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, 2017, 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 <uapi/linux/lustre/lustre_idl.h>
40 #include <linux/percpu_counter.h>
41 #include <linux/ctype.h>
42 #include <obd_target.h>
45 struct proc_dir_entry;
51 * lu_* data-types represent server-side entities shared by data and meta-data
56 * -# support for layering.
58 * Server side object is split into layers, one per device in the
59 * corresponding device stack. Individual layer is represented by struct
60 * lu_object. Compound layered object --- by struct lu_object_header. Most
61 * interface functions take lu_object as an argument and operate on the
62 * whole compound object. This decision was made due to the following
65 * - it's envisaged that lu_object will be used much more often than
68 * - we want lower (non-top) layers to be able to initiate operations
69 * on the whole object.
71 * Generic code supports layering more complex than simple stacking, e.g.,
72 * it is possible that at some layer object "spawns" multiple sub-objects
75 * -# fid-based identification.
77 * Compound object is uniquely identified by its fid. Objects are indexed
78 * by their fids (hash table is used for index).
80 * -# caching and life-cycle management.
82 * Object's life-time is controlled by reference counting. When reference
83 * count drops to 0, object is returned to cache. Cached objects still
84 * retain their identity (i.e., fid), and can be recovered from cache.
86 * Objects are kept in the global LRU list, and lu_site_purge() function
87 * can be used to reclaim given number of unused objects from the tail of
90 * -# avoiding recursion.
92 * Generic code tries to replace recursion through layers by iterations
93 * where possible. Additionally to the end of reducing stack consumption,
94 * data, when practically possible, are allocated through lu_context_key
95 * interface rather than on stack.
102 struct lu_object_header;
107 * Operations common for data and meta-data devices.
109 struct lu_device_operations {
111 * Allocate object for the given device (without lower-layer
112 * parts). This is called by lu_object_operations::loo_object_init()
113 * from the parent layer, and should setup at least lu_object::lo_dev
114 * and lu_object::lo_ops fields of resulting lu_object.
116 * Object creation protocol.
118 * Due to design goal of avoiding recursion, object creation (see
119 * lu_object_alloc()) is somewhat involved:
121 * - first, lu_device_operations::ldo_object_alloc() method of the
122 * top-level device in the stack is called. It should allocate top
123 * level object (including lu_object_header), but without any
124 * lower-layer sub-object(s).
126 * - then lu_object_alloc() sets fid in the header of newly created
129 * - then lu_object_operations::loo_object_init() is called. It has
130 * to allocate lower-layer object(s). To do this,
131 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
132 * of the lower-layer device(s).
134 * - for all new objects allocated by
135 * lu_object_operations::loo_object_init() (and inserted into object
136 * stack), lu_object_operations::loo_object_init() is called again
137 * repeatedly, until no new objects are created.
139 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
140 * result->lo_ops != NULL);
142 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
143 const struct lu_object_header *h,
144 struct lu_device *d);
146 * process config specific for device.
148 int (*ldo_process_config)(const struct lu_env *env,
149 struct lu_device *, struct lustre_cfg *);
150 int (*ldo_recovery_complete)(const struct lu_env *,
154 * initialize local objects for device. this method called after layer has
155 * been initialized (after LCFG_SETUP stage) and before it starts serving
159 int (*ldo_prepare)(const struct lu_env *,
160 struct lu_device *parent,
161 struct lu_device *dev);
166 * For lu_object_conf flags
169 /* This is a new object to be allocated, or the file
170 * corresponding to the object does not exists. */
171 LOC_F_NEW = 0x00000001,
175 * Object configuration, describing particulars of object being created. On
176 * server this is not used, as server objects are full identified by fid. On
177 * client configuration contains struct lustre_md.
179 struct lu_object_conf {
181 * Some hints for obj find and alloc.
183 loc_flags_t loc_flags;
187 * Type of "printer" function used by lu_object_operations::loo_object_print()
190 * Printer function is needed to provide some flexibility in (semi-)debugging
191 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
193 typedef int (*lu_printer_t)(const struct lu_env *env,
194 void *cookie, const char *format, ...)
195 __attribute__ ((format (printf, 3, 4)));
198 * Operations specific for particular lu_object.
200 struct lu_object_operations {
203 * Allocate lower-layer parts of the object by calling
204 * lu_device_operations::ldo_object_alloc() of the corresponding
207 * This method is called once for each object inserted into object
208 * stack. It's responsibility of this method to insert lower-layer
209 * object(s) it create into appropriate places of object stack.
211 int (*loo_object_init)(const struct lu_env *env,
213 const struct lu_object_conf *conf);
215 * Called (in top-to-bottom order) during object allocation after all
216 * layers were allocated and initialized. Can be used to perform
217 * initialization depending on lower layers.
219 int (*loo_object_start)(const struct lu_env *env,
220 struct lu_object *o);
222 * Called before lu_object_operations::loo_object_free() to signal
223 * that object is being destroyed. Dual to
224 * lu_object_operations::loo_object_init().
