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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef __LUSTRE_LU_OBJECT_H
34 #define __LUSTRE_LU_OBJECT_H
37 #include <libcfs/libcfs.h>
38 #include <lustre/lustre_idl.h>
40 #include <linux/percpu_counter.h>
43 struct proc_dir_entry;
48 * lu_* data-types represent server-side entities shared by data and meta-data
53 * -# support for layering.
55 * Server side object is split into layers, one per device in the
56 * corresponding device stack. Individual layer is represented by struct
57 * lu_object. Compound layered object --- by struct lu_object_header. Most
58 * interface functions take lu_object as an argument and operate on the
59 * whole compound object. This decision was made due to the following
62 * - it's envisaged that lu_object will be used much more often than
65 * - we want lower (non-top) layers to be able to initiate operations
66 * on the whole object.
68 * Generic code supports layering more complex than simple stacking, e.g.,
69 * it is possible that at some layer object "spawns" multiple sub-objects
72 * -# fid-based identification.
74 * Compound object is uniquely identified by its fid. Objects are indexed
75 * by their fids (hash table is used for index).
77 * -# caching and life-cycle management.
79 * Object's life-time is controlled by reference counting. When reference
80 * count drops to 0, object is returned to cache. Cached objects still
81 * retain their identity (i.e., fid), and can be recovered from cache.
83 * Objects are kept in the global LRU list, and lu_site_purge() function
84 * can be used to reclaim given number of unused objects from the tail of
87 * -# avoiding recursion.
89 * Generic code tries to replace recursion through layers by iterations
90 * where possible. Additionally to the end of reducing stack consumption,
91 * data, when practically possible, are allocated through lu_context_key
92 * interface rather than on stack.
99 struct lu_object_header;
104 * Operations common for data and meta-data devices.
106 struct lu_device_operations {
108 * Allocate object for the given device (without lower-layer
109 * parts). This is called by lu_object_operations::loo_object_init()
110 * from the parent layer, and should setup at least lu_object::lo_dev
111 * and lu_object::lo_ops fields of resulting lu_object.
113 * Object creation protocol.
115 * Due to design goal of avoiding recursion, object creation (see
116 * lu_object_alloc()) is somewhat involved:
118 * - first, lu_device_operations::ldo_object_alloc() method of the
119 * top-level device in the stack is called. It should allocate top
120 * level object (including lu_object_header), but without any
121 * lower-layer sub-object(s).
123 * - then lu_object_alloc() sets fid in the header of newly created
126 * - then lu_object_operations::loo_object_init() is called. It has
127 * to allocate lower-layer object(s). To do this,
128 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
129 * of the lower-layer device(s).
131 * - for all new objects allocated by
132 * lu_object_operations::loo_object_init() (and inserted into object
133 * stack), lu_object_operations::loo_object_init() is called again
134 * repeatedly, until no new objects are created.
136 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
137 * result->lo_ops != NULL);
139 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
140 const struct lu_object_header *h,
141 struct lu_device *d);
143 * process config specific for device.
145 int (*ldo_process_config)(const struct lu_env *env,
146 struct lu_device *, struct lustre_cfg *);
147 int (*ldo_recovery_complete)(const struct lu_env *,
151 * initialize local objects for device. this method called after layer has
152 * been initialized (after LCFG_SETUP stage) and before it starts serving
156 int (*ldo_prepare)(const struct lu_env *,
157 struct lu_device *parent,
158 struct lu_device *dev);
163 * For lu_object_conf flags
166 /* This is a new object to be allocated, or the file
167 * corresponding to the object does not exists. */
168 LOC_F_NEW = 0x00000001,
170 /* When find a dying object, just return -EAGAIN at once instead of
171 * blocking the thread. */
172 LOC_F_NOWAIT = 0x00000002,
176 * Object configuration, describing particulars of object being created. On
177 * server this is not used, as server objects are full identified by fid. On
178 * client configuration contains struct lustre_md.
