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 struct list_head 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 atomic_t ldt_device_nr;
338 * Linkage into a global list of all device types.
340 * \see lu_device_types.
342 struct list_head 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 * Common object attributes.
406 /** modification time in seconds since Epoch */
408 /** access time in seconds since Epoch */
410 /** change time in seconds since Epoch */
412 /** 512-byte blocks allocated to object */
414 /** permission bits and file type */
422 /** number of persistent references to this object */
424 /** blk bits of the object*/
426 /** blk size of the object*/
438 /** Bit-mask of valid attributes */
452 LA_BLKSIZE = 1 << 12,
453 LA_KILL_SUID = 1 << 13,
454 LA_KILL_SGID = 1 << 14,
458 * Layer in the layered object.
462 * Header for this object.
464 struct lu_object_header *lo_header;
466 * Device for this layer.
468 struct lu_device *lo_dev;
470 * Operations for this object.
472 const struct lu_object_operations *lo_ops;
474 * Linkage into list of all layers.
476 struct list_head lo_linkage;
478 * Link to the device, for debugging.
480 struct lu_ref_link lo_dev_ref;
483 enum lu_object_header_flags {
485 * Don't keep this object in cache. Object will be destroyed as soon
486 * as last reference to it is released. This flag cannot be cleared
489 LU_OBJECT_HEARD_BANSHEE = 0,
491 * Mark this object has already been taken out of cache.
493 LU_OBJECT_UNHASHED = 1,
496 enum lu_object_header_attr {
497 LOHA_EXISTS = 1 << 0,
498 LOHA_REMOTE = 1 << 1,
500 * UNIX file type is stored in S_IFMT bits.
502 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
503 LOHA_FT_END = 017 << 12, /**< S_IFMT */
507 * "Compound" object, consisting of multiple layers.
509 * Compound object with given fid is unique with given lu_site.
511 * Note, that object does *not* necessary correspond to the real object in the
512 * persistent storage: object is an anchor for locking and method calling, so
513 * it is created for things like not-yet-existing child created by mkdir or
514 * create calls. lu_object_operations::loo_exists() can be used to check
515 * whether object is backed by persistent storage entity.
517 struct lu_object_header {
519 * Fid, uniquely identifying this object.
521 struct lu_fid loh_fid;
523 * Object flags from enum lu_object_header_flags. Set and checked
526 unsigned long loh_flags;
528 * Object reference count. Protected by lu_site::ls_guard.
532 * Common object attributes, cached for efficiency. From enum
533 * lu_object_header_attr.
537 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
539 struct hlist_node loh_hash;
541 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
543 struct list_head loh_lru;
545 * Linkage into list of layers. Never modified once set (except lately
546 * during object destruction). No locking is necessary.
548 struct list_head loh_layers;
550 * A list of references to this object, for debugging.
552 struct lu_ref loh_reference;
557 struct lu_site_bkt_data {
559 * number of busy object on this bucket
563 * LRU list, updated on each access to object. Protected by
564 * bucket lock of lu_site::ls_obj_hash.
566 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
567 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
568 * of list_for_each_entry_safe_reverse()).
570 struct list_head lsb_lru;
572 * Wait-queue signaled when an object in this site is ultimately
573 * destroyed (lu_object_free()). It is used by lu_object_find() to
574 * wait before re-trying when object in the process of destruction is
575 * found in the hash table.
577 * \see htable_lookup().
579 wait_queue_head_t lsb_marche_funebre;
587 LU_SS_CACHE_DEATH_RACE,
593 * lu_site is a "compartment" within which objects are unique, and LRU
594 * discipline is maintained.
596 * lu_site exists so that multiple layered stacks can co-exist in the same
599 * lu_site has the same relation to lu_device as lu_object_header to
606 cfs_hash_t *ls_obj_hash;
608 * index of bucket on hash table while purging
612 * Top-level device for this stack.
614 struct lu_device *ls_top_dev;
616 * Bottom-level device for this stack
618 struct lu_device *ls_bottom_dev;
620 * Linkage into global list of sites.
622 struct list_head ls_linkage;
624 * List for lu device for this site, protected
627 struct list_head ls_ld_linkage;
628 spinlock_t ls_ld_lock;
630 * Lock to serialize site purge.