226 void (*loo_object_delete)(const struct lu_env *env,
227 struct lu_object *o);
229 * Dual to lu_device_operations::ldo_object_alloc(). Called when
230 * object is removed from memory. Must use call_rcu or kfree_rcu
231 * if the object contains an lu_object_header.
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: context tags used by obd_*() calls.
331 * Number of existing device type instances.
333 atomic_t ldt_device_nr;
337 * Operations on a device type.
339 struct lu_device_type_operations {
341 * Allocate new device.
343 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
344 struct lu_device_type *t,
345 struct lustre_cfg *lcfg);
347 * Free device. Dual to
348 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
349 * the next device in the stack.
351 struct lu_device *(*ldto_device_free)(const struct lu_env *,
355 * Initialize the devices after allocation
357 int (*ldto_device_init)(const struct lu_env *env,
358 struct lu_device *, const char *,
361 * Finalize device. Dual to
362 * lu_device_type_operations::ldto_device_init(). Returns pointer to
363 * the next device in the stack.
365 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
368 * Initialize device type. This is called on module load.
370 int (*ldto_init)(struct lu_device_type *t);
372 * Finalize device type. Dual to
373 * lu_device_type_operations::ldto_init(). Called on module unload.
375 void (*ldto_fini)(struct lu_device_type *t);
377 * Called when the first device is created.
379 void (*ldto_start)(struct lu_device_type *t);
381 * Called when number of devices drops to 0.
383 void (*ldto_stop)(struct lu_device_type *t);
386 static inline int lu_device_is_md(const struct lu_device *d)
388 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
392 * Common object attributes.
403 /** modification time in seconds since Epoch */
405 /** access time in seconds since Epoch */
407 /** change time in seconds since Epoch */
409 /** 512-byte blocks allocated to object */
411 /** permission bits and file type */
419 /** number of persistent references to this object */
421 /** blk bits of the object*/
423 /** blk size of the object*/
429 /** set layout version to OST objects. */
430 __u32 la_layout_version;
434 * Layer in the layered object.
438 * Header for this object.
440 struct lu_object_header *lo_header;
442 * Device for this layer.
444 struct lu_device *lo_dev;
446 * Operations for this object.
448 const struct lu_object_operations *lo_ops;
450 * Linkage into list of all layers.
452 struct list_head lo_linkage;
454 * Link to the device, for debugging.
456 struct lu_ref_link lo_dev_ref;
459 enum lu_object_header_flags {
461 * Don't keep this object in cache. Object will be destroyed as soon
462 * as last reference to it is released. This flag cannot be cleared
465 LU_OBJECT_HEARD_BANSHEE = 0,
467 * Mark this object has already been taken out of cache.
469 LU_OBJECT_UNHASHED = 1,
471 * Object is initialized, when object is found in cache, it may not be
472 * intialized yet, the object allocator will initialize it.
474 LU_OBJECT_INITED = 2,
476 * Object is being purged, so mustn't be returned by
479 LU_OBJECT_PURGING = 3,
482 enum lu_object_header_attr {
483 LOHA_EXISTS = 1 << 0,
484 LOHA_REMOTE = 1 << 1,
485 LOHA_HAS_AGENT_ENTRY = 1 << 2,
487 * UNIX file type is stored in S_IFMT bits.
489 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
490 LOHA_FT_END = 017 << 12, /**< S_IFMT */
494 * "Compound" object, consisting of multiple layers.
496 * Compound object with given fid is unique with given lu_site.
498 * Note, that object does *not* necessary correspond to the real object in the
499 * persistent storage: object is an anchor for locking and method calling, so
500 * it is created for things like not-yet-existing child created by mkdir or
501 * create calls. lu_object_operations::loo_exists() can be used to check
502 * whether object is backed by persistent storage entity.
504 struct lu_object_header {
506 * Fid, uniquely identifying this object.
508 struct lu_fid loh_fid;
510 * Object flags from enum lu_object_header_flags. Set and checked
513 unsigned long loh_flags;
515 * Object reference count. Protected by lu_site::ls_guard.
519 * Common object attributes, cached for efficiency. From enum
520 * lu_object_header_attr.
524 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
526 struct hlist_node loh_hash;
528 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
530 struct list_head loh_lru;
532 * Linkage into list of layers. Never modified once set (except lately
533 * during object destruction). No locking is necessary.
535 struct list_head loh_layers;
537 * A list of references to this object, for debugging.
539 struct lu_ref loh_reference;
541 * Handle used for kfree_rcu() or similar.
543 struct rcu_head loh_rcu;
553 LU_SS_CACHE_DEATH_RACE,
559 * lu_site is a "compartment" within which objects are unique, and LRU
560 * discipline is maintained.