180 struct lu_object_conf {
182 * Some hints for obj find and alloc.
184 loc_flags_t loc_flags;
188 * Type of "printer" function used by lu_object_operations::loo_object_print()
191 * Printer function is needed to provide some flexibility in (semi-)debugging
192 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
194 typedef int (*lu_printer_t)(const struct lu_env *env,
195 void *cookie, const char *format, ...)
196 __attribute__ ((format (printf, 3, 4)));
199 * Operations specific for particular lu_object.
201 struct lu_object_operations {
204 * Allocate lower-layer parts of the object by calling
205 * lu_device_operations::ldo_object_alloc() of the corresponding
208 * This method is called once for each object inserted into object
209 * stack. It's responsibility of this method to insert lower-layer
210 * object(s) it create into appropriate places of object stack.
212 int (*loo_object_init)(const struct lu_env *env,
214 const struct lu_object_conf *conf);
216 * Called (in top-to-bottom order) during object allocation after all
217 * layers were allocated and initialized. Can be used to perform
218 * initialization depending on lower layers.
220 int (*loo_object_start)(const struct lu_env *env,
221 struct lu_object *o);
223 * Called before lu_object_operations::loo_object_free() to signal
224 * that object is being destroyed. Dual to
225 * lu_object_operations::loo_object_init().
227 void (*loo_object_delete)(const struct lu_env *env,
228 struct lu_object *o);
230 * Dual to lu_device_operations::ldo_object_alloc(). Called when
231 * object is removed from memory.
233 void (*loo_object_free)(const struct lu_env *env,
234 struct lu_object *o);
236 * Called when last active reference to the object is released (and
237 * object returns to the cache). This method is optional.
239 void (*loo_object_release)(const struct lu_env *env,
240 struct lu_object *o);
242 * Optional debugging helper. Print given object.
244 int (*loo_object_print)(const struct lu_env *env, void *cookie,
245 lu_printer_t p, const struct lu_object *o);
247 * Optional debugging method. Returns true iff method is internally
250 int (*loo_object_invariant)(const struct lu_object *o);
256 struct lu_device_type;
259 * Device: a layer in the server side abstraction stacking.
263 * reference count. This is incremented, in particular, on each object
264 * created at this layer.
266 * \todo XXX which means that atomic_t is probably too small.
270 * Pointer to device type. Never modified once set.
272 struct lu_device_type *ld_type;
274 * Operation vector for this device.
276 const struct lu_device_operations *ld_ops;
278 * Stack this device belongs to.
280 struct lu_site *ld_site;
281 struct proc_dir_entry *ld_proc_entry;
283 /** \todo XXX: temporary back pointer into obd. */
284 struct obd_device *ld_obd;
286 * A list of references to this object, for debugging.
288 struct lu_ref ld_reference;
290 * Link the device to the site.
292 struct list_head ld_linkage;
295 struct lu_device_type_operations;
298 * Tag bits for device type. They are used to distinguish certain groups of
302 /** this is meta-data device */
303 LU_DEVICE_MD = (1 << 0),
304 /** this is data device */
305 LU_DEVICE_DT = (1 << 1),
306 /** data device in the client stack */
307 LU_DEVICE_CL = (1 << 2)
313 struct lu_device_type {
315 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
319 * Name of this class. Unique system-wide. Never modified once set.
323 * Operations for this type.
325 const struct lu_device_type_operations *ldt_ops;
327 * \todo XXX: temporary pointer to associated obd_type.
329 struct obd_type *ldt_obd_type;
331 * \todo XXX: temporary: context tags used by obd_*() calls.
335 * Number of existing device type instances.
337 atomic_t ldt_device_nr;
339 * Linkage into a global list of all device types.
341 * \see lu_device_types.
343 struct list_head ldt_linkage;
347 * Operations on a device type.
349 struct lu_device_type_operations {
351 * Allocate new device.
