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, 2015, 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>
44 #include <linux/percpu_counter.h>
47 struct proc_dir_entry;
52 * lu_* data-types represent server-side entities shared by data and meta-data
57 * -# support for layering.
59 * Server side object is split into layers, one per device in the
60 * corresponding device stack. Individual layer is represented by struct
61 * lu_object. Compound layered object --- by struct lu_object_header. Most
62 * interface functions take lu_object as an argument and operate on the
63 * whole compound object. This decision was made due to the following
66 * - it's envisaged that lu_object will be used much more often than
69 * - we want lower (non-top) layers to be able to initiate operations
70 * on the whole object.
72 * Generic code supports layering more complex than simple stacking, e.g.,
73 * it is possible that at some layer object "spawns" multiple sub-objects
76 * -# fid-based identification.
78 * Compound object is uniquely identified by its fid. Objects are indexed
79 * by their fids (hash table is used for index).
81 * -# caching and life-cycle management.
83 * Object's life-time is controlled by reference counting. When reference
84 * count drops to 0, object is returned to cache. Cached objects still
85 * retain their identity (i.e., fid), and can be recovered from cache.
87 * Objects are kept in the global LRU list, and lu_site_purge() function
88 * can be used to reclaim given number of unused objects from the tail of
91 * -# avoiding recursion.
93 * Generic code tries to replace recursion through layers by iterations
94 * where possible. Additionally to the end of reducing stack consumption,
95 * data, when practically possible, are allocated through lu_context_key
96 * interface rather than on stack.
103 struct lu_object_header;
108 * Operations common for data and meta-data devices.
110 struct lu_device_operations {
112 * Allocate object for the given device (without lower-layer
113 * parts). This is called by lu_object_operations::loo_object_init()
114 * from the parent layer, and should setup at least lu_object::lo_dev
115 * and lu_object::lo_ops fields of resulting lu_object.
117 * Object creation protocol.
119 * Due to design goal of avoiding recursion, object creation (see
120 * lu_object_alloc()) is somewhat involved:
122 * - first, lu_device_operations::ldo_object_alloc() method of the
123 * top-level device in the stack is called. It should allocate top
124 * level object (including lu_object_header), but without any
125 * lower-layer sub-object(s).
127 * - then lu_object_alloc() sets fid in the header of newly created
130 * - then lu_object_operations::loo_object_init() is called. It has
131 * to allocate lower-layer object(s). To do this,
132 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
133 * of the lower-layer device(s).
135 * - for all new objects allocated by
136 * lu_object_operations::loo_object_init() (and inserted into object
137 * stack), lu_object_operations::loo_object_init() is called again
138 * repeatedly, until no new objects are created.
140 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
141 * result->lo_ops != NULL);
143 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
144 const struct lu_object_header *h,
145 struct lu_device *d);
147 * process config specific for device.
149 int (*ldo_process_config)(const struct lu_env *env,
150 struct lu_device *, struct lustre_cfg *);
151 int (*ldo_recovery_complete)(const struct lu_env *,
155 * initialize local objects for device. this method called after layer has
156 * been initialized (after LCFG_SETUP stage) and before it starts serving
160 int (*ldo_prepare)(const struct lu_env *,
161 struct lu_device *parent,
162 struct lu_device *dev);
167 * For lu_object_conf flags
170 /* This is a new object to be allocated, or the file
171 * corresponding to the object does not exists. */
172 LOC_F_NEW = 0x00000001,
174 /* When find a dying object, just return -EAGAIN at once instead of
175 * blocking the thread. */
176 LOC_F_NOWAIT = 0x00000002,
180 * Object configuration, describing particulars of object being created. On
181 * server this is not used, as server objects are full identified by fid. On
182 * client configuration contains struct lustre_md.
184 struct lu_object_conf {
186 * Some hints for obj find and alloc.
