MD_THREAD = 1 << 0,
986 * Thread on dt server
988 LCT_DT_THREAD = 1 << 1,
990 * Context for transaction handle
992 LCT_TX_HANDLE = 1 << 2,
996 LCT_CL_THREAD = 1 << 3,
998 * A per-request session on a server, and a per-system-call session on
1001 LCT_SESSION = 1 << 4,
1004 * Set when at least one of keys, having values in this context has
1005 * non-NULL lu_context_key::lct_exit() method. This is used to
1006 * optimize lu_context_exit() call.
1008 LCT_HAS_EXIT = 1 << 28,
1010 * Don't add references for modules creating key values in that context.
1011 * This is only for contexts used internally by lu_object framework.
1013 LCT_NOREF = 1 << 29,
1015 * Key is being prepared for retiring, don't create new values for it.
1017 LCT_QUIESCENT = 1 << 30,
1019 * Context should be remembered.
1021 LCT_REMEMBER = 1 << 31,
1023 * Contexts usable in cache shrinker thread.
1025 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1029 * Key. Represents per-context value slot.
1031 * Keys are usually registered when module owning the key is initialized, and
1032 * de-registered when module is unloaded. Once key is registered, all new
1033 * contexts with matching tags, will get key value. "Old" contexts, already
1034 * initialized at the time of key registration, can be forced to get key value
1035 * by calling lu_context_refill().
1037 * Every key value is counted in lu_context_key::lct_used and acquires a
1038 * reference on an owning module. This means, that all key values have to be
1039 * destroyed before module can be unloaded. This is usually achieved by
1040 * stopping threads started by the module, that created contexts in their
1041 * entry functions. Situation is complicated by the threads shared by multiple
1042 * modules, like ptlrpcd daemon on a client. To work around this problem,
1043 * contexts, created in such threads, are `remembered' (see
1044 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1045 * for unloading it does the following:
1047 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1048 * preventing new key values from being allocated in the new contexts,
1051 * - scans a list of remembered contexts, destroying values of module
1052 * keys, thus releasing references to the module.
1054 * This is done by lu_context_key_quiesce(). If module is re-activated
1055 * before key has been de-registered, lu_context_key_revive() call clears
1056 * `quiescent' marker.
1058 * lu_context code doesn't provide any internal synchronization for these
1059 * activities---it's assumed that startup (including threads start-up) and
1060 * shutdown are serialized by some external means.
1064 struct lu_context_key {
1066 * Set of tags for which values of this key are to be instantiated.
1070 * Value constructor. This is called when new value is created for a
1071 * context. Returns pointer to new value of error pointer.
1073 void *(*lct_init)(const struct lu_context *ctx,
1074 struct lu_context_key *key);
1076 * Value destructor. Called when context with previously allocated
1077 * value of this slot is destroyed. \a data is a value that was returned
1078 * by a matching call to lu_context_key::lct_init().
1080 void (*lct_fini)(const struct lu_context *ctx,
1081 struct lu_context_key *key, void *data);
1083 * Optional method called on lu_context_exit() for all allocated
1084 * keys. Can be used by debugging code checking that locks are
1087 void (*lct_exit)(const struct lu_context *ctx,
1088 struct lu_context_key *key, void *data);
1090 * Internal implementation detail: index within lu_context::lc_value[]
1091 * reserved for this key.
1095 * Internal implementation detail: number of values created for this
1098 cfs_atomic_t lct_used;
1100 * Internal implementation detail: module for this key.
1102 cfs_module_t *lct_owner;
1104 * References to this key. For debugging.