353 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
354 struct lu_device_type *t,
355 struct lustre_cfg *lcfg);
357 * Free device. Dual to
358 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
359 * the next device in the stack.
361 struct lu_device *(*ldto_device_free)(const struct lu_env *,
365 * Initialize the devices after allocation
367 int (*ldto_device_init)(const struct lu_env *env,
368 struct lu_device *, const char *,
371 * Finalize device. Dual to
372 * lu_device_type_operations::ldto_device_init(). Returns pointer to
373 * the next device in the stack.
375 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
378 * Initialize device type. This is called on module load.
380 int (*ldto_init)(struct lu_device_type *t);
382 * Finalize device type. Dual to
383 * lu_device_type_operations::ldto_init(). Called on module unload.
385 void (*ldto_fini)(struct lu_device_type *t);
387 * Called when the first device is created.
389 void (*ldto_start)(struct lu_device_type *t);
391 * Called when number of devices drops to 0.
393 void (*ldto_stop)(struct lu_device_type *t);
396 static inline int lu_device_is_md(const struct lu_device *d)
398 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
402 * Common object attributes.
407 /** modification time in seconds since Epoch */
409 /** access time in seconds since Epoch */
411 /** change time in seconds since Epoch */
413 /** 512-byte blocks allocated to object */
415 /** permission bits and file type */
423 /** number of persistent references to this object */
425 /** blk bits of the object*/
427 /** blk size of the object*/
439 /** Bit-mask of valid attributes */
453 LA_BLKSIZE = 1 << 12,
454 LA_KILL_SUID = 1 << 13,
455 LA_KILL_SGID = 1 << 14,
459 * Layer in the layered object.
463 * Header for this object.
465 struct lu_object_header *lo_header;
467 * Device for this layer.
469 struct lu_device *lo_dev;
471 * Operations for this object.
473 const struct lu_object_operations *lo_ops;
475 * Linkage into list of all layers.
477 struct list_head lo_linkage;
479 * Link to the device, for debugging.
481 struct lu_ref_link lo_dev_ref;
484 enum lu_object_header_flags {
486 * Don't keep this object in cache. Object will be destroyed as soon
487 * as last reference to it is released. This flag cannot be cleared
490 LU_OBJECT_HEARD_BANSHEE = 0,
492 * Mark this object has already been taken out of cache.
494 LU_OBJECT_UNHASHED = 1,
497 enum lu_object_header_attr {
498 LOHA_EXISTS = 1 << 0,
499 LOHA_REMOTE = 1 << 1,
501 * UNIX file type is stored in S_IFMT bits.
503 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
504 LOHA_FT_END = 017 << 12, /**< S_IFMT */
508 * "Compound" object, consisting of multiple layers.
510 * Compound object with given fid is unique with given lu_site.
512 * Note, that object does *not* necessary correspond to the real object in the
513 * persistent storage: object is an anchor for locking and method calling, so
514 * it is created for things like not-yet-existing child created by mkdir or
515 * create calls. lu_object_operations::loo_exists() can be used to check
516 * whether object is backed by persistent storage entity.
518 struct lu_object_header {
520 * Fid, uniquely identifying this object.
522 struct lu_fid loh_fid;
524 * Object flags from enum lu_object_header_flags. Set and checked
527 unsigned long loh_flags;
529 * Object reference count. Protected by lu_site::ls_guard.
533 * Common object attributes, cached for efficiency. From enum
534 * lu_object_header_attr.
538 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
540 struct hlist_node loh_hash;
542 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
544 struct list_head loh_lru;
546 * Linkage into list of layers. Never modified once set (except lately
547 * during object destruction). No locking is necessary.
549 struct list_head loh_layers;
551 * A list of references to this object, for debugging.
553 struct lu_ref loh_reference;
558 struct lu_site_bkt_data {
560 * number of object in this bucket on the lsb_lru list.
564 * LRU list, updated on each access to object. Protected by
565 * bucket lock of lu_site::ls_obj_hash.
