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;
341 * Operations on a device type.
343 struct lu_device_type_operations {
345 * Allocate new device.
347 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
348 struct lu_device_type *t,
349 struct lustre_cfg *lcfg);
351 * Free device. Dual to
352 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
353 * the next device in the stack.
355 struct lu_device *(*ldto_device_free)(const struct lu_env *,
359 * Initialize the devices after allocation
361 int (*ldto_device_init)(const struct lu_env *env,
362 struct lu_device *, const char *,
365 * Finalize device. Dual to
366 * lu_device_type_operations::ldto_device_init(). Returns pointer to
367 * the next device in the stack.
369 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
372 * Initialize device type. This is called on module load.
374 int (*ldto_init)(struct lu_device_type *t);
376 * Finalize device type. Dual to
377 * lu_device_type_operations::ldto_init(). Called on module unload.
379 void (*ldto_fini)(struct lu_device_type *t);
381 * Called when the first device is created.
383 void (*ldto_start)(struct lu_device_type *t);
385 * Called when number of devices drops to 0.
387 void (*ldto_stop)(struct lu_device_type *t);
390 static inline int lu_device_is_md(const struct lu_device *d)
392 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
396 * Common object attributes.
401 /** modification time in seconds since Epoch */
403 /** access time in seconds since Epoch */
405 /** change time in seconds since Epoch */
407 /** 512-byte blocks allocated to object */
409 /** permission bits and file type */
417 /** number of persistent references to this object */
419 /** blk bits of the object*/
421 /** blk size of the object*/
433 /** Bit-mask of valid attributes */
447 LA_BLKSIZE = 1 << 12,
448 LA_KILL_SUID = 1 << 13,
449 LA_KILL_SGID = 1 << 14,
453 * Layer in the layered object.
457 * Header for this object.
459 struct lu_object_header *lo_header;
461 * Device for this layer.
463 struct lu_device *lo_dev;
465 * Operations for this object.
467 const struct lu_object_operations *lo_ops;
469 * Linkage into list of all layers.
471 struct list_head lo_linkage;
473 * Link to the device, for debugging.
475 struct lu_ref_link lo_dev_ref;
478 enum lu_object_header_flags {
480 * Don't keep this object in cache. Object will be destroyed as soon
481 * as last reference to it is released. This flag cannot be cleared
484 LU_OBJECT_HEARD_BANSHEE = 0,
486 * Mark this object has already been taken out of cache.
488 LU_OBJECT_UNHASHED = 1,
491 enum lu_object_header_attr {
492 LOHA_EXISTS = 1 << 0,
493 LOHA_REMOTE = 1 << 1,
495 * UNIX file type is stored in S_IFMT bits.
497 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
498 LOHA_FT_END = 017 << 12, /**< S_IFMT */
502 * "Compound" object, consisting of multiple layers.
504 * Compound object with given fid is unique with given lu_site.
506 * Note, that object does *not* necessary correspond to the real object in the
507 * persistent storage: object is an anchor for locking and method calling, so
508 * it is created for things like not-yet-existing child created by mkdir or
509 * create calls. lu_object_operations::loo_exists() can be used to check
510 * whether object is backed by persistent storage entity.
512 struct lu_object_header {
514 * Fid, uniquely identifying this object.
516 struct lu_fid loh_fid;
518 * Object flags from enum lu_object_header_flags. Set and checked
521 unsigned long loh_flags;
523 * Object reference count. Protected by lu_site::ls_guard.
527 * Common object attributes, cached for efficiency. From enum
528 * lu_object_header_attr.
532 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
534 struct hlist_node loh_hash;
536 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
538 struct list_head loh_lru;
540 * Linkage into list of layers. Never modified once set (except lately
541 * during object destruction). No locking is necessary.
543 struct list_head loh_layers;
545 * A list of references to this object, for debugging.
547 struct lu_ref loh_reference;
552 struct lu_site_bkt_data {
554 * number of object in this bucket on the lsb_lru list.