188 loc_flags_t loc_flags;
192 * Type of "printer" function used by lu_object_operations::loo_object_print()
195 * Printer function is needed to provide some flexibility in (semi-)debugging
196 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
198 typedef int (*lu_printer_t)(const struct lu_env *env,
199 void *cookie, const char *format, ...)
200 __attribute__ ((format (printf, 3, 4)));
203 * Operations specific for particular lu_object.
205 struct lu_object_operations {
208 * Allocate lower-layer parts of the object by calling
209 * lu_device_operations::ldo_object_alloc() of the corresponding
212 * This method is called once for each object inserted into object
213 * stack. It's responsibility of this method to insert lower-layer
214 * object(s) it create into appropriate places of object stack.
216 int (*loo_object_init)(const struct lu_env *env,
218 const struct lu_object_conf *conf);
220 * Called (in top-to-bottom order) during object allocation after all
221 * layers were allocated and initialized. Can be used to perform
222 * initialization depending on lower layers.
224 int (*loo_object_start)(const struct lu_env *env,
225 struct lu_object *o);
227 * Called before lu_object_operations::loo_object_free() to signal
228 * that object is being destroyed. Dual to
229 * lu_object_operations::loo_object_init().
231 void (*loo_object_delete)(const struct lu_env *env,
232 struct lu_object *o);
234 * Dual to lu_device_operations::ldo_object_alloc(). Called when
235 * object is removed from memory.
237 void (*loo_object_free)(const struct lu_env *env,
238 struct lu_object *o);
240 * Called when last active reference to the object is released (and
241 * object returns to the cache). This method is optional.
243 void (*loo_object_release)(const struct lu_env *env,
244 struct lu_object *o);
246 * Optional debugging helper. Print given object.
248 int (*loo_object_print)(const struct lu_env *env, void *cookie,
249 lu_printer_t p, const struct lu_object *o);
251 * Optional debugging method. Returns true iff method is internally
254 int (*loo_object_invariant)(const struct lu_object *o);
260 struct lu_device_type;
263 * Device: a layer in the server side abstraction stacking.
267 * reference count. This is incremented, in particular, on each object
268 * created at this layer.
270 * \todo XXX which means that atomic_t is probably too small.
274 * Pointer to device type. Never modified once set.
276 struct lu_device_type *ld_type;
278 * Operation vector for this device.
280 const struct lu_device_operations *ld_ops;
282 * Stack this device belongs to.
284 struct lu_site *ld_site;
285 struct proc_dir_entry *ld_proc_entry;
287 /** \todo XXX: temporary back pointer into obd. */
288 struct obd_device *ld_obd;
290 * A list of references to this object, for debugging.
292 struct lu_ref ld_reference;
294 * Link the device to the site.
296 struct list_head ld_linkage;
299 struct lu_device_type_operations;
302 * Tag bits for device type. They are used to distinguish certain groups of
306 /** this is meta-data device */
307 LU_DEVICE_MD = (1 << 0),
308 /** this is data device */
309 LU_DEVICE_DT = (1 << 1),
310 /** data device in the client stack */
311 LU_DEVICE_CL = (1 << 2)
317 struct lu_device_type {
319 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
323 * Name of this class. Unique system-wide. Never modified once set.
327 * Operations for this type.
329 const struct lu_device_type_operations *ldt_ops;
331 * \todo XXX: temporary pointer to associated obd_type.
333 struct obd_type *ldt_obd_type;
335 * \todo XXX: temporary: context tags used by obd_*() calls.
339 * Number of existing device type instances.
341 atomic_t ldt_device_nr;
343 * Linkage into a global list of all device types.
345 * \see lu_device_types.
347 struct list_head ldt_linkage;
351 * Operations on a device type.
353 struct lu_device_type_operations {
355 * Allocate new device.
357 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
358 struct lu_device_type *t,
359 struct lustre_cfg *lcfg);
361 * Free device. Dual to
362 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
363 * the next device in the stack.