567 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
568 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
569 * of list_for_each_entry_safe_reverse()).
571 struct list_head lsb_lru;
573 * Wait-queue signaled when an object in this site is ultimately
574 * destroyed (lu_object_free()). It is used by lu_object_find() to
575 * wait before re-trying when object in the process of destruction is
576 * found in the hash table.
578 * \see htable_lookup().
580 wait_queue_head_t lsb_marche_funebre;
588 LU_SS_CACHE_DEATH_RACE,
594 * lu_site is a "compartment" within which objects are unique, and LRU
595 * discipline is maintained.
597 * lu_site exists so that multiple layered stacks can co-exist in the same
600 * lu_site has the same relation to lu_device as lu_object_header to
607 struct cfs_hash *ls_obj_hash;
609 * index of bucket on hash table while purging
611 unsigned int ls_purge_start;
613 * Top-level device for this stack.
615 struct lu_device *ls_top_dev;
617 * Bottom-level device for this stack
619 struct lu_device *ls_bottom_dev;
621 * Linkage into global list of sites.
623 struct list_head ls_linkage;
625 * List for lu device for this site, protected
628 struct list_head ls_ld_linkage;
629 spinlock_t ls_ld_lock;
631 * Lock to serialize site purge.
633 struct mutex ls_purge_mutex;
637 struct lprocfs_stats *ls_stats;
639 * XXX: a hack! fld has to find md_site via site, remove when possible
641 struct seq_server_site *ld_seq_site;
643 * Pointer to the lu_target for this site.
645 struct lu_target *ls_tgt;
648 * Number of objects in lsb_lru_lists - used for shrinking
650 struct percpu_counter ls_lru_len_counter;
653 static inline struct lu_site_bkt_data *
654 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
656 struct cfs_hash_bd bd;
658 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
659 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
662 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
664 return s->ld_seq_site;
668 * Constructors/destructors.
672 int lu_site_init (struct lu_site *s, struct lu_device *d);
673 void lu_site_fini (struct lu_site *s);
674 int lu_site_init_finish (struct lu_site *s);
675 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
676 void lu_device_get (struct lu_device *d);
677 void lu_device_put (struct lu_device *d);
678 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
679 void lu_device_fini (struct lu_device *d);
680 int lu_object_header_init(struct lu_object_header *h);
681 void lu_object_header_fini(struct lu_object_header *h);
682 int lu_object_init (struct lu_object *o,
683 struct lu_object_header *h, struct lu_device *d);
684 void lu_object_fini (struct lu_object *o);
685 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
686 void lu_object_add (struct lu_object *before, struct lu_object *o);
688 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
689 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
692 * Helpers to initialize and finalize device types.
695 int lu_device_type_init(struct lu_device_type *ldt);
696 void lu_device_type_fini(struct lu_device_type *ldt);
701 * Caching and reference counting.
706 * Acquire additional reference to the given object. This function is used to
707 * attain additional reference. To acquire initial reference use
710 static inline void lu_object_get(struct lu_object *o)
712 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
713 atomic_inc(&o->lo_header->loh_ref);
717 * Return true of object will not be cached after last reference to it is
720 static inline int lu_object_is_dying(const struct lu_object_header *h)
722 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
725 void lu_object_put(const struct lu_env *env, struct lu_object *o);
726 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
727 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
728 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
731 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
734 return lu_site_purge_objects(env, s, nr, 1);
737 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
738 lu_printer_t printer);
739 struct lu_object *lu_object_find(const struct lu_env *env,
740 struct lu_device *dev, const struct lu_fid *f,
741 const struct lu_object_conf *conf);
742 struct lu_object *lu_object_find_at(const struct lu_env *env,
743 struct lu_device *dev,
744 const struct lu_fid *f,
745 const struct lu_object_conf *conf);
746 struct lu_object *lu_object_find_slice(const struct lu_env *env,
747 struct lu_device *dev,
748 const struct lu_fid *f,
749 const struct lu_object_conf *conf);
758 * First (topmost) sub-object of given compound object
760 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
762 LASSERT(!list_empty(&h->loh_layers));
763 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
767 * Next sub-object in the layering
769 static inline struct lu_object *lu_object_next(const struct lu_object *o)
771 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
775 * Pointer to the fid of this object.