558 * LRU list, updated on each access to object. Protected by
559 * bucket lock of lu_site::ls_obj_hash.
561 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
562 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
563 * of list_for_each_entry_safe_reverse()).
565 struct list_head lsb_lru;
567 * Wait-queue signaled when an object in this site is ultimately
568 * destroyed (lu_object_free()). It is used by lu_object_find() to
569 * wait before re-trying when object in the process of destruction is
570 * found in the hash table.
572 * \see htable_lookup().
574 wait_queue_head_t lsb_marche_funebre;
582 LU_SS_CACHE_DEATH_RACE,
588 * lu_site is a "compartment" within which objects are unique, and LRU
589 * discipline is maintained.
591 * lu_site exists so that multiple layered stacks can co-exist in the same
594 * lu_site has the same relation to lu_device as lu_object_header to
601 struct cfs_hash *ls_obj_hash;
603 * index of bucket on hash table while purging
605 unsigned int ls_purge_start;
607 * Top-level device for this stack.
609 struct lu_device *ls_top_dev;
611 * Bottom-level device for this stack
613 struct lu_device *ls_bottom_dev;
615 * Linkage into global list of sites.
617 struct list_head ls_linkage;
619 * List for lu device for this site, protected
622 struct list_head ls_ld_linkage;
623 spinlock_t ls_ld_lock;
625 * Lock to serialize site purge.
627 struct mutex ls_purge_mutex;
631 struct lprocfs_stats *ls_stats;
633 * XXX: a hack! fld has to find md_site via site, remove when possible
635 struct seq_server_site *ld_seq_site;
637 * Pointer to the lu_target for this site.
639 struct lu_target *ls_tgt;
642 * Number of objects in lsb_lru_lists - used for shrinking
644 struct percpu_counter ls_lru_len_counter;
647 static inline struct lu_site_bkt_data *
648 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
650 struct cfs_hash_bd bd;
652 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
653 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
656 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
658 return s->ld_seq_site;
662 * Constructors/destructors.
666 int lu_site_init (struct lu_site *s, struct lu_device *d);
667 void lu_site_fini (struct lu_site *s);
668 int lu_site_init_finish (struct lu_site *s);
669 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
670 void lu_device_get (struct lu_device *d);
671 void lu_device_put (struct lu_device *d);
672 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
673 void lu_device_fini (struct lu_device *d);
674 int lu_object_header_init(struct lu_object_header *h);
675 void lu_object_header_fini(struct lu_object_header *h);
676 int lu_object_init (struct lu_object *o,
677 struct lu_object_header *h, struct lu_device *d);
678 void lu_object_fini (struct lu_object *o);
679 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
680 void lu_object_add (struct lu_object *before, struct lu_object *o);
682 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
683 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
686 * Helpers to initialize and finalize device types.
689 int lu_device_type_init(struct lu_device_type *ldt);
690 void lu_device_type_fini(struct lu_device_type *ldt);
695 * Caching and reference counting.
700 * Acquire additional reference to the given object. This function is used to
701 * attain additional reference. To acquire initial reference use
704 static inline void lu_object_get(struct lu_object *o)
706 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
707 atomic_inc(&o->lo_header->loh_ref);
711 * Return true of object will not be cached after last reference to it is
714 static inline int lu_object_is_dying(const struct lu_object_header *h)
716 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
719 void lu_object_put(const struct lu_env *env, struct lu_object *o);
720 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
721 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
722 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
725 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
728 return lu_site_purge_objects(env, s, nr, 1);
731 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
732 lu_printer_t printer);
733 struct lu_object *lu_object_find(const struct lu_env *env,
734 struct lu_device *dev, const struct lu_fid *f,
735 const struct lu_object_conf *conf);
736 struct lu_object *lu_object_find_at(const struct lu_env *env,
737 struct lu_device *dev,
738 const struct lu_fid *f,
739 const struct lu_object_conf *conf);
740 struct lu_object *lu_object_find_slice(const struct lu_env *env,
741 struct lu_device *dev,
742 const struct lu_fid *f,
743 const struct lu_object_conf *conf);
752 * First (topmost) sub-object of given compound object
754 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
756 LASSERT(!list_empty(&h->loh_layers));
757 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
761 * Next sub-object in the layering
763 static inline struct lu_object *lu_object_next(const struct lu_object *o)
765 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
769 * Pointer to the fid of this object.