365 struct lu_device *(*ldto_device_free)(const struct lu_env *,
369 * Initialize the devices after allocation
371 int (*ldto_device_init)(const struct lu_env *env,
372 struct lu_device *, const char *,
375 * Finalize device. Dual to
376 * lu_device_type_operations::ldto_device_init(). Returns pointer to
377 * the next device in the stack.
379 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
382 * Initialize device type. This is called on module load.
384 int (*ldto_init)(struct lu_device_type *t);
386 * Finalize device type. Dual to
387 * lu_device_type_operations::ldto_init(). Called on module unload.
389 void (*ldto_fini)(struct lu_device_type *t);
391 * Called when the first device is created.
393 void (*ldto_start)(struct lu_device_type *t);
395 * Called when number of devices drops to 0.
397 void (*ldto_stop)(struct lu_device_type *t);
400 static inline int lu_device_is_md(const struct lu_device *d)
402 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
406 * Common object attributes.
411 /** modification time in seconds since Epoch */
413 /** access time in seconds since Epoch */
415 /** change time in seconds since Epoch */
417 /** 512-byte blocks allocated to object */
419 /** permission bits and file type */
427 /** number of persistent references to this object */
429 /** blk bits of the object*/
431 /** blk size of the object*/
443 /** Bit-mask of valid attributes */
457 LA_BLKSIZE = 1 << 12,
458 LA_KILL_SUID = 1 << 13,
459 LA_KILL_SGID = 1 << 14,
463 * Layer in the layered object.
467 * Header for this object.
469 struct lu_object_header *lo_header;
471 * Device for this layer.
473 struct lu_device *lo_dev;
475 * Operations for this object.
477 const struct lu_object_operations *lo_ops;
479 * Linkage into list of all layers.
481 struct list_head lo_linkage;
483 * Link to the device, for debugging.
485 struct lu_ref_link lo_dev_ref;
488 enum lu_object_header_flags {
490 * Don't keep this object in cache. Object will be destroyed as soon
491 * as last reference to it is released. This flag cannot be cleared
494 LU_OBJECT_HEARD_BANSHEE = 0,
496 * Mark this object has already been taken out of cache.
498 LU_OBJECT_UNHASHED = 1,
501 enum lu_object_header_attr {
502 LOHA_EXISTS = 1 << 0,
503 LOHA_REMOTE = 1 << 1,
505 * UNIX file type is stored in S_IFMT bits.
507 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
508 LOHA_FT_END = 017 << 12, /**< S_IFMT */
512 * "Compound" object, consisting of multiple layers.
514 * Compound object with given fid is unique with given lu_site.
516 * Note, that object does *not* necessary correspond to the real object in the
517 * persistent storage: object is an anchor for locking and method calling, so
518 * it is created for things like not-yet-existing child created by mkdir or
519 * create calls. lu_object_operations::loo_exists() can be used to check
520 * whether object is backed by persistent storage entity.
522 struct lu_object_header {
524 * Fid, uniquely identifying this object.
526 struct lu_fid loh_fid;
528 * Object flags from enum lu_object_header_flags. Set and checked
531 unsigned long loh_flags;
533 * Object reference count. Protected by lu_site::ls_guard.
537 * Common object attributes, cached for efficiency. From enum
538 * lu_object_header_attr.
542 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
544 struct hlist_node loh_hash;
546 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
548 struct list_head loh_lru;
550 * Linkage into list of layers. Never modified once set (except lately
551 * during object destruction). No locking is necessary.
553 struct list_head loh_layers;
555 * A list of references to this object, for debugging.
557 struct lu_ref loh_reference;
562 struct lu_site_bkt_data {
564 * number of object in this bucket on the lsb_lru list.
568 * LRU list, updated on each access to object. Protected by
569 * bucket lock of lu_site::ls_obj_hash.
571 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
572 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
573 * of list_for_each_entry_safe_reverse()).
