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>
44 struct proc_dir_entry;
50 * lu_* data-types represent server-side entities shared by data and meta-data
55 * -# support for layering.
57 * Server side object is split into layers, one per device in the
58 * corresponding device stack. Individual layer is represented by struct
59 * lu_object. Compound layered object --- by struct lu_object_header. Most
60 * interface functions take lu_object as an argument and operate on the
61 * whole compound object. This decision was made due to the following
64 * - it's envisaged that lu_object will be used much more often than
67 * - we want lower (non-top) layers to be able to initiate operations
68 * on the whole object.
70 * Generic code supports layering more complex than simple stacking, e.g.,
71 * it is possible that at some layer object "spawns" multiple sub-objects
74 * -# fid-based identification.
76 * Compound object is uniquely identified by its fid. Objects are indexed
77 * by their fids (hash table is used for index).
79 * -# caching and life-cycle management.
81 * Object's life-time is controlled by reference counting. When reference
82 * count drops to 0, object is returned to cache. Cached objects still
83 * retain their identity (i.e., fid), and can be recovered from cache.
85 * Objects are kept in the global LRU list, and lu_site_purge() function
86 * can be used to reclaim given number of unused objects from the tail of
89 * -# avoiding recursion.
91 * Generic code tries to replace recursion through layers by iterations
92 * where possible. Additionally to the end of reducing stack consumption,
93 * data, when practically possible, are allocated through lu_context_key
94 * interface rather than on stack.
101 struct lu_object_header;
106 * Operations common for data and meta-data devices.
108 struct lu_device_operations {
110 * Allocate object for the given device (without lower-layer
111 * parts). This is called by lu_object_operations::loo_object_init()
112 * from the parent layer, and should setup at least lu_object::lo_dev
113 * and lu_object::lo_ops fields of resulting lu_object.
115 * Object creation protocol.
117 * Due to design goal of avoiding recursion, object creation (see
118 * lu_object_alloc()) is somewhat involved:
120 * - first, lu_device_operations::ldo_object_alloc() method of the
121 * top-level device in the stack is called. It should allocate top
122 * level object (including lu_object_header), but without any
123 * lower-layer sub-object(s).
125 * - then lu_object_alloc() sets fid in the header of newly created
128 * - then lu_object_operations::loo_object_init() is called. It has
129 * to allocate lower-layer object(s). To do this,
130 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
131 * of the lower-layer device(s).
133 * - for all new objects allocated by
134 * lu_object_operations::loo_object_init() (and inserted into object
135 * stack), lu_object_operations::loo_object_init() is called again
136 * repeatedly, until no new objects are created.
138 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
139 * result->lo_ops != NULL);
141 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
142 const struct lu_object_header *h,
143 struct lu_device *d);
145 * process config specific for device.
147 int (*ldo_process_config)(const struct lu_env *env,
148 struct lu_device *, struct lustre_cfg *);
149 int (*ldo_recovery_complete)(const struct lu_env *,
153 * initialize local objects for device. this method called after layer has
154 * been initialized (after LCFG_SETUP stage) and before it starts serving
158 int (*ldo_prepare)(const struct lu_env *,
159 struct lu_device *parent,
160 struct lu_device *dev);
165 * For lu_object_conf flags
168 /* This is a new object to be allocated, or the file
169 * corresponding to the object does not exists. */
170 LOC_F_NEW = 0x00000001,
174 * Object configuration, describing particulars of object being created. On
175 * server this is not used, as server objects are full identified by fid. On
176 * client configuration contains struct lustre_md.
178 struct lu_object_conf {
180 * Some hints for obj find and alloc.
182 loc_flags_t loc_flags;
186 * Type of "printer" function used by lu_object_operations::loo_object_print()
189 * Printer function is needed to provide some flexibility in (semi-)debugging
190 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
192 typedef int (*lu_printer_t)(const struct lu_env *env,
193 void *cookie, const char *format, ...)
194 __attribute__ ((format (printf, 3, 4)));
197 * Operations specific for particular lu_object.
199 struct lu_object_operations {
202 * Allocate lower-layer parts of the object by calling
203 * lu_device_operations::ldo_object_alloc() of the corresponding
206 * This method is called once for each object inserted into object
207 * stack. It's responsibility of this method to insert lower-layer
208 * object(s) it create into appropriate places of object stack.