771 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
773 return &o->lo_header->loh_fid;
777 * return device operations vector for this object
779 static const inline struct lu_device_operations *
780 lu_object_ops(const struct lu_object *o)
782 return o->lo_dev->ld_ops;
786 * Given a compound object, find its slice, corresponding to the device type
789 struct lu_object *lu_object_locate(struct lu_object_header *h,
790 const struct lu_device_type *dtype);
793 * Printer function emitting messages through libcfs_debug_msg().
795 int lu_cdebug_printer(const struct lu_env *env,
796 void *cookie, const char *format, ...);
799 * Print object description followed by a user-supplied message.
801 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
803 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
804 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
805 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
806 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
811 * Print short object description followed by a user-supplied message.
813 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
815 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
816 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
817 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
818 (object)->lo_header); \
819 lu_cdebug_printer(env, &msgdata, "\n"); \
820 CDEBUG(mask, format , ## __VA_ARGS__); \
824 void lu_object_print (const struct lu_env *env, void *cookie,
825 lu_printer_t printer, const struct lu_object *o);
826 void lu_object_header_print(const struct lu_env *env, void *cookie,
827 lu_printer_t printer,
828 const struct lu_object_header *hdr);
831 * Check object consistency.
833 int lu_object_invariant(const struct lu_object *o);
837 * Check whether object exists, no matter on local or remote storage.
838 * Note: LOHA_EXISTS will be set once some one created the object,
839 * and it does not needs to be committed to storage.
841 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
844 * Check whether object on the remote storage.
846 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
848 static inline int lu_object_assert_exists(const struct lu_object *o)
850 return lu_object_exists(o);
853 static inline int lu_object_assert_not_exists(const struct lu_object *o)
855 return !lu_object_exists(o);
859 * Attr of this object.
861 static inline __u32 lu_object_attr(const struct lu_object *o)
863 LASSERT(lu_object_exists(o) != 0);
864 return o->lo_header->loh_attr;
867 static inline void lu_object_ref_add(struct lu_object *o,
871 lu_ref_add(&o->lo_header->loh_reference, scope, source);
874 static inline void lu_object_ref_add_at(struct lu_object *o,
875 struct lu_ref_link *link,
879 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
882 static inline void lu_object_ref_del(struct lu_object *o,
883 const char *scope, const void *source)
885 lu_ref_del(&o->lo_header->loh_reference, scope, source);
888 static inline void lu_object_ref_del_at(struct lu_object *o,
889 struct lu_ref_link *link,
890 const char *scope, const void *source)
892 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
895 /** input params, should be filled out by mdt */
899 /** count in bytes */
900 unsigned int rp_count;
901 /** number of pages */
902 unsigned int rp_npages;
903 /** requested attr */
905 /** pointers to pages */
906 struct page **rp_pages;
909 enum lu_xattr_flags {
910 LU_XATTR_REPLACE = (1 << 0),
911 LU_XATTR_CREATE = (1 << 1)
919 /** For lu_context health-checks */
920 enum lu_context_state {
928 * lu_context. Execution context for lu_object methods. Currently associated
931 * All lu_object methods, except device and device type methods (called during
932 * system initialization and shutdown) are executed "within" some
933 * lu_context. This means, that pointer to some "current" lu_context is passed
934 * as an argument to all methods.
936 * All service ptlrpc threads create lu_context as part of their
937 * initialization. It is possible to create "stand-alone" context for other
938 * execution environments (like system calls).