1106 struct lu_ref lct_reference;
1109 #define LU_KEY_INIT(mod, type) \
1110 static void* mod##_key_init(const struct lu_context *ctx, \
1111 struct lu_context_key *key) \
1115 CLASSERT(CFS_PAGE_SIZE >= sizeof (*value)); \
1117 OBD_ALLOC_PTR(value); \
1118 if (value == NULL) \
1119 value = ERR_PTR(-ENOMEM); \
1123 struct __##mod##__dummy_init {;} /* semicolon catcher */
1125 #define LU_KEY_FINI(mod, type) \
1126 static void mod##_key_fini(const struct lu_context *ctx, \
1127 struct lu_context_key *key, void* data) \
1129 type *info = data; \
1131 OBD_FREE_PTR(info); \
1133 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1135 #define LU_KEY_INIT_FINI(mod, type) \
1136 LU_KEY_INIT(mod,type); \
1137 LU_KEY_FINI(mod,type)
1139 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1140 struct lu_context_key mod##_thread_key = { \
1142 .lct_init = mod##_key_init, \
1143 .lct_fini = mod##_key_fini \
1146 #define LU_CONTEXT_KEY_INIT(key) \
1148 (key)->lct_owner = THIS_MODULE; \
1151 int lu_context_key_register(struct lu_context_key *key);
1152 void lu_context_key_degister(struct lu_context_key *key);
1153 void *lu_context_key_get (const struct lu_context *ctx,
1154 const struct lu_context_key *key);
1155 void lu_context_key_quiesce (struct lu_context_key *key);
1156 void lu_context_key_revive (struct lu_context_key *key);
1160 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1164 #define LU_KEY_INIT_GENERIC(mod) \
1165 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1167 struct lu_context_key *key = k; \
1170 va_start(args, k); \
1172 LU_CONTEXT_KEY_INIT(key); \
1173 key = va_arg(args, struct lu_context_key *); \
1174 } while (key != NULL); \
1178 #define LU_TYPE_INIT(mod, ...) \
1179 LU_KEY_INIT_GENERIC(mod) \
1180 static int mod##_type_init(struct lu_device_type *t) \
1182 mod##_key_init_generic(__VA_ARGS__, NULL); \
1183 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1185 struct __##mod##_dummy_type_init {;}
1187 #define LU_TYPE_FINI(mod, ...) \
1188 static void mod##_type_fini(struct lu_device_type *t) \
1190 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1192 struct __##mod##_dummy_type_fini {;}
1194 #define LU_TYPE_START(mod, ...) \
1195 static void mod##_type_start(struct lu_device_type *t) \
1197 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1199 struct __##mod##_dummy_type_start {;}
1201 #define LU_TYPE_STOP(mod, ...) \
1202 static void mod##_type_stop(struct lu_device_type *t) \
1204 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1206 struct __##mod##_dummy_type_stop {;}
1210 #define LU_TYPE_INIT_FINI(mod, ...) \
1211 LU_TYPE_INIT(mod, __VA_ARGS__); \
1212 LU_TYPE_FINI(mod, __VA_ARGS__); \
1213 LU_TYPE_START(mod, __VA_ARGS__); \
1214 LU_TYPE_STOP(mod, __VA_ARGS__)
1216 int lu_context_init (struct lu_context *ctx, __u32 tags);
1217 void lu_context_fini (struct lu_context *ctx);
1218 void lu_context_enter (struct lu_context *ctx);
1219 void lu_context_exit (struct lu_context *ctx);
1220 int lu_context_refill(struct lu_context *ctx);
1223 * Helper functions to operate on multiple keys. These are used by the default
1224 * device type operations, defined by LU_TYPE_INIT_FINI().
1227 int lu_context_key_register_many(struct lu_context_key *k, ...);
1228 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1229 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1230 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1233 * update/clear ctx/ses tags.
1235 void lu_context_tags_update(__u32 tags);
1236 void lu_context_tags_clear(__u32 tags);
1237 void lu_session_tags_update(__u32 tags);
1238 void lu_session_tags_clear(__u32 tags);
1245 * "Local" context, used to store data instead of stack.
1247 struct lu_context le_ctx;
1249 * "Session" context for per-request data.
1251 struct lu_context *le_ses;
1254 int lu_env_init (struct lu_env *env, __u32 tags);
1255 void lu_env_fini (struct lu_env *env);
1256 int lu_env_refill(struct lu_env *env);
1257 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1259 /** @} lu_context */
1262 * Output site statistical counters into a buffer. Suitable for
1263 * ll_rd_*()-style functions.
1265 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1268 * Common name structure to be passed around for various name related methods.
1271 const char *ln_name;
1276 * Common buffer structure to be passed around for various xattr_{s,g}et()
1285 extern struct lu_buf LU_BUF_NULL;
1287 #define DLUBUF "(%p %z)"
1288 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1290 * One-time initializers, called at obdclass module initialization, not
1295 * Initialization of global lu_* data.
1297 int lu_global_init(void);
1300 * Dual to lu_global_init().
1302 void lu_global_fini(void);
1305 LU_TIME_FIND_LOOKUP,
1307 LU_TIME_FIND_INSERT,
1311 extern const char *lu_time_names[LU_TIME_NR];
1313 struct lu_kmem_descr {
1314 cfs_mem_cache_t **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);
1323 #endif /* __LUSTRE_LU_OBJECT_H */