562 * lu_site exists so that multiple layered stacks can co-exist in the same
565 * lu_site has the same relation to lu_device as lu_object_header to
572 struct cfs_hash *ls_obj_hash;
574 * buckets for summary data
576 struct lu_site_bkt_data *ls_bkts;
580 * index of bucket on hash table while purging
582 unsigned int ls_purge_start;
584 * Top-level device for this stack.
586 struct lu_device *ls_top_dev;
588 * Bottom-level device for this stack
590 struct lu_device *ls_bottom_dev;
592 * Linkage into global list of sites.
594 struct list_head ls_linkage;
596 * List for lu device for this site, protected
599 struct list_head ls_ld_linkage;
600 spinlock_t ls_ld_lock;
602 * Lock to serialize site purge.
604 struct mutex ls_purge_mutex;
608 struct lprocfs_stats *ls_stats;
610 * XXX: a hack! fld has to find md_site via site, remove when possible
612 struct seq_server_site *ld_seq_site;
614 * Pointer to the lu_target for this site.
616 struct lu_target *ls_tgt;
619 * Number of objects in lsb_lru_lists - used for shrinking
621 struct percpu_counter ls_lru_len_counter;
625 lu_site_wq_from_fid(struct lu_site *site, struct lu_fid *fid);
627 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
629 return s->ld_seq_site;
633 * Constructors/destructors.
637 int lu_site_init (struct lu_site *s, struct lu_device *d);
638 void lu_site_fini (struct lu_site *s);
639 int lu_site_init_finish (struct lu_site *s);
640 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
641 void lu_device_get (struct lu_device *d);
642 void lu_device_put (struct lu_device *d);
643 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
644 void lu_device_fini (struct lu_device *d);
645 int lu_object_header_init(struct lu_object_header *h);
646 void lu_object_header_fini(struct lu_object_header *h);
647 int lu_object_init (struct lu_object *o,
648 struct lu_object_header *h, struct lu_device *d);
649 void lu_object_fini (struct lu_object *o);
650 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
651 void lu_object_add (struct lu_object *before, struct lu_object *o);
653 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
654 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
657 * Helpers to initialize and finalize device types.
660 int lu_device_type_init(struct lu_device_type *ldt);
661 void lu_device_type_fini(struct lu_device_type *ldt);
666 * Caching and reference counting.
671 * Acquire additional reference to the given object. This function is used to
672 * attain additional reference. To acquire initial reference use
675 static inline void lu_object_get(struct lu_object *o)
677 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
678 atomic_inc(&o->lo_header->loh_ref);
682 * Return true if object will not be cached after last reference to it is
685 static inline int lu_object_is_dying(const struct lu_object_header *h)
687 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
691 * Return true if object is initialized.
693 static inline int lu_object_is_inited(const struct lu_object_header *h)
695 return test_bit(LU_OBJECT_INITED, &h->loh_flags);
698 void lu_object_put(const struct lu_env *env, struct lu_object *o);
699 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
700 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
701 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
704 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
707 return lu_site_purge_objects(env, s, nr, 1);
710 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
711 lu_printer_t printer);
712 struct lu_object *lu_object_find(const struct lu_env *env,
713 struct lu_device *dev, const struct lu_fid *f,
714 const struct lu_object_conf *conf);
715 struct lu_object *lu_object_find_at(const struct lu_env *env,
716 struct lu_device *dev,
717 const struct lu_fid *f,
718 const struct lu_object_conf *conf);
719 struct lu_object *lu_object_find_slice(const struct lu_env *env,
720 struct lu_device *dev,
721 const struct lu_fid *f,
722 const struct lu_object_conf *conf);
731 * First (topmost) sub-object of given compound object
733 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
735 LASSERT(!list_empty(&h->loh_layers));
736 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
740 * Next sub-object in the layering
742 static inline struct lu_object *lu_object_next(const struct lu_object *o)
744 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
748 * Pointer to the fid of this object.
750 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
752 return &o->lo_header->loh_fid;
756 * return device operations vector for this object
758 static const inline struct lu_device_operations *
759 lu_object_ops(const struct lu_object *o)
761 return o->lo_dev->ld_ops;
765 * Given a compound object, find its slice, corresponding to the device type
768 struct lu_object *lu_object_locate(struct lu_object_header *h,
769 const struct lu_device_type *dtype);
772 * Printer function emitting messages through libcfs_debug_msg().
774 int lu_cdebug_printer(const struct lu_env *env,
775 void *cookie, const char *format, ...);
778 * Print object description followed by a user-supplied message.
780 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
782 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
783 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
784 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
785 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
790 * Print short object description followed by a user-supplied message.
792 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
794 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
795 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
796 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
797 (object)->lo_header); \
798 lu_cdebug_printer(env, &msgdata, "\n"); \
799 CDEBUG(mask, format , ## __VA_ARGS__); \
803 void lu_object_print (const struct lu_env *env, void *cookie,
804 lu_printer_t printer, const struct lu_object *o);
805 void lu_object_header_print(const struct lu_env *env, void *cookie,
806 lu_printer_t printer,
807 const struct lu_object_header *hdr);
810 * Check object consistency.