940 * lu_object methods mainly use lu_context through lu_context_key interface
941 * that allows each layer to associate arbitrary pieces of data with each
942 * context (see pthread_key_create(3) for similar interface).
944 * On a client, lu_context is bound to a thread, see cl_env_get().
946 * \see lu_context_key
950 * lu_context is used on the client side too. Yet we don't want to
951 * allocate values of server-side keys for the client contexts and
954 * To achieve this, set of tags in introduced. Contexts and keys are
955 * marked with tags. Key value are created only for context whose set
956 * of tags has non-empty intersection with one for key. Tags are taken
957 * from enum lu_context_tag.
960 enum lu_context_state lc_state;
962 * Pointer to the home service thread. NULL for other execution
965 struct ptlrpc_thread *lc_thread;
967 * Pointer to an array with key values. Internal implementation
972 * Linkage into a list of all remembered contexts. Only
973 * `non-transient' contexts, i.e., ones created for service threads
976 struct list_head lc_remember;
978 * Version counter used to skip calls to lu_context_refill() when no
979 * keys were registered.
989 * lu_context_key interface. Similar to pthread_key.
992 enum lu_context_tag {
994 * Thread on md server
996 LCT_MD_THREAD = 1 << 0,
998 * Thread on dt server
1000 LCT_DT_THREAD = 1 << 1,
1002 * Context for transaction handle
1004 LCT_TX_HANDLE = 1 << 2,
1008 LCT_CL_THREAD = 1 << 3,
1010 * A per-request session on a server, and a per-system-call session on
1013 LCT_SESSION = 1 << 4,
1015 * A per-request data on OSP device
1017 LCT_OSP_THREAD = 1 << 5,
1021 LCT_MG_THREAD = 1 << 6,
1023 * Context for local operations
1027 * session for server thread
1029 LCT_SERVER_SESSION = 1 << 8,
1031 * Set when at least one of keys, having values in this context has
1032 * non-NULL lu_context_key::lct_exit() method. This is used to
1033 * optimize lu_context_exit() call.
1035 LCT_HAS_EXIT = 1 << 28,
1037 * Don't add references for modules creating key values in that context.
1038 * This is only for contexts used internally by lu_object framework.
1040 LCT_NOREF = 1 << 29,
1042 * Key is being prepared for retiring, don't create new values for it.
1044 LCT_QUIESCENT = 1 << 30,
1046 * Context should be remembered.
1048 LCT_REMEMBER = 1 << 31,
1050 * Contexts usable in cache shrinker thread.
1052 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1056 * Key. Represents per-context value slot.
1058 * Keys are usually registered when module owning the key is initialized, and
1059 * de-registered when module is unloaded. Once key is registered, all new
1060 * contexts with matching tags, will get key value. "Old" contexts, already
1061 * initialized at the time of key registration, can be forced to get key value
1062 * by calling lu_context_refill().
1064 * Every key value is counted in lu_context_key::lct_used and acquires a
1065 * reference on an owning module. This means, that all key values have to be
1066 * destroyed before module can be unloaded. This is usually achieved by
1067 * stopping threads started by the module, that created contexts in their
1068 * entry functions. Situation is complicated by the threads shared by multiple
1069 * modules, like ptlrpcd daemon on a client. To work around this problem,
1070 * contexts, created in such threads, are `remembered' (see
1071 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1072 * for unloading it does the following:
1074 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1075 * preventing new key values from being allocated in the new contexts,
1078 * - scans a list of remembered contexts, destroying values of module
1079 * keys, thus releasing references to the module.
1081 * This is done by lu_context_key_quiesce(). If module is re-activated
1082 * before key has been de-registered, lu_context_key_revive() call clears
1083 * `quiescent' marker.
1085 * lu_context code doesn't provide any internal synchronization for these
1086 * activities---it's assumed that startup (including threads start-up) and
1087 * shutdown are serialized by some external means.
1091 struct lu_context_key {
1093 * Set of tags for which values of this key are to be instantiated.
1097 * Value constructor. This is called when new value is created for a
1098 * context. Returns pointer to new value of error pointer.