575 struct list_head lsb_lru;
577 * Wait-queue signaled when an object in this site is ultimately
578 * destroyed (lu_object_free()). It is used by lu_object_find() to
579 * wait before re-trying when object in the process of destruction is
580 * found in the hash table.
582 * \see htable_lookup().
584 wait_queue_head_t lsb_marche_funebre;
592 LU_SS_CACHE_DEATH_RACE,
598 * lu_site is a "compartment" within which objects are unique, and LRU
599 * discipline is maintained.
601 * lu_site exists so that multiple layered stacks can co-exist in the same
604 * lu_site has the same relation to lu_device as lu_object_header to
611 struct cfs_hash *ls_obj_hash;
613 * index of bucket on hash table while purging
615 unsigned int ls_purge_start;
617 * Top-level device for this stack.
619 struct lu_device *ls_top_dev;
621 * Bottom-level device for this stack
623 struct lu_device *ls_bottom_dev;
625 * Linkage into global list of sites.
627 struct list_head ls_linkage;
629 * List for lu device for this site, protected
632 struct list_head ls_ld_linkage;
633 spinlock_t ls_ld_lock;
635 * Lock to serialize site purge.
637 struct mutex ls_purge_mutex;
641 struct lprocfs_stats *ls_stats;
643 * XXX: a hack! fld has to find md_site via site, remove when possible
645 struct seq_server_site *ld_seq_site;
647 * Pointer to the lu_target for this site.
649 struct lu_target *ls_tgt;
652 * Number of objects in lsb_lru_lists - used for shrinking
654 struct percpu_counter ls_lru_len_counter;
657 static inline struct lu_site_bkt_data *
658 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
660 struct cfs_hash_bd bd;
662 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
663 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
666 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
668 return s->ld_seq_site;
672 * Constructors/destructors.
676 int lu_site_init (struct lu_site *s, struct lu_device *d);
677 void lu_site_fini (struct lu_site *s);
678 int lu_site_init_finish (struct lu_site *s);
679 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
680 void lu_device_get (struct lu_device *d);
681 void lu_device_put (struct lu_device *d);
682 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
683 void lu_device_fini (struct lu_device *d);
684 int lu_object_header_init(struct lu_object_header *h);
685 void lu_object_header_fini(struct lu_object_header *h);
686 int lu_object_init (struct lu_object *o,
687 struct lu_object_header *h, struct lu_device *d);
688 void lu_object_fini (struct lu_object *o);
689 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
690 void lu_object_add (struct lu_object *before, struct lu_object *o);
692 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
693 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
696 * Helpers to initialize and finalize device types.
699 int lu_device_type_init(struct lu_device_type *ldt);
700 void lu_device_type_fini(struct lu_device_type *ldt);
705 * Caching and reference counting.
710 * Acquire additional reference to the given object. This function is used to
711 * attain additional reference. To acquire initial reference use
714 static inline void lu_object_get(struct lu_object *o)
716 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
717 atomic_inc(&o->lo_header->loh_ref);
721 * Return true of object will not be cached after last reference to it is
724 static inline int lu_object_is_dying(const struct lu_object_header *h)
726 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
729 void lu_object_put(const struct lu_env *env, struct lu_object *o);
730 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
731 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
732 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
735 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
738 return lu_site_purge_objects(env, s, nr, 1);
741 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
742 lu_printer_t printer);
743 struct lu_object *lu_object_find(const struct lu_env *env,
744 struct lu_device *dev, const struct lu_fid *f,
745 const struct lu_object_conf *conf);
746 struct lu_object *lu_object_find_at(const struct lu_env *env,
747 struct lu_device *dev,
748 const struct lu_fid *f,
749 const struct lu_object_conf *conf);
750 struct lu_object *lu_object_find_slice(const struct lu_env *env,
751 struct lu_device *dev,
752 const struct lu_fid *f,
753 const struct lu_object_conf *conf);
762 * First (topmost) sub-object of given compound object
764 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
766 LASSERT(!list_empty(&h->loh_layers));
767 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
771 * Next sub-object in the layering
773 static inline struct lu_object *lu_object_next(const struct lu_object *o)
775 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
779 * Pointer to the fid of this object.