812 int lu_object_invariant(const struct lu_object *o);
816 * Check whether object exists, no matter on local or remote storage.
817 * Note: LOHA_EXISTS will be set once some one created the object,
818 * and it does not needs to be committed to storage.
820 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
823 * Check whether object on the remote storage.
825 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
828 * Check whether the object as agent entry on current target
830 #define lu_object_has_agent_entry(o) \
831 unlikely((o)->lo_header->loh_attr & LOHA_HAS_AGENT_ENTRY)
833 static inline void lu_object_set_agent_entry(struct lu_object *o)
835 o->lo_header->loh_attr |= LOHA_HAS_AGENT_ENTRY;
838 static inline void lu_object_clear_agent_entry(struct lu_object *o)
840 o->lo_header->loh_attr &= ~LOHA_HAS_AGENT_ENTRY;
843 static inline int lu_object_assert_exists(const struct lu_object *o)
845 return lu_object_exists(o);
848 static inline int lu_object_assert_not_exists(const struct lu_object *o)
850 return !lu_object_exists(o);
854 * Attr of this object.
856 static inline __u32 lu_object_attr(const struct lu_object *o)
858 LASSERT(lu_object_exists(o) != 0);
860 return o->lo_header->loh_attr & S_IFMT;
863 static inline void lu_object_ref_add(struct lu_object *o,
867 lu_ref_add(&o->lo_header->loh_reference, scope, source);
870 static inline void lu_object_ref_add_at(struct lu_object *o,
871 struct lu_ref_link *link,
875 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
878 static inline void lu_object_ref_del(struct lu_object *o,
879 const char *scope, const void *source)
881 lu_ref_del(&o->lo_header->loh_reference, scope, source);
884 static inline void lu_object_ref_del_at(struct lu_object *o,
885 struct lu_ref_link *link,
886 const char *scope, const void *source)
888 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
891 /** input params, should be filled out by mdt */
895 /** count in bytes */
896 unsigned int rp_count;
897 /** number of pages */
898 unsigned int rp_npages;
899 /** requested attr */
901 /** pointers to pages */
902 struct page **rp_pages;
905 enum lu_xattr_flags {
906 LU_XATTR_REPLACE = (1 << 0),
907 LU_XATTR_CREATE = (1 << 1),
908 LU_XATTR_MERGE = (1 << 2),
909 LU_XATTR_SPLIT = (1 << 3),
917 /** For lu_context health-checks */
918 enum lu_context_state {
927 * lu_context. Execution context for lu_object methods. Currently associated
930 * All lu_object methods, except device and device type methods (called during
931 * system initialization and shutdown) are executed "within" some
932 * lu_context. This means, that pointer to some "current" lu_context is passed
933 * as an argument to all methods.
935 * All service ptlrpc threads create lu_context as part of their
936 * initialization. It is possible to create "stand-alone" context for other
937 * execution environments (like system calls).
939 * lu_object methods mainly use lu_context through lu_context_key interface
940 * that allows each layer to associate arbitrary pieces of data with each
941 * context (see pthread_key_create(3) for similar interface).
943 * On a client, lu_context is bound to a thread, see cl_env_get().
945 * \see lu_context_key
949 * lu_context is used on the client side too. Yet we don't want to
950 * allocate values of server-side keys for the client contexts and
953 * To achieve this, set of tags in introduced. Contexts and keys are
954 * marked with tags. Key value are created only for context whose set
955 * of tags has non-empty intersection with one for key. Tags are taken
956 * from enum lu_context_tag.
959 enum lu_context_state lc_state;
961 * Pointer to the home service thread. NULL for other execution
964 struct ptlrpc_thread *lc_thread;
966 * Pointer to an array with key values. Internal implementation
971 * Linkage into a list of all remembered contexts. Only
972 * `non-transient' contexts, i.e., ones created for service threads
975 struct list_head lc_remember;
977 * Version counter used to skip calls to lu_context_refill() when no
978 * keys were registered.
988 * lu_context_key interface. Similar to pthread_key.
991 enum lu_context_tag {
993 * Thread on md server
995 LCT_MD_THREAD = 1 << 0,
997 * Thread on dt server
999 LCT_DT_THREAD = 1 << 1,
1003 LCT_CL_THREAD = 1 << 3,
1005 * A per-request session on a server, and a per-system-call session on
1008 LCT_SESSION = 1 << 4,
1010 * A per-request data on OSP device
1012 LCT_OSP_THREAD = 1 << 5,
1016 LCT_MG_THREAD = 1 << 6,
1018 * Context for local operations
1022 * session for server thread
1024 LCT_SERVER_SESSION = 1 << 8,
1026 * Set when at least one of keys, having values in this context has
1027 * non-NULL lu_context_key::lct_exit() method. This is used to
1028 * optimize lu_context_exit() call.