1100 void *(*lct_init)(const struct lu_context *ctx,
1101 struct lu_context_key *key);
1103 * Value destructor. Called when context with previously allocated
1104 * value of this slot is destroyed. \a data is a value that was returned
1105 * by a matching call to lu_context_key::lct_init().
1107 void (*lct_fini)(const struct lu_context *ctx,
1108 struct lu_context_key *key, void *data);
1110 * Optional method called on lu_context_exit() for all allocated
1111 * keys. Can be used by debugging code checking that locks are
1114 void (*lct_exit)(const struct lu_context *ctx,
1115 struct lu_context_key *key, void *data);
1117 * Internal implementation detail: index within lu_context::lc_value[]
1118 * reserved for this key.
1122 * Internal implementation detail: number of values created for this
1127 * Internal implementation detail: module for this key.
1129 struct module *lct_owner;
1131 * References to this key. For debugging.
1133 struct lu_ref lct_reference;
1136 #define LU_KEY_INIT(mod, type) \
1137 static void* mod##_key_init(const struct lu_context *ctx, \
1138 struct lu_context_key *key) \
1142 CLASSERT(PAGE_SIZE >= sizeof(*value)); \
1144 OBD_ALLOC_PTR(value); \
1145 if (value == NULL) \
1146 value = ERR_PTR(-ENOMEM); \
1150 struct __##mod##__dummy_init {;} /* semicolon catcher */
1152 #define LU_KEY_FINI(mod, type) \
1153 static void mod##_key_fini(const struct lu_context *ctx, \
1154 struct lu_context_key *key, void* data) \
1156 type *info = data; \
1158 OBD_FREE_PTR(info); \
1160 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1162 #define LU_KEY_INIT_FINI(mod, type) \
1163 LU_KEY_INIT(mod,type); \
1164 LU_KEY_FINI(mod,type)
1166 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1167 struct lu_context_key mod##_thread_key = { \
1169 .lct_init = mod##_key_init, \
1170 .lct_fini = mod##_key_fini \
1173 #define LU_CONTEXT_KEY_INIT(key) \
1175 (key)->lct_owner = THIS_MODULE; \
1178 int lu_context_key_register(struct lu_context_key *key);
1179 void lu_context_key_degister(struct lu_context_key *key);
1180 void *lu_context_key_get (const struct lu_context *ctx,
1181 const struct lu_context_key *key);
1182 void lu_context_key_quiesce (struct lu_context_key *key);
1183 void lu_context_key_revive (struct lu_context_key *key);
1187 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1191 #define LU_KEY_INIT_GENERIC(mod) \
1192 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1194 struct lu_context_key *key = k; \
1197 va_start(args, k); \
1199 LU_CONTEXT_KEY_INIT(key); \
1200 key = va_arg(args, struct lu_context_key *); \
1201 } while (key != NULL); \
1205 #define LU_TYPE_INIT(mod, ...) \
1206 LU_KEY_INIT_GENERIC(mod) \
1207 static int mod##_type_init(struct lu_device_type *t) \
1209 mod##_key_init_generic(__VA_ARGS__, NULL); \
1210 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1212 struct __##mod##_dummy_type_init {;}
1214 #define LU_TYPE_FINI(mod, ...) \
1215 static void mod##_type_fini(struct lu_device_type *t) \
1217 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1219 struct __##mod##_dummy_type_fini {;}
1221 #define LU_TYPE_START(mod, ...) \
1222 static void mod##_type_start(struct lu_device_type *t) \
1224 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1226 struct __##mod##_dummy_type_start {;}
1228 #define LU_TYPE_STOP(mod, ...) \
1229 static void mod##_type_stop(struct lu_device_type *t) \
1231 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1233 struct __##mod##_dummy_type_stop {;}
1237 #define LU_TYPE_INIT_FINI(mod, ...) \
1238 LU_TYPE_INIT(mod, __VA_ARGS__); \
1239 LU_TYPE_FINI(mod, __VA_ARGS__); \
1240 LU_TYPE_START(mod, __VA_ARGS__); \
1241 LU_TYPE_STOP(mod, __VA_ARGS__)
1243 int lu_context_init (struct lu_context *ctx, __u32 tags);
1244 void lu_context_fini (struct lu_context *ctx);
1245 void lu_context_enter (struct lu_context *ctx);
1246 void lu_context_exit (struct lu_context *ctx);
1247 int lu_context_refill(struct lu_context *ctx);
1250 * Helper functions to operate on multiple keys. These are used by the default
1251 * device type operations, defined by LU_TYPE_INIT_FINI().