781 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
783 return &o->lo_header->loh_fid;
787 * return device operations vector for this object
789 static const inline struct lu_device_operations *
790 lu_object_ops(const struct lu_object *o)
792 return o->lo_dev->ld_ops;
796 * Given a compound object, find its slice, corresponding to the device type
799 struct lu_object *lu_object_locate(struct lu_object_header *h,
800 const struct lu_device_type *dtype);
803 * Printer function emitting messages through libcfs_debug_msg().
805 int lu_cdebug_printer(const struct lu_env *env,
806 void *cookie, const char *format, ...);
809 * Print object description followed by a user-supplied message.
811 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
813 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
814 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
815 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
816 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
821 * Print short object description followed by a user-supplied message.
823 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
825 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
826 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
827 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
828 (object)->lo_header); \
829 lu_cdebug_printer(env, &msgdata, "\n"); \
830 CDEBUG(mask, format , ## __VA_ARGS__); \
834 void lu_object_print (const struct lu_env *env, void *cookie,
835 lu_printer_t printer, const struct lu_object *o);
836 void lu_object_header_print(const struct lu_env *env, void *cookie,
837 lu_printer_t printer,
838 const struct lu_object_header *hdr);
841 * Check object consistency.
843 int lu_object_invariant(const struct lu_object *o);
847 * Check whether object exists, no matter on local or remote storage.
848 * Note: LOHA_EXISTS will be set once some one created the object,
849 * and it does not needs to be committed to storage.
851 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
854 * Check whether object on the remote storage.
856 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
858 static inline int lu_object_assert_exists(const struct lu_object *o)
860 return lu_object_exists(o);
863 static inline int lu_object_assert_not_exists(const struct lu_object *o)
865 return !lu_object_exists(o);
869 * Attr of this object.
871 static inline __u32 lu_object_attr(const struct lu_object *o)
873 LASSERT(lu_object_exists(o) != 0);
874 return o->lo_header->loh_attr;
877 static inline void lu_object_ref_add(struct lu_object *o,
881 lu_ref_add(&o->lo_header->loh_reference, scope, source);
884 static inline void lu_object_ref_add_at(struct lu_object *o,
885 struct lu_ref_link *link,
889 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
892 static inline void lu_object_ref_del(struct lu_object *o,
893 const char *scope, const void *source)
895 lu_ref_del(&o->lo_header->loh_reference, scope, source);
898 static inline void lu_object_ref_del_at(struct lu_object *o,
899 struct lu_ref_link *link,
900 const char *scope, const void *source)
902 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
905 /** input params, should be filled out by mdt */
909 /** count in bytes */
910 unsigned int rp_count;
911 /** number of pages */
912 unsigned int rp_npages;
913 /** requested attr */
915 /** pointers to pages */
916 struct page **rp_pages;
919 enum lu_xattr_flags {
920 LU_XATTR_REPLACE = (1 << 0),
921 LU_XATTR_CREATE = (1 << 1)
929 /** For lu_context health-checks */
930 enum lu_context_state {
938 * lu_context. Execution context for lu_object methods. Currently associated
941 * All lu_object methods, except device and device type methods (called during
942 * system initialization and shutdown) are executed "within" some
943 * lu_context. This means, that pointer to some "current" lu_context is passed
944 * as an argument to all methods.
946 * All service ptlrpc threads create lu_context as part of their
947 * initialization. It is possible to create "stand-alone" context for other
948 * execution environments (like system calls).
950 * lu_object methods mainly use lu_context through lu_context_key interface
951 * that allows each layer to associate arbitrary pieces of data with each
952 * context (see pthread_key_create(3) for similar interface).
954 * On a client, lu_context is bound to a thread, see cl_env_get().