777 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
779 return &o->lo_header->loh_fid;
783 * return device operations vector for this object
785 static const inline struct lu_device_operations *
786 lu_object_ops(const struct lu_object *o)
788 return o->lo_dev->ld_ops;
792 * Given a compound object, find its slice, corresponding to the device type
795 struct lu_object *lu_object_locate(struct lu_object_header *h,
796 const struct lu_device_type *dtype);
799 * Printer function emitting messages through libcfs_debug_msg().
801 int lu_cdebug_printer(const struct lu_env *env,
802 void *cookie, const char *format, ...);
805 * Print object description followed by a user-supplied message.
807 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
809 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
810 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
811 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
812 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
817 * Print short object description followed by a user-supplied message.
819 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
821 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
822 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
823 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
824 (object)->lo_header); \
825 lu_cdebug_printer(env, &msgdata, "\n"); \
826 CDEBUG(mask, format , ## __VA_ARGS__); \
830 void lu_object_print (const struct lu_env *env, void *cookie,
831 lu_printer_t printer, const struct lu_object *o);
832 void lu_object_header_print(const struct lu_env *env, void *cookie,
833 lu_printer_t printer,
834 const struct lu_object_header *hdr);
837 * Check object consistency.
839 int lu_object_invariant(const struct lu_object *o);
843 * Check whether object exists, no matter on local or remote storage.
844 * Note: LOHA_EXISTS will be set once some one created the object,
845 * and it does not needs to be committed to storage.
847 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
850 * Check whether object on the remote storage.
852 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
854 static inline int lu_object_assert_exists(const struct lu_object *o)
856 return lu_object_exists(o);
859 static inline int lu_object_assert_not_exists(const struct lu_object *o)
861 return !lu_object_exists(o);
865 * Attr of this object.
867 static inline __u32 lu_object_attr(const struct lu_object *o)
869 LASSERT(lu_object_exists(o) != 0);
870 return o->lo_header->loh_attr;
873 static inline void lu_object_ref_add(struct lu_object *o,
877 lu_ref_add(&o->lo_header->loh_reference, scope, source);
880 static inline void lu_object_ref_add_at(struct lu_object *o,
881 struct lu_ref_link *link,
885 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
888 static inline void lu_object_ref_del(struct lu_object *o,
889 const char *scope, const void *source)
891 lu_ref_del(&o->lo_header->loh_reference, scope, source);
894 static inline void lu_object_ref_del_at(struct lu_object *o,
895 struct lu_ref_link *link,
896 const char *scope, const void *source)
898 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
901 /** input params, should be filled out by mdt */
905 /** count in bytes */
906 unsigned int rp_count;
907 /** number of pages */
908 unsigned int rp_npages;
909 /** requested attr */
911 /** pointers to pages */
912 struct page **rp_pages;
915 enum lu_xattr_flags {
916 LU_XATTR_REPLACE = (1 << 0),
917 LU_XATTR_CREATE = (1 << 1)
925 /** For lu_context health-checks */
926 enum lu_context_state {
934 * lu_context. Execution context for lu_object methods. Currently associated
937 * All lu_object methods, except device and device type methods (called during
938 * system initialization and shutdown) are executed "within" some
939 * lu_context. This means, that pointer to some "current" lu_context is passed
940 * as an argument to all methods.
942 * All service ptlrpc threads create lu_context as part of their
943 * initialization. It is possible to create "stand-alone" context for other
944 * execution environments (like system calls).
946 * lu_object methods mainly use lu_context through lu_context_key interface
947 * that allows each layer to associate arbitrary pieces of data with each
948 * context (see pthread_key_create(3) for similar interface).