632 struct mutex ls_purge_mutex;
636 struct lprocfs_stats *ls_stats;
638 * XXX: a hack! fld has to find md_site via site, remove when possible
640 struct seq_server_site *ld_seq_site;
643 static inline struct lu_site_bkt_data *
644 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
648 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
649 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
652 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
654 return s->ld_seq_site;
658 * Constructors/destructors.
662 int lu_site_init (struct lu_site *s, struct lu_device *d);
663 void lu_site_fini (struct lu_site *s);
664 int lu_site_init_finish (struct lu_site *s);
665 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
666 void lu_device_get (struct lu_device *d);
667 void lu_device_put (struct lu_device *d);
668 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
669 void lu_device_fini (struct lu_device *d);
670 int lu_object_header_init(struct lu_object_header *h);
671 void lu_object_header_fini(struct lu_object_header *h);
672 int lu_object_init (struct lu_object *o,
673 struct lu_object_header *h, struct lu_device *d);
674 void lu_object_fini (struct lu_object *o);
675 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
676 void lu_object_add (struct lu_object *before, struct lu_object *o);
678 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
679 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
682 * Helpers to initialize and finalize device types.
685 int lu_device_type_init(struct lu_device_type *ldt);
686 void lu_device_type_fini(struct lu_device_type *ldt);
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(atomic_read(&o->lo_header->loh_ref) > 0);
703 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 test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
715 void lu_object_put(const struct lu_env *env, struct lu_object *o);
716 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
717 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
719 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
721 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
722 lu_printer_t printer);
723 struct lu_object *lu_object_find(const struct lu_env *env,
724 struct lu_device *dev, const struct lu_fid *f,
725 const struct lu_object_conf *conf);
726 struct lu_object *lu_object_find_at(const struct lu_env *env,
727 struct lu_device *dev,
728 const struct lu_fid *f,
729 const struct lu_object_conf *conf);
730 void lu_object_purge(const struct lu_env *env, struct lu_device *dev,
731 const struct lu_fid *f);
732 struct lu_object *lu_object_find_slice(const struct lu_env *env,
733 struct lu_device *dev,
734 const struct lu_fid *f,
735 const struct lu_object_conf *conf);
744 * First (topmost) sub-object of given compound object
746 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
748 LASSERT(!list_empty(&h->loh_layers));
749 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
753 * Next sub-object in the layering
755 static inline struct lu_object *lu_object_next(const struct lu_object *o)
757 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
761 * Pointer to the fid of this object.
763 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
765 return &o->lo_header->loh_fid;
769 * return device operations vector for this object
771 static const inline struct lu_device_operations *
772 lu_object_ops(const struct lu_object *o)
774 return o->lo_dev->ld_ops;
778 * Given a compound object, find its slice, corresponding to the device type
781 struct lu_object *lu_object_locate(struct lu_object_header *h,
782 const struct lu_device_type *dtype);
785 * Printer function emitting messages through libcfs_debug_msg().
787 int lu_cdebug_printer(const struct lu_env *env,
788 void *cookie, const char *format, ...);
791 * Print object description followed by a user-supplied message.
793 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
795 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
796 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
797 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
798 CDEBUG(mask, format , ## __VA_ARGS__); \
803 * Print short object description followed by a user-supplied message.
805 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
807 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
808 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
809 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
810 (object)->lo_header); \
811 lu_cdebug_printer(env, &msgdata, "\n"); \
812 CDEBUG(mask, format , ## __VA_ARGS__); \
816 void lu_object_print (const struct lu_env *env, void *cookie,
817 lu_printer_t printer, const struct lu_object *o);
818 void lu_object_header_print(const struct lu_env *env, void *cookie,
819 lu_printer_t printer,
820 const struct lu_object_header *hdr);
823 * Check object consistency.
825 int lu_object_invariant(const struct lu_object *o);
829 * Check whether object exists, no matter on local or remote storage.
830 * Note: LOHA_EXISTS will be set once some one created the object,
831 * and it does not needs to be committed to storage.
833 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
836 * Check whether object on the remote storage.
838 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
840 static inline int lu_object_assert_exists(const struct lu_object *o)
842 return lu_object_exists(o);
845 static inline int lu_object_assert_not_exists(const struct lu_object *o)
847 return !lu_object_exists(o);
851 * Attr of this object.