956 * \see lu_context_key
960 * lu_context is used on the client side too. Yet we don't want to
961 * allocate values of server-side keys for the client contexts and
964 * To achieve this, set of tags in introduced. Contexts and keys are
965 * marked with tags. Key value are created only for context whose set
966 * of tags has non-empty intersection with one for key. Tags are taken
967 * from enum lu_context_tag.
970 enum lu_context_state lc_state;
972 * Pointer to the home service thread. NULL for other execution
975 struct ptlrpc_thread *lc_thread;
977 * Pointer to an array with key values. Internal implementation
982 * Linkage into a list of all remembered contexts. Only
983 * `non-transient' contexts, i.e., ones created for service threads
986 struct list_head lc_remember;
988 * Version counter used to skip calls to lu_context_refill() when no
989 * keys were registered.
999 * lu_context_key interface. Similar to pthread_key.
1002 enum lu_context_tag {
1004 * Thread on md server
1006 LCT_MD_THREAD = 1 << 0,
1008 * Thread on dt server
1010 LCT_DT_THREAD = 1 << 1,
1012 * Context for transaction handle
1014 LCT_TX_HANDLE = 1 << 2,
1018 LCT_CL_THREAD = 1 << 3,
1020 * A per-request session on a server, and a per-system-call session on
1023 LCT_SESSION = 1 << 4,
1025 * A per-request data on OSP device
1027 LCT_OSP_THREAD = 1 << 5,
1031 LCT_MG_THREAD = 1 << 6,
1033 * Context for local operations
1037 * session for server thread
1039 LCT_SERVER_SESSION = 1 << 8,
1041 * Set when at least one of keys, having values in this context has
1042 * non-NULL lu_context_key::lct_exit() method. This is used to
1043 * optimize lu_context_exit() call.
1045 LCT_HAS_EXIT = 1 << 28,
1047 * Don't add references for modules creating key values in that context.
1048 * This is only for contexts used internally by lu_object framework.
1050 LCT_NOREF = 1 << 29,
1052 * Key is being prepared for retiring, don't create new values for it.
1054 LCT_QUIESCENT = 1 << 30,
1056 * Context should be remembered.
1058 LCT_REMEMBER = 1 << 31,
1060 * Contexts usable in cache shrinker thread.
1062 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1066 * Key. Represents per-context value slot.
1068 * Keys are usually registered when module owning the key is initialized, and
1069 * de-registered when module is unloaded. Once key is registered, all new
1070 * contexts with matching tags, will get key value. "Old" contexts, already
1071 * initialized at the time of key registration, can be forced to get key value
1072 * by calling lu_context_refill().
1074 * Every key value is counted in lu_context_key::lct_used and acquires a
1075 * reference on an owning module. This means, that all key values have to be
1076 * destroyed before module can be unloaded. This is usually achieved by
1077 * stopping threads started by the module, that created contexts in their
1078 * entry functions. Situation is complicated by the threads shared by multiple
1079 * modules, like ptlrpcd daemon on a client. To work around this problem,
1080 * contexts, created in such threads, are `remembered' (see
1081 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1082 * for unloading it does the following:
1084 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1085 * preventing new key values from being allocated in the new contexts,
1088 * - scans a list of remembered contexts, destroying values of module
1089 * keys, thus releasing references to the module.
1091 * This is done by lu_context_key_quiesce(). If module is re-activated
1092 * before key has been de-registered, lu_context_key_revive() call clears
1093 * `quiescent' marker.
1095 * lu_context code doesn't provide any internal synchronization for these
1096 * activities---it's assumed that startup (including threads start-up) and
1097 * shutdown are serialized by some external means.
1101 struct lu_context_key {
1103 * Set of tags for which values of this key are to be instantiated.
1107 * Value constructor. This is called when new value is created for a
1108 * context. Returns pointer to new value of error pointer.