1030 LCT_HAS_EXIT = 1 << 28,
1032 * Don't add references for modules creating key values in that context.
1033 * This is only for contexts used internally by lu_object framework.
1035 LCT_NOREF = 1 << 29,
1037 * Key is being prepared for retiring, don't create new values for it.
1039 LCT_QUIESCENT = 1 << 30,
1041 * Context should be remembered.
1043 LCT_REMEMBER = 1 << 31,
1045 * Contexts usable in cache shrinker thread.
1047 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1051 * Key. Represents per-context value slot.
1053 * Keys are usually registered when module owning the key is initialized, and
1054 * de-registered when module is unloaded. Once key is registered, all new
1055 * contexts with matching tags, will get key value. "Old" contexts, already
1056 * initialized at the time of key registration, can be forced to get key value
1057 * by calling lu_context_refill().
1059 * Every key value is counted in lu_context_key::lct_used and acquires a
1060 * reference on an owning module. This means, that all key values have to be
1061 * destroyed before module can be unloaded. This is usually achieved by
1062 * stopping threads started by the module, that created contexts in their
1063 * entry functions. Situation is complicated by the threads shared by multiple
1064 * modules, like ptlrpcd daemon on a client. To work around this problem,
1065 * contexts, created in such threads, are `remembered' (see
1066 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1067 * for unloading it does the following:
1069 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1070 * preventing new key values from being allocated in the new contexts,
1073 * - scans a list of remembered contexts, destroying values of module
1074 * keys, thus releasing references to the module.
1076 * This is done by lu_context_key_quiesce(). If module is re-activated
1077 * before key has been de-registered, lu_context_key_revive() call clears
1078 * `quiescent' marker.
1080 * lu_context code doesn't provide any internal synchronization for these
1081 * activities---it's assumed that startup (including threads start-up) and
1082 * shutdown are serialized by some external means.
1086 struct lu_context_key {
1088 * Set of tags for which values of this key are to be instantiated.
1092 * Value constructor. This is called when new value is created for a
1093 * context. Returns pointer to new value of error pointer.
1095 void *(*lct_init)(const struct lu_context *ctx,
1096 struct lu_context_key *key);
1098 * Value destructor. Called when context with previously allocated
1099 * value of this slot is destroyed. \a data is a value that was returned
1100 * by a matching call to lu_context_key::lct_init().
1102 void (*lct_fini)(const struct lu_context *ctx,
1103 struct lu_context_key *key, void *data);
1105 * Optional method called on lu_context_exit() for all allocated
1106 * keys. Can be used by debugging code checking that locks are
1109 void (*lct_exit)(const struct lu_context *ctx,
1110 struct lu_context_key *key, void *data);
1112 * Internal implementation detail: index within lu_context::lc_value[]
1113 * reserved for this key.
1117 * Internal implementation detail: number of values created for this
1122 * Internal implementation detail: module for this key.
1124 struct module *lct_owner;
1126 * References to this key. For debugging.
1128 struct lu_ref lct_reference;
1131 #define LU_KEY_INIT(mod, type) \
1132 static void *mod##_key_init(const struct lu_context *ctx, \
1133 struct lu_context_key *key) \
1137 BUILD_BUG_ON(PAGE_SIZE < sizeof(*value)); \
1139 OBD_ALLOC_PTR(value); \
1140 if (value == NULL) \
1141 value = ERR_PTR(-ENOMEM); \
1145 struct __##mod##__dummy_init { ; } /* semicolon catcher */
1147 #define LU_KEY_FINI(mod, type) \
1148 static void mod##_key_fini(const struct lu_context *ctx, \
1149 struct lu_context_key *key, void* data) \
1151 type *info = data; \
1153 OBD_FREE_PTR(info); \
1155 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1157 #define LU_KEY_INIT_FINI(mod, type) \
1158 LU_KEY_INIT(mod,type); \
1159 LU_KEY_FINI(mod,type)
1161 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1162 struct lu_context_key mod##_thread_key = { \
1164 .lct_init = mod##_key_init, \
1165 .lct_fini = mod##_key_fini \
1168 #define LU_CONTEXT_KEY_INIT(key) \
1170 (key)->lct_owner = THIS_MODULE; \
1173 int lu_context_key_register(struct lu_context_key *key);
1174 void lu_context_key_degister(struct lu_context_key *key);
1175 void *lu_context_key_get (const struct lu_context *ctx,