853 static inline __u32 lu_object_attr(const struct lu_object *o)
855 LASSERT(lu_object_exists(o) != 0);
856 return o->lo_header->loh_attr;
859 static inline void lu_object_ref_add(struct lu_object *o,
863 lu_ref_add(&o->lo_header->loh_reference, scope, source);
866 static inline void lu_object_ref_add_at(struct lu_object *o,
867 struct lu_ref_link *link,
871 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
874 static inline void lu_object_ref_del(struct lu_object *o,
875 const char *scope, const void *source)
877 lu_ref_del(&o->lo_header->loh_reference, scope, source);
880 static inline void lu_object_ref_del_at(struct lu_object *o,
881 struct lu_ref_link *link,
882 const char *scope, const void *source)
884 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
887 /** input params, should be filled out by mdt */
891 /** count in bytes */
892 unsigned int rp_count;
893 /** number of pages */
894 unsigned int rp_npages;
895 /** requested attr */
897 /** pointers to pages */
898 struct page **rp_pages;
901 enum lu_xattr_flags {
902 LU_XATTR_REPLACE = (1 << 0),
903 LU_XATTR_CREATE = (1 << 1)
911 /** For lu_context health-checks */
912 enum lu_context_state {
920 * lu_context. Execution context for lu_object methods. Currently associated
923 * All lu_object methods, except device and device type methods (called during
924 * system initialization and shutdown) are executed "within" some
925 * lu_context. This means, that pointer to some "current" lu_context is passed
926 * as an argument to all methods.
928 * All service ptlrpc threads create lu_context as part of their
929 * initialization. It is possible to create "stand-alone" context for other
930 * execution environments (like system calls).
932 * lu_object methods mainly use lu_context through lu_context_key interface
933 * that allows each layer to associate arbitrary pieces of data with each
934 * context (see pthread_key_create(3) for similar interface).
936 * On a client, lu_context is bound to a thread, see cl_env_get().
938 * \see lu_context_key
942 * lu_context is used on the client side too. Yet we don't want to
943 * allocate values of server-side keys for the client contexts and
946 * To achieve this, set of tags in introduced. Contexts and keys are
947 * marked with tags. Key value are created only for context whose set
948 * of tags has non-empty intersection with one for key. Tags are taken
949 * from enum lu_context_tag.
952 enum lu_context_state lc_state;
954 * Pointer to the home service thread. NULL for other execution
957 struct ptlrpc_thread *lc_thread;
959 * Pointer to an array with key values. Internal implementation
964 * Linkage into a list of all remembered contexts. Only
965 * `non-transient' contexts, i.e., ones created for service threads
968 struct list_head lc_remember;
970 * Version counter used to skip calls to lu_context_refill() when no
971 * keys were registered.
981 * lu_context_key interface. Similar to pthread_key.
984 enum lu_context_tag {
986 * Thread on md server
988 LCT_MD_THREAD = 1 << 0,
990 * Thread on dt server
992 LCT_DT_THREAD = 1 << 1,
994 * Context for transaction handle
996 LCT_TX_HANDLE = 1 << 2,
1000 LCT_CL_THREAD = 1 << 3,
1002 * A per-request session on a server, and a per-system-call session on
1005 LCT_SESSION = 1 << 4,
1007 * A per-request data on OSP device
1009 LCT_OSP_THREAD = 1 << 5,
1013 LCT_MG_THREAD = 1 << 6,
1015 * Context for local operations
1019 * session for server thread
1021 LCT_SERVER_SESSION = 1 << 8,
1023 * Set when at least one of keys, having values in this context has
1024 * non-NULL lu_context_key::lct_exit() method. This is used to
1025 * optimize lu_context_exit() call.
1027 LCT_HAS_EXIT = 1 << 28,
1029 * Don't add references for modules creating key values in that context.
1030 * This is only for contexts used internally by lu_object framework.
1032 LCT_NOREF = 1 << 29,
1034 * Key is being prepared for retiring, don't create new values for it.
1036 LCT_QUIESCENT = 1 << 30,
1038 * Context should be remembered.
1040 LCT_REMEMBER = 1 << 31,
1042 * Contexts usable in cache shrinker thread.
1044 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1048 * Key. Represents per-context value slot.
1050 * Keys are usually registered when module owning the key is initialized, and
1051 * de-registered when module is unloaded. Once key is registered, all new
1052 * contexts with matching tags, will get key value. "Old" contexts, already
1053 * initialized at the time of key registration, can be forced to get key value
1054 * by calling lu_context_refill().