1110 void *(*lct_init)(const struct lu_context *ctx,
1111 struct lu_context_key *key);
1113 * Value destructor. Called when context with previously allocated
1114 * value of this slot is destroyed. \a data is a value that was returned
1115 * by a matching call to lu_context_key::lct_init().
1117 void (*lct_fini)(const struct lu_context *ctx,
1118 struct lu_context_key *key, void *data);
1120 * Optional method called on lu_context_exit() for all allocated
1121 * keys. Can be used by debugging code checking that locks are
1124 void (*lct_exit)(const struct lu_context *ctx,
1125 struct lu_context_key *key, void *data);
1127 * Internal implementation detail: index within lu_context::lc_value[]
1128 * reserved for this key.
1132 * Internal implementation detail: number of values created for this
1137 * Internal implementation detail: module for this key.
1139 struct module *lct_owner;
1141 * References to this key. For debugging.
1143 struct lu_ref lct_reference;
1146 #define LU_KEY_INIT(mod, type) \
1147 static void* mod##_key_init(const struct lu_context *ctx, \
1148 struct lu_context_key *key) \
1152 CLASSERT(PAGE_CACHE_SIZE >= sizeof (*value)); \
1154 OBD_ALLOC_PTR(value); \
1155 if (value == NULL) \
1156 value = ERR_PTR(-ENOMEM); \
1160 struct __##mod##__dummy_init {;} /* semicolon catcher */
1162 #define LU_KEY_FINI(mod, type) \
1163 static void mod##_key_fini(const struct lu_context *ctx, \
1164 struct lu_context_key *key, void* data) \
1166 type *info = data; \
1168 OBD_FREE_PTR(info); \
1170 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1172 #define LU_KEY_INIT_FINI(mod, type) \
1173 LU_KEY_INIT(mod,type); \
1174 LU_KEY_FINI(mod,type)
1176 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1177 struct lu_context_key mod##_thread_key = { \
1179 .lct_init = mod##_key_init, \
1180 .lct_fini = mod##_key_fini \
1183 #define LU_CONTEXT_KEY_INIT(key) \
1185 (key)->lct_owner = THIS_MODULE; \
1188 int lu_context_key_register(struct lu_context_key *key);
1189 void lu_context_key_degister(struct lu_context_key *key);
1190 void *lu_context_key_get (const struct lu_context *ctx,
1191 const struct lu_context_key *key);
1192 void lu_context_key_quiesce (struct lu_context_key *key);
1193 void lu_context_key_revive (struct lu_context_key *key);
1197 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1201 #define LU_KEY_INIT_GENERIC(mod) \
1202 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1204 struct lu_context_key *key = k; \
1207 va_start(args, k); \
1209 LU_CONTEXT_KEY_INIT(key); \
1210 key = va_arg(args, struct lu_context_key *); \
1211 } while (key != NULL); \
1215 #define LU_TYPE_INIT(mod, ...) \
1216 LU_KEY_INIT_GENERIC(mod) \
1217 static int mod##_type_init(struct lu_device_type *t) \
1219 mod##_key_init_generic(__VA_ARGS__, NULL); \
1220 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1222 struct __##mod##_dummy_type_init {;}
1224 #define LU_TYPE_FINI(mod, ...) \
1225 static void mod##_type_fini(struct lu_device_type *t) \
1227 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1229 struct __##mod##_dummy_type_fini {;}
1231 #define LU_TYPE_START(mod, ...) \
1232 static void mod##_type_start(struct lu_device_type *t) \
1234 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1236 struct __##mod##_dummy_type_start {;}
1238 #define LU_TYPE_STOP(mod, ...) \
1239 static void mod##_type_stop(struct lu_device_type *t) \
1241 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1243 struct __##mod##_dummy_type_stop {;}
1247 #define LU_TYPE_INIT_FINI(mod, ...) \
1248 LU_TYPE_INIT(mod, __VA_ARGS__); \
1249 LU_TYPE_FINI(mod, __VA_ARGS__); \
1250 LU_TYPE_START(mod, __VA_ARGS__); \
1251 LU_TYPE_STOP(mod, __VA_ARGS__)
1253 int lu_context_init (struct lu_context *ctx, __u32 tags);
1254 void lu_context_fini (struct lu_context *ctx);
1255 void lu_context_enter (struct lu_context *ctx);
1256 void lu_context_exit (struct lu_context *ctx);
1257 int lu_context_refill(struct lu_context *ctx);
1260 * Helper functions to operate on multiple keys. These are used by the default
1261 * device type operations, defined by LU_TYPE_INIT_FINI().