1176 const struct lu_context_key *key);
1177 void lu_context_key_quiesce (struct lu_context_key *key);
1178 void lu_context_key_revive (struct lu_context_key *key);
1182 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1186 #define LU_KEY_INIT_GENERIC(mod) \
1187 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1189 struct lu_context_key *key = k; \
1192 va_start(args, k); \
1194 LU_CONTEXT_KEY_INIT(key); \
1195 key = va_arg(args, struct lu_context_key *); \
1196 } while (key != NULL); \
1200 #define LU_TYPE_INIT(mod, ...) \
1201 LU_KEY_INIT_GENERIC(mod) \
1202 static int mod##_type_init(struct lu_device_type *t) \
1204 mod##_key_init_generic(__VA_ARGS__, NULL); \
1205 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1207 struct __##mod##_dummy_type_init {;}
1209 #define LU_TYPE_FINI(mod, ...) \
1210 static void mod##_type_fini(struct lu_device_type *t) \
1212 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1214 struct __##mod##_dummy_type_fini {;}
1216 #define LU_TYPE_START(mod, ...) \
1217 static void mod##_type_start(struct lu_device_type *t) \
1219 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1221 struct __##mod##_dummy_type_start {;}
1223 #define LU_TYPE_STOP(mod, ...) \
1224 static void mod##_type_stop(struct lu_device_type *t) \
1226 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1228 struct __##mod##_dummy_type_stop {;}
1232 #define LU_TYPE_INIT_FINI(mod, ...) \
1233 LU_TYPE_INIT(mod, __VA_ARGS__); \
1234 LU_TYPE_FINI(mod, __VA_ARGS__); \
1235 LU_TYPE_START(mod, __VA_ARGS__); \
1236 LU_TYPE_STOP(mod, __VA_ARGS__)
1238 int lu_context_init (struct lu_context *ctx, __u32 tags);
1239 void lu_context_fini (struct lu_context *ctx);
1240 void lu_context_enter (struct lu_context *ctx);
1241 void lu_context_exit (struct lu_context *ctx);
1242 int lu_context_refill(struct lu_context *ctx);
1245 * Helper functions to operate on multiple keys. These are used by the default
1246 * device type operations, defined by LU_TYPE_INIT_FINI().
1249 int lu_context_key_register_many(struct lu_context_key *k, ...);
1250 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1251 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1252 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1255 * update/clear ctx/ses tags.
1257 void lu_context_tags_update(__u32 tags);
1258 void lu_context_tags_clear(__u32 tags);
1259 void lu_session_tags_update(__u32 tags);
1260 void lu_session_tags_clear(__u32 tags);
1267 * "Local" context, used to store data instead of stack.
1269 struct lu_context le_ctx;
1271 * "Session" context for per-request data.
1273 struct lu_context *le_ses;
1276 int lu_env_init (struct lu_env *env, __u32 tags);
1277 void lu_env_fini (struct lu_env *env);
1278 int lu_env_refill(struct lu_env *env);
1279 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1281 static inline void* lu_env_info(const struct lu_env *env,
1282 const struct lu_context_key *key)
1285 info = lu_context_key_get(&env->le_ctx, key);
1287 if (!lu_env_refill((struct lu_env *)env))
1288 info = lu_context_key_get(&env->le_ctx, key);
1294 struct lu_env *lu_env_find(void);
1295 int lu_env_add(struct lu_env *env);
1296 int lu_env_add_task(struct lu_env *env, struct task_struct *task);
1297 void lu_env_remove(struct lu_env *env);
1299 /** @} lu_context */
1302 * Output site statistical counters into a buffer. Suitable for
1303 * ll_rd_*()-style functions.
1305 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1308 * Common name structure to be passed around for various name related methods.
1311 const char *ln_name;
1315 static inline bool name_is_dot_or_dotdot(const char *name, int namelen)
1317 return name[0] == '.' &&
1318 (namelen == 1 || (namelen == 2 && name[1] == '.'));
1321 static inline bool lu_name_is_dot_or_dotdot(const struct lu_name *lname)
1323 return name_is_dot_or_dotdot(lname->ln_name, lname->ln_namelen);
1326 static inline bool lu_name_is_temp_file(const char *name, int namelen,
1327 bool dot_prefix, int suffixlen)
1332 int len = suffixlen;
1334 if (dot_prefix && name[0] != '.')
1337 if (namelen < dot_prefix + suffixlen + 2 ||
1338 name[namelen - suffixlen - 1] != '.')
1342 lower += islower(name[namelen - len]);
1343 upper += isupper(name[namelen - len]);
1344 digit += isdigit(name[namelen - len]);
1347 /* mktemp() filename suffixes will have a mix of upper- and lower-case
1348 * letters and/or numbers, not all numbers, or all upper or lower-case.
1349 * About 0.07% of randomly-generated names will slip through,
1350 * but this avoids 99.93% of cross-MDT renames for those files.