950 * On a client, lu_context is bound to a thread, see cl_env_get().
952 * \see lu_context_key
956 * lu_context is used on the client side too. Yet we don't want to
957 * allocate values of server-side keys for the client contexts and
960 * To achieve this, set of tags in introduced. Contexts and keys are
961 * marked with tags. Key value are created only for context whose set
962 * of tags has non-empty intersection with one for key. Tags are taken
963 * from enum lu_context_tag.
966 enum lu_context_state lc_state;
968 * Pointer to the home service thread. NULL for other execution
971 struct ptlrpc_thread *lc_thread;
973 * Pointer to an array with key values. Internal implementation
978 * Linkage into a list of all remembered contexts. Only
979 * `non-transient' contexts, i.e., ones created for service threads
982 struct list_head lc_remember;
984 * Version counter used to skip calls to lu_context_refill() when no
985 * keys were registered.
995 * lu_context_key interface. Similar to pthread_key.
998 enum lu_context_tag {
1000 * Thread on md server
1002 LCT_MD_THREAD = 1 << 0,
1004 * Thread on dt server
1006 LCT_DT_THREAD = 1 << 1,
1008 * Context for transaction handle
1010 LCT_TX_HANDLE = 1 << 2,
1014 LCT_CL_THREAD = 1 << 3,
1016 * A per-request session on a server, and a per-system-call session on
1019 LCT_SESSION = 1 << 4,
1021 * A per-request data on OSP device
1023 LCT_OSP_THREAD = 1 << 5,
1027 LCT_MG_THREAD = 1 << 6,
1029 * Context for local operations
1033 * session for server thread
1035 LCT_SERVER_SESSION = 1 << 8,
1037 * Set when at least one of keys, having values in this context has
1038 * non-NULL lu_context_key::lct_exit() method. This is used to
1039 * optimize lu_context_exit() call.
1041 LCT_HAS_EXIT = 1 << 28,
1043 * Don't add references for modules creating key values in that context.
1044 * This is only for contexts used internally by lu_object framework.
1046 LCT_NOREF = 1 << 29,
1048 * Key is being prepared for retiring, don't create new values for it.
1050 LCT_QUIESCENT = 1 << 30,
1052 * Context should be remembered.
1054 LCT_REMEMBER = 1 << 31,
1056 * Contexts usable in cache shrinker thread.
1058 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1062 * Key. Represents per-context value slot.
1064 * Keys are usually registered when module owning the key is initialized, and
1065 * de-registered when module is unloaded. Once key is registered, all new
1066 * contexts with matching tags, will get key value. "Old" contexts, already
1067 * initialized at the time of key registration, can be forced to get key value
1068 * by calling lu_context_refill().
1070 * Every key value is counted in lu_context_key::lct_used and acquires a
1071 * reference on an owning module. This means, that all key values have to be
1072 * destroyed before module can be unloaded. This is usually achieved by
1073 * stopping threads started by the module, that created contexts in their
1074 * entry functions. Situation is complicated by the threads shared by multiple
1075 * modules, like ptlrpcd daemon on a client. To work around this problem,
1076 * contexts, created in such threads, are `remembered' (see
1077 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1078 * for unloading it does the following:
1080 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1081 * preventing new key values from being allocated in the new contexts,
1084 * - scans a list of remembered contexts, destroying values of module
1085 * keys, thus releasing references to the module.
1087 * This is done by lu_context_key_quiesce(). If module is re-activated
1088 * before key has been de-registered, lu_context_key_revive() call clears
1089 * `quiescent' marker.
1091 * lu_context code doesn't provide any internal synchronization for these
1092 * activities---it's assumed that startup (including threads start-up) and
1093 * shutdown are serialized by some external means.
1097 struct lu_context_key {
1099 * Set of tags for which values of this key are to be instantiated.
1103 * Value constructor. This is called when new value is created for a
1104 * context. Returns pointer to new value of error pointer.