1264 int lu_context_key_register_many(struct lu_context_key *k, ...);
1265 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1266 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1267 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1270 * update/clear ctx/ses tags.
1272 void lu_context_tags_update(__u32 tags);
1273 void lu_context_tags_clear(__u32 tags);
1274 void lu_session_tags_update(__u32 tags);
1275 void lu_session_tags_clear(__u32 tags);
1282 * "Local" context, used to store data instead of stack.
1284 struct lu_context le_ctx;
1286 * "Session" context for per-request data.
1288 struct lu_context *le_ses;
1291 int lu_env_init (struct lu_env *env, __u32 tags);
1292 void lu_env_fini (struct lu_env *env);
1293 int lu_env_refill(struct lu_env *env);
1294 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1296 /** @} lu_context */
1299 * Output site statistical counters into a buffer. Suitable for
1300 * ll_rd_*()-style functions.
1302 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1305 * Common name structure to be passed around for various name related methods.
1308 const char *ln_name;
1313 * Validate names (path components)
1315 * To be valid \a name must be non-empty, '\0' terminated of length \a
1316 * name_len, and not contain '/'. The maximum length of a name (before
1317 * say -ENAMETOOLONG will be returned) is really controlled by llite
1318 * and the server. We only check for something insane coming from bad
1319 * integer handling here.
1321 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1323 return name != NULL &&
1325 name_len < INT_MAX &&
1326 name[name_len] == '\0' &&
1327 strlen(name) == name_len &&
1328 memchr(name, '/', name_len) == NULL;
1331 static inline bool lu_name_is_valid(const struct lu_name *ln)
1333 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1336 #define DNAME "%.*s"
1338 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1339 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1342 * Common buffer structure to be passed around for various xattr_{s,g}et()
1350 #define DLUBUF "(%p %zu)"
1351 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1353 /* read buffer params, should be filled out by out */
1355 /** number of buffers */
1356 unsigned int rb_nbufs;
1357 /** pointers to buffers */
1358 struct lu_buf rb_bufs[];
1362 * One-time initializers, called at obdclass module initialization, not
1367 * Initialization of global lu_* data.
1369 int lu_global_init(void);
1372 * Dual to lu_global_init().
1374 void lu_global_fini(void);
1376 struct lu_kmem_descr {
1377 struct kmem_cache **ckd_cache;
1378 const char *ckd_name;
1379 const size_t ckd_size;
1382 int lu_kmem_init(struct lu_kmem_descr *caches);
1383 void lu_kmem_fini(struct lu_kmem_descr *caches);
1385 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1386 const struct lu_fid *fid);
1387 struct lu_object *lu_object_anon(const struct lu_env *env,
1388 struct lu_device *dev,
1389 const struct lu_object_conf *conf);
1392 extern struct lu_buf LU_BUF_NULL;
1394 void lu_buf_free(struct lu_buf *buf);
1395 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1396 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1398 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1399 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1401 extern __u32 lu_context_tags_default;
1402 extern __u32 lu_session_tags_default;
1404 static inline bool lu_device_is_cl(const struct lu_device *d)
1406 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1409 static inline bool lu_object_is_cl(const struct lu_object *o)
1411 return lu_device_is_cl(o->lo_dev);
1415 #endif /* __LUSTRE_LU_OBJECT_H */