1056 * Every key value is counted in lu_context_key::lct_used and acquires a
1057 * reference on an owning module. This means, that all key values have to be
1058 * destroyed before module can be unloaded. This is usually achieved by
1059 * stopping threads started by the module, that created contexts in their
1060 * entry functions. Situation is complicated by the threads shared by multiple
1061 * modules, like ptlrpcd daemon on a client. To work around this problem,
1062 * contexts, created in such threads, are `remembered' (see
1063 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1064 * for unloading it does the following:
1066 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1067 * preventing new key values from being allocated in the new contexts,
1070 * - scans a list of remembered contexts, destroying values of module
1071 * keys, thus releasing references to the module.
1073 * This is done by lu_context_key_quiesce(). If module is re-activated
1074 * before key has been de-registered, lu_context_key_revive() call clears
1075 * `quiescent' marker.
1077 * lu_context code doesn't provide any internal synchronization for these
1078 * activities---it's assumed that startup (including threads start-up) and
1079 * shutdown are serialized by some external means.
1083 struct lu_context_key {
1085 * Set of tags for which values of this key are to be instantiated.
1089 * Value constructor. This is called when new value is created for a
1090 * context. Returns pointer to new value of error pointer.
1092 void *(*lct_init)(const struct lu_context *ctx,
1093 struct lu_context_key *key);
1095 * Value destructor. Called when context with previously allocated
1096 * value of this slot is destroyed. \a data is a value that was returned
1097 * by a matching call to lu_context_key::lct_init().
1099 void (*lct_fini)(const struct lu_context *ctx,
1100 struct lu_context_key *key, void *data);
1102 * Optional method called on lu_context_exit() for all allocated
1103 * keys. Can be used by debugging code checking that locks are
1106 void (*lct_exit)(const struct lu_context *ctx,
1107 struct lu_context_key *key, void *data);
1109 * Internal implementation detail: index within lu_context::lc_value[]
1110 * reserved for this key.
1114 * Internal implementation detail: number of values created for this
1119 * Internal implementation detail: module for this key.
1121 struct module *lct_owner;
1123 * References to this key. For debugging.
1125 struct lu_ref lct_reference;
1128 #define LU_KEY_INIT(mod, type) \
1129 static void* mod##_key_init(const struct lu_context *ctx, \
1130 struct lu_context_key *key) \
1134 CLASSERT(PAGE_CACHE_SIZE >= sizeof (*value)); \
1136 OBD_ALLOC_PTR(value); \
1137 if (value == NULL) \
1138 value = ERR_PTR(-ENOMEM); \
1142 struct __##mod##__dummy_init {;} /* semicolon catcher */
1144 #define LU_KEY_FINI(mod, type) \
1145 static void mod##_key_fini(const struct lu_context *ctx, \
1146 struct lu_context_key *key, void* data) \
1148 type *info = data; \
1150 OBD_FREE_PTR(info); \
1152 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1154 #define LU_KEY_INIT_FINI(mod, type) \
1155 LU_KEY_INIT(mod,type); \
1156 LU_KEY_FINI(mod,type)
1158 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1159 struct lu_context_key mod##_thread_key = { \
1161 .lct_init = mod##_key_init, \
1162 .lct_fini = mod##_key_fini \
1165 #define LU_CONTEXT_KEY_INIT(key) \
1167 (key)->lct_owner = THIS_MODULE; \
1170 int lu_context_key_register(struct lu_context_key *key);
1171 void lu_context_key_degister(struct lu_context_key *key);
1172 void *lu_context_key_get (const struct lu_context *ctx,
1173 const struct lu_context_key *key);
1174 void lu_context_key_quiesce (struct lu_context_key *key);
1175 void lu_context_key_revive (struct lu_context_key *key);
1179 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1183 #define LU_KEY_INIT_GENERIC(mod) \
1184 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1186 struct lu_context_key *key = k; \
1189 va_start(args, k); \
1191 LU_CONTEXT_KEY_INIT(key); \
1192 key = va_arg(args, struct lu_context_key *); \
1193 } while (key != NULL); \
1197 #define LU_TYPE_INIT(mod, ...) \
1198 LU_KEY_INIT_GENERIC(mod) \
1199 static int mod##_type_init(struct lu_device_type *t) \
1201 mod##_key_init_generic(__VA_ARGS__, NULL); \
1202 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1204 struct __##mod##_dummy_type_init {;}