1352 if (digit >= suffixlen - 2 || upper == suffixlen || lower == suffixlen)
1358 static inline bool lu_name_is_backup_file(const char *name, int namelen,
1362 name[namelen - 2] != '.' && name[namelen - 1] == '~') {
1368 if (namelen > 4 && name[namelen - 4] == '.' &&
1369 (!strncasecmp(name + namelen - 3, "bak", 3) ||
1370 !strncasecmp(name + namelen - 3, "sav", 3))) {
1376 if (namelen > 5 && name[namelen - 5] == '.' &&
1377 !strncasecmp(name + namelen - 4, "orig", 4)) {
1386 static inline bool lu_name_is_valid_len(const char *name, size_t name_len)
1388 return name != NULL &&
1390 name_len < INT_MAX &&
1391 strlen(name) == name_len &&
1392 memchr(name, '/', name_len) == NULL;
1396 * Validate names (path components)
1398 * To be valid \a name must be non-empty, '\0' terminated of length \a
1399 * name_len, and not contain '/'. The maximum length of a name (before
1400 * say -ENAMETOOLONG will be returned) is really controlled by llite
1401 * and the server. We only check for something insane coming from bad
1402 * integer handling here.
1404 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1406 return lu_name_is_valid_len(name, name_len) && name[name_len] == '\0';
1409 static inline bool lu_name_is_valid(const struct lu_name *ln)
1411 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1414 #define DNAME "%.*s"
1416 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1417 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1420 * Common buffer structure to be passed around for various xattr_{s,g}et()
1428 #define DLUBUF "(%p %zu)"
1429 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1431 /* read buffer params, should be filled out by out */
1433 /** number of buffers */
1434 unsigned int rb_nbufs;
1435 /** pointers to buffers */
1436 struct lu_buf rb_bufs[];
1440 * One-time initializers, called at obdclass module initialization, not
1445 * Initialization of global lu_* data.
1447 int lu_global_init(void);
1450 * Dual to lu_global_init().
1452 void lu_global_fini(void);
1454 struct lu_kmem_descr {
1455 struct kmem_cache **ckd_cache;
1456 const char *ckd_name;
1457 const size_t ckd_size;
1460 int lu_kmem_init(struct lu_kmem_descr *caches);
1461 void lu_kmem_fini(struct lu_kmem_descr *caches);
1463 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1464 const struct lu_fid *fid);
1465 struct lu_object *lu_object_anon(const struct lu_env *env,
1466 struct lu_device *dev,
1467 const struct lu_object_conf *conf);
1470 extern struct lu_buf LU_BUF_NULL;
1472 void lu_buf_free(struct lu_buf *buf);
1473 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1474 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1476 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1477 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1479 extern __u32 lu_context_tags_default;
1480 extern __u32 lu_session_tags_default;
1482 static inline bool lu_device_is_cl(const struct lu_device *d)
1484 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1487 static inline bool lu_object_is_cl(const struct lu_object *o)
1489 return lu_device_is_cl(o->lo_dev);
1492 /* round-robin QoS data for LOD/LMV */
1494 spinlock_t lqr_alloc; /* protect allocation index */
1495 __u32 lqr_start_idx; /* start index of new inode */
1496 __u32 lqr_offset_idx;/* aliasing for start_idx */
1497 int lqr_start_count;/* reseed counter */
1498 struct lu_tgt_pool lqr_pool; /* round-robin optimized list */
1499 unsigned long lqr_dirty:1; /* recalc round-robin list */
1502 /* QoS data per MDS/OSS */
1504 struct obd_uuid lsq_uuid; /* ptlrpc's c_remote_uuid */
1505 struct list_head lsq_svr_list; /* link to lq_svr_list */
1506 __u64 lsq_bavail; /* total bytes avail on svr */
1507 __u64 lsq_iavail; /* tital inode avail on svr */
1508 __u64 lsq_penalty; /* current penalty */
1509 __u64 lsq_penalty_per_obj; /* penalty decrease
1511 time64_t lsq_used; /* last used time, seconds */
1512 __u32 lsq_tgt_count; /* number of tgts on this svr */
1513 __u32 lsq_id; /* unique svr id */
1516 /* QoS data per MDT/OST */
1518 struct lu_svr_qos *ltq_svr; /* svr info */
1519 __u64 ltq_penalty; /* current penalty */
1520 __u64 ltq_penalty_per_obj; /* penalty decrease
1522 __u64 ltq_weight; /* net weighting */
1523 time64_t ltq_used; /* last used time, seconds */
1524 bool ltq_usable:1; /* usable for striping */
1527 /* target descriptor */
1528 struct lu_tgt_desc {
1530 struct dt_device *ltd_tgt;
1531 struct obd_device *ltd_obd;
1533 struct obd_export *ltd_exp;
1534 struct obd_uuid ltd_uuid;
1537 struct list_head ltd_kill;
1538 struct task_struct *ltd_recovery_task;
1539 struct mutex ltd_fid_mutex;
1540 struct lu_tgt_qos ltd_qos; /* qos info per target */
1541 struct obd_statfs ltd_statfs;
1542 time64_t ltd_statfs_age;
1543 unsigned long ltd_active:1,/* is this target up for requests */
1544 ltd_activate:1,/* should target be activated */
1545 ltd_reap:1, /* should this target be deleted */
1546 ltd_got_update_log:1, /* Already got update log */
1547 ltd_connecting:1; /* target is connecting */
1550 /* number of pointers at 1st level */
1551 #define TGT_PTRS (PAGE_SIZE / sizeof(void *))
1552 /* number of pointers at 2nd level */
1553 #define TGT_PTRS_PER_BLOCK (PAGE_SIZE / sizeof(void *))
1555 struct lu_tgt_desc_idx {
1556 struct lu_tgt_desc *ldi_tgt[TGT_PTRS_PER_BLOCK];
1559 /* QoS data for LOD/LMV */
1561 struct list_head lq_svr_list; /* lu_svr_qos list */
1562 struct rw_semaphore lq_rw_sem;
1563 __u32 lq_active_svr_count;
1564 unsigned int lq_prio_free; /* priority for free space */
1565 unsigned int lq_threshold_rr;/* priority for rr */
1566 struct lu_qos_rr lq_rr; /* round robin qos data */
1567 unsigned long lq_dirty:1, /* recalc qos data */
1568 lq_same_space:1,/* the servers all have approx.