1106 void *(*lct_init)(const struct lu_context *ctx,
1107 struct lu_context_key *key);
1109 * Value destructor. Called when context with previously allocated
1110 * value of this slot is destroyed. \a data is a value that was returned
1111 * by a matching call to lu_context_key::lct_init().
1113 void (*lct_fini)(const struct lu_context *ctx,
1114 struct lu_context_key *key, void *data);
1116 * Optional method called on lu_context_exit() for all allocated
1117 * keys. Can be used by debugging code checking that locks are
1120 void (*lct_exit)(const struct lu_context *ctx,
1121 struct lu_context_key *key, void *data);
1123 * Internal implementation detail: index within lu_context::lc_value[]
1124 * reserved for this key.
1128 * Internal implementation detail: number of values created for this
1133 * Internal implementation detail: module for this key.
1135 struct module *lct_owner;
1137 * References to this key. For debugging.
1139 struct lu_ref lct_reference;
1142 #define LU_KEY_INIT(mod, type) \
1143 static void* mod##_key_init(const struct lu_context *ctx, \
1144 struct lu_context_key *key) \
1148 CLASSERT(PAGE_SIZE >= sizeof(*value)); \
1150 OBD_ALLOC_PTR(value); \
1151 if (value == NULL) \
1152 value = ERR_PTR(-ENOMEM); \
1156 struct __##mod##__dummy_init {;} /* semicolon catcher */
1158 #define LU_KEY_FINI(mod, type) \
1159 static void mod##_key_fini(const struct lu_context *ctx, \
1160 struct lu_context_key *key, void* data) \
1162 type *info = data; \
1164 OBD_FREE_PTR(info); \
1166 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1168 #define LU_KEY_INIT_FINI(mod, type) \
1169 LU_KEY_INIT(mod,type); \
1170 LU_KEY_FINI(mod,type)
1172 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1173 struct lu_context_key mod##_thread_key = { \
1175 .lct_init = mod##_key_init, \
1176 .lct_fini = mod##_key_fini \
1179 #define LU_CONTEXT_KEY_INIT(key) \
1181 (key)->lct_owner = THIS_MODULE; \
1184 int lu_context_key_register(struct lu_context_key *key);
1185 void lu_context_key_degister(struct lu_context_key *key);
1186 void *lu_context_key_get (const struct lu_context *ctx,
1187 const struct lu_context_key *key);
1188 void lu_context_key_quiesce (struct lu_context_key *key);
1189 void lu_context_key_revive (struct lu_context_key *key);
1193 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1197 #define LU_KEY_INIT_GENERIC(mod) \
1198 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1200 struct lu_context_key *key = k; \
1203 va_start(args, k); \
1205 LU_CONTEXT_KEY_INIT(key); \
1206 key = va_arg(args, struct lu_context_key *); \
1207 } while (key != NULL); \
1211 #define LU_TYPE_INIT(mod, ...) \
1212 LU_KEY_INIT_GENERIC(mod) \
1213 static int mod##_type_init(struct lu_device_type *t) \
1215 mod##_key_init_generic(__VA_ARGS__, NULL); \
1216 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1218 struct __##mod##_dummy_type_init {;}
1220 #define LU_TYPE_FINI(mod, ...) \
1221 static void mod##_type_fini(struct lu_device_type *t) \
1223 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1225 struct __##mod##_dummy_type_fini {;}
1227 #define LU_TYPE_START(mod, ...) \
1228 static void mod##_type_start(struct lu_device_type *t) \
1230 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1232 struct __##mod##_dummy_type_start {;}
1234 #define LU_TYPE_STOP(mod, ...) \
1235 static void mod##_type_stop(struct lu_device_type *t) \
1237 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1239 struct __##mod##_dummy_type_stop {;}
1243 #define LU_TYPE_INIT_FINI(mod, ...) \
1244 LU_TYPE_INIT(mod, __VA_ARGS__); \
1245 LU_TYPE_FINI(mod, __VA_ARGS__); \
1246 LU_TYPE_START(mod, __VA_ARGS__); \
1247 LU_TYPE_STOP(mod, __VA_ARGS__)
1249 int lu_context_init (struct lu_context *ctx, __u32 tags);
1250 void lu_context_fini (struct lu_context *ctx);
1251 void lu_context_enter (struct lu_context *ctx);
1252 void lu_context_exit (struct lu_context *ctx);
1253 int lu_context_refill(struct lu_context *ctx);
1256 * Helper functions to operate on multiple keys. These are used by the default
1257 * device type operations, defined by LU_TYPE_INIT_FINI().