1254 int lu_context_key_register_many(struct lu_context_key *k, ...);
1255 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1256 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1257 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1260 * update/clear ctx/ses tags.
1262 void lu_context_tags_update(__u32 tags);
1263 void lu_context_tags_clear(__u32 tags);
1264 void lu_session_tags_update(__u32 tags);
1265 void lu_session_tags_clear(__u32 tags);
1272 * "Local" context, used to store data instead of stack.
1274 struct lu_context le_ctx;
1276 * "Session" context for per-request data.
1278 struct lu_context *le_ses;
1281 int lu_env_init (struct lu_env *env, __u32 tags);
1282 void lu_env_fini (struct lu_env *env);
1283 int lu_env_refill(struct lu_env *env);
1284 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1286 /** @} lu_context */
1289 * Output site statistical counters into a buffer. Suitable for
1290 * ll_rd_*()-style functions.
1292 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1295 * Common name structure to be passed around for various name related methods.
1298 const char *ln_name;
1303 * Validate names (path components)
1305 * To be valid \a name must be non-empty, '\0' terminated of length \a
1306 * name_len, and not contain '/'. The maximum length of a name (before
1307 * say -ENAMETOOLONG will be returned) is really controlled by llite
1308 * and the server. We only check for something insane coming from bad
1309 * integer handling here.
1311 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1313 return name != NULL &&
1315 name_len < INT_MAX &&
1316 name[name_len] == '\0' &&
1317 strlen(name) == name_len &&
1318 memchr(name, '/', name_len) == NULL;
1321 static inline bool lu_name_is_valid(const struct lu_name *ln)
1323 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1326 #define DNAME "%.*s"
1328 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1329 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1332 * Common buffer structure to be passed around for various xattr_{s,g}et()
1340 #define DLUBUF "(%p %zu)"
1341 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1343 /* read buffer params, should be filled out by out */
1345 /** number of buffers */
1346 unsigned int rb_nbufs;
1347 /** pointers to buffers */
1348 struct lu_buf rb_bufs[];
1352 * One-time initializers, called at obdclass module initialization, not
1357 * Initialization of global lu_* data.
1359 int lu_global_init(void);
1362 * Dual to lu_global_init().
1364 void lu_global_fini(void);
1366 struct lu_kmem_descr {
1367 struct kmem_cache **ckd_cache;
1368 const char *ckd_name;
1369 const size_t ckd_size;
1372 int lu_kmem_init(struct lu_kmem_descr *caches);
1373 void lu_kmem_fini(struct lu_kmem_descr *caches);
1375 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1376 const struct lu_fid *fid);
1377 struct lu_object *lu_object_anon(const struct lu_env *env,
1378 struct lu_device *dev,
1379 const struct lu_object_conf *conf);
1382 extern struct lu_buf LU_BUF_NULL;
1384 void lu_buf_free(struct lu_buf *buf);
1385 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1386 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1388 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1389 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1391 extern __u32 lu_context_tags_default;
1392 extern __u32 lu_session_tags_default;
1394 static inline bool lu_device_is_cl(const struct lu_device *d)
1396 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1399 static inline bool lu_object_is_cl(const struct lu_object *o)
1401 return lu_device_is_cl(o->lo_dev);
1405 #endif /* __LUSTRE_LU_OBJECT_H */