1569 * the same space avail */
1570 lq_reset:1; /* zero current penalties */
1573 struct lu_tgt_descs {
1575 struct lov_desc ltd_lov_desc;
1576 struct lmv_desc ltd_lmv_desc;
1578 /* list of known TGTs */
1579 struct lu_tgt_desc_idx *ltd_tgt_idx[TGT_PTRS];
1580 /* Size of the lu_tgts array, granted to be a power of 2 */
1581 __u32 ltd_tgts_size;
1582 /* bitmap of TGTs available */
1583 struct cfs_bitmap *ltd_tgt_bitmap;
1584 /* TGTs scheduled to be deleted */
1585 __u32 ltd_death_row;
1586 /* Table refcount used for delayed deletion */
1588 /* mutex to serialize concurrent updates to the tgt table */
1589 struct mutex ltd_mutex;
1590 /* read/write semaphore used for array relocation */
1591 struct rw_semaphore ltd_rw_sem;
1593 struct lu_qos ltd_qos;
1594 /* all tgts in a packed array */
1595 struct lu_tgt_pool ltd_tgt_pool;
1596 /* true if tgt is MDT */
1600 #define LTD_TGT(ltd, index) \
1601 (ltd)->ltd_tgt_idx[(index) / \
1602 TGT_PTRS_PER_BLOCK]->ldi_tgt[(index) % TGT_PTRS_PER_BLOCK]
1604 u64 lu_prandom_u64_max(u64 ep_ro);
1605 void lu_qos_rr_init(struct lu_qos_rr *lqr);
1606 int lu_qos_add_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1607 void lu_tgt_qos_weight_calc(struct lu_tgt_desc *tgt);
1609 int lu_tgt_descs_init(struct lu_tgt_descs *ltd, bool is_mdt);
1610 void lu_tgt_descs_fini(struct lu_tgt_descs *ltd);
1611 int ltd_add_tgt(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt);
1612 void ltd_del_tgt(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt);
1613 bool ltd_qos_is_usable(struct lu_tgt_descs *ltd);
1614 int ltd_qos_penalties_calc(struct lu_tgt_descs *ltd);
1615 int ltd_qos_update(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt,
1618 static inline struct lu_tgt_desc *ltd_first_tgt(struct lu_tgt_descs *ltd)
1622 index = find_first_bit(ltd->ltd_tgt_bitmap->data,
1623 ltd->ltd_tgt_bitmap->size);
1624 return (index < ltd->ltd_tgt_bitmap->size) ? LTD_TGT(ltd, index) : NULL;
1627 static inline struct lu_tgt_desc *ltd_next_tgt(struct lu_tgt_descs *ltd,
1628 struct lu_tgt_desc *tgt)
1635 index = tgt->ltd_index;
1636 LASSERT(index < ltd->ltd_tgt_bitmap->size);
1637 index = find_next_bit(ltd->ltd_tgt_bitmap->data,
1638 ltd->ltd_tgt_bitmap->size, index + 1);
1639 return (index < ltd->ltd_tgt_bitmap->size) ? LTD_TGT(ltd, index) : NULL;
1642 #define ltd_foreach_tgt(ltd, tgt) \
1643 for (tgt = ltd_first_tgt(ltd); tgt; tgt = ltd_next_tgt(ltd, tgt))
1645 #define ltd_foreach_tgt_safe(ltd, tgt, tmp) \
1646 for (tgt = ltd_first_tgt(ltd), tmp = ltd_next_tgt(ltd, tgt); tgt; \
1647 tgt = tmp, tmp = ltd_next_tgt(ltd, tgt))
1650 #endif /* __LUSTRE_LU_OBJECT_H */