1260 int lu_context_key_register_many(struct lu_context_key *k, ...);
1261 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1262 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1263 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1266 * update/clear ctx/ses tags.
1268 void lu_context_tags_update(__u32 tags);
1269 void lu_context_tags_clear(__u32 tags);
1270 void lu_session_tags_update(__u32 tags);
1271 void lu_session_tags_clear(__u32 tags);
1278 * "Local" context, used to store data instead of stack.
1280 struct lu_context le_ctx;
1282 * "Session" context for per-request data.
1284 struct lu_context *le_ses;
1287 int lu_env_init (struct lu_env *env, __u32 tags);
1288 void lu_env_fini (struct lu_env *env);
1289 int lu_env_refill(struct lu_env *env);
1290 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1292 /** @} lu_context */
1295 * Output site statistical counters into a buffer. Suitable for
1296 * ll_rd_*()-style functions.
1298 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1301 * Common name structure to be passed around for various name related methods.
1304 const char *ln_name;
1309 * Validate names (path components)
1311 * To be valid \a name must be non-empty, '\0' terminated of length \a
1312 * name_len, and not contain '/'. The maximum length of a name (before
1313 * say -ENAMETOOLONG will be returned) is really controlled by llite
1314 * and the server. We only check for something insane coming from bad
1315 * integer handling here.
1317 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1319 return name != NULL &&
1321 name_len < INT_MAX &&
1322 name[name_len] == '\0' &&
1323 strlen(name) == name_len &&
1324 memchr(name, '/', name_len) == NULL;
1327 static inline bool lu_name_is_valid(const struct lu_name *ln)
1329 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1332 #define DNAME "%.*s"
1334 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1335 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1338 * Common buffer structure to be passed around for various xattr_{s,g}et()
1346 #define DLUBUF "(%p %zu)"
1347 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1349 /* read buffer params, should be filled out by out */
1351 /** number of buffers */
1352 unsigned int rb_nbufs;
1353 /** pointers to buffers */
1354 struct lu_buf rb_bufs[];
1358 * One-time initializers, called at obdclass module initialization, not
1363 * Initialization of global lu_* data.
1365 int lu_global_init(void);
1368 * Dual to lu_global_init().
1370 void lu_global_fini(void);
1372 struct lu_kmem_descr {
1373 struct kmem_cache **ckd_cache;
1374 const char *ckd_name;
1375 const size_t ckd_size;
1378 int lu_kmem_init(struct lu_kmem_descr *caches);
1379 void lu_kmem_fini(struct lu_kmem_descr *caches);
1381 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1382 const struct lu_fid *fid);
1383 struct lu_object *lu_object_anon(const struct lu_env *env,
1384 struct lu_device *dev,
1385 const struct lu_object_conf *conf);
1388 extern struct lu_buf LU_BUF_NULL;
1390 void lu_buf_free(struct lu_buf *buf);
1391 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1392 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1394 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1395 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1397 extern __u32 lu_context_tags_default;
1398 extern __u32 lu_session_tags_default;
1400 static inline bool lu_device_is_cl(const struct lu_device *d)
1402 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1405 static inline bool lu_object_is_cl(const struct lu_object *o)
1407 return lu_device_is_cl(o->lo_dev);
1411 #endif /* __LUSTRE_LU_OBJECT_H */