1206 #define LU_TYPE_FINI(mod, ...) \
1207 static void mod##_type_fini(struct lu_device_type *t) \
1209 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1211 struct __##mod##_dummy_type_fini {;}
1213 #define LU_TYPE_START(mod, ...) \
1214 static void mod##_type_start(struct lu_device_type *t) \
1216 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1218 struct __##mod##_dummy_type_start {;}
1220 #define LU_TYPE_STOP(mod, ...) \
1221 static void mod##_type_stop(struct lu_device_type *t) \
1223 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1225 struct __##mod##_dummy_type_stop {;}
1229 #define LU_TYPE_INIT_FINI(mod, ...) \
1230 LU_TYPE_INIT(mod, __VA_ARGS__); \
1231 LU_TYPE_FINI(mod, __VA_ARGS__); \
1232 LU_TYPE_START(mod, __VA_ARGS__); \
1233 LU_TYPE_STOP(mod, __VA_ARGS__)
1235 int lu_context_init (struct lu_context *ctx, __u32 tags);
1236 void lu_context_fini (struct lu_context *ctx);
1237 void lu_context_enter (struct lu_context *ctx);
1238 void lu_context_exit (struct lu_context *ctx);
1239 int lu_context_refill(struct lu_context *ctx);
1242 * Helper functions to operate on multiple keys. These are used by the default
1243 * device type operations, defined by LU_TYPE_INIT_FINI().
1246 int lu_context_key_register_many(struct lu_context_key *k, ...);
1247 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1248 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1249 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1252 * update/clear ctx/ses tags.
1254 void lu_context_tags_update(__u32 tags);
1255 void lu_context_tags_clear(__u32 tags);
1256 void lu_session_tags_update(__u32 tags);
1257 void lu_session_tags_clear(__u32 tags);
1264 * "Local" context, used to store data instead of stack.
1266 struct lu_context le_ctx;
1268 * "Session" context for per-request data.
1270 struct lu_context *le_ses;
1273 int lu_env_init (struct lu_env *env, __u32 tags);
1274 void lu_env_fini (struct lu_env *env);
1275 int lu_env_refill(struct lu_env *env);
1276 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1278 /** @} lu_context */
1281 * Output site statistical counters into a buffer. Suitable for
1282 * ll_rd_*()-style functions.
1284 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1285 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1288 * Common name structure to be passed around for various name related methods.
1291 const char *ln_name;
1296 * Validate names (path components)
1298 * To be valid \a name must be non-empty, '\0' terminated of length \a
1299 * name_len, and not contain '/'. The maximum length of a name (before
1300 * say -ENAMETOOLONG will be returned) is really controlled by llite
1301 * and the server. We only check for something insane coming from bad
1302 * integer handling here.
1304 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1306 return name != NULL &&
1308 name_len < INT_MAX &&
1309 name[name_len] == '\0' &&
1310 strlen(name) == name_len &&
1311 memchr(name, '/', name_len) == NULL;
1314 static inline bool lu_name_is_valid(const struct lu_name *ln)
1316 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1319 #define DNAME "%.*s"
1321 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1322 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1325 * Common buffer structure to be passed around for various xattr_{s,g}et()
1333 #define DLUBUF "(%p %zu)"
1334 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1336 * One-time initializers, called at obdclass module initialization, not
1341 * Initialization of global lu_* data.
1343 int lu_global_init(void);
1346 * Dual to lu_global_init().
1348 void lu_global_fini(void);
1350 struct lu_kmem_descr {
1351 struct kmem_cache **ckd_cache;
1352 const char *ckd_name;
1353 const size_t ckd_size;
1356 int lu_kmem_init(struct lu_kmem_descr *caches);
1357 void lu_kmem_fini(struct lu_kmem_descr *caches);
1359 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1360 const struct lu_fid *fid);
1361 struct lu_object *lu_object_anon(const struct lu_env *env,
1362 struct lu_device *dev,
1363 const struct lu_object_conf *conf);
1366 extern struct lu_buf LU_BUF_NULL;
1368 void lu_buf_free(struct lu_buf *buf);
1369 void lu_buf_alloc(struct lu_buf *buf, int size);
1370 void lu_buf_realloc(struct lu_buf *buf, int size);
1372 int lu_buf_check_and_grow(struct lu_buf *buf, int len);
1373 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len);
1376 #endif /* __LUSTRE_LU_OBJECT_H */