210 int (*loo_object_init)(const struct lu_env *env,
212 const struct lu_object_conf *conf);
214 * Called (in top-to-bottom order) during object allocation after all
215 * layers were allocated and initialized. Can be used to perform
216 * initialization depending on lower layers.
218 int (*loo_object_start)(const struct lu_env *env,
219 struct lu_object *o);
221 * Called before lu_object_operations::loo_object_free() to signal
222 * that object is being destroyed. Dual to
223 * lu_object_operations::loo_object_init().
225 void (*loo_object_delete)(const struct lu_env *env,
226 struct lu_object *o);
228 * Dual to lu_device_operations::ldo_object_alloc(). Called when
229 * object is removed from memory.
231 void (*loo_object_free)(const struct lu_env *env,
232 struct lu_object *o);
234 * Called when last active reference to the object is released (and
235 * object returns to the cache). This method is optional.
237 void (*loo_object_release)(const struct lu_env *env,
238 struct lu_object *o);
240 * Optional debugging helper. Print given object.
242 int (*loo_object_print)(const struct lu_env *env, void *cookie,
243 lu_printer_t p, const struct lu_object *o);
245 * Optional debugging method. Returns true iff method is internally
248 int (*loo_object_invariant)(const struct lu_object *o);
254 struct lu_device_type;
257 * Device: a layer in the server side abstraction stacking.
261 * reference count. This is incremented, in particular, on each object
262 * created at this layer.
264 * \todo XXX which means that atomic_t is probably too small.
268 * Pointer to device type. Never modified once set.
270 struct lu_device_type *ld_type;
272 * Operation vector for this device.
274 const struct lu_device_operations *ld_ops;
276 * Stack this device belongs to.
278 struct lu_site *ld_site;
279 struct proc_dir_entry *ld_proc_entry;
281 /** \todo XXX: temporary back pointer into obd. */
282 struct obd_device *ld_obd;
284 * A list of references to this object, for debugging.
286 struct lu_ref ld_reference;
288 * Link the device to the site.
290 struct list_head ld_linkage;
293 struct lu_device_type_operations;
296 * Tag bits for device type. They are used to distinguish certain groups of
300 /** this is meta-data device */
301 LU_DEVICE_MD = (1 << 0),
302 /** this is data device */
303 LU_DEVICE_DT = (1 << 1),
304 /** data device in the client stack */
305 LU_DEVICE_CL = (1 << 2)
311 struct lu_device_type {
313 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
317 * Name of this class. Unique system-wide. Never modified once set.
321 * Operations for this type.
323 const struct lu_device_type_operations *ldt_ops;
325 * \todo XXX: temporary: context tags used by obd_*() calls.
329 * Number of existing device type instances.
331 atomic_t ldt_device_nr;
335 * Operations on a device type.
337 struct lu_device_type_operations {
339 * Allocate new device.
341 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
342 struct lu_device_type *t,
343 struct lustre_cfg *lcfg);
345 * Free device. Dual to
346 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
347 * the next device in the stack.
349 struct lu_device *(*ldto_device_free)(const struct lu_env *,
353 * Initialize the devices after allocation
355 int (*ldto_device_init)(const struct lu_env *env,
356 struct lu_device *, const char *,
359 * Finalize device. Dual to
360 * lu_device_type_operations::ldto_device_init(). Returns pointer to
361 * the next device in the stack.
363 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
366 * Initialize device type. This is called on module load.
368 int (*ldto_init)(struct lu_device_type *t);
370 * Finalize device type. Dual to
371 * lu_device_type_operations::ldto_init(). Called on module unload.
373 void (*ldto_fini)(struct lu_device_type *t);
375 * Called when the first device is created.
377 void (*ldto_start)(struct lu_device_type *t);
379 * Called when number of devices drops to 0.
381 void (*ldto_stop)(struct lu_device_type *t);
384 static inline int lu_device_is_md(const struct lu_device *d)
386 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
390 * 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*/
427 /** set layout version to OST objects. */
428 __u32 la_layout_version;
432 * Layer in the layered object.
436 * Header for this object.
438 struct lu_object_header *lo_header;
440 * Device for this layer.
442 struct lu_device *lo_dev;
444 * Operations for this object.
446 const struct lu_object_operations *lo_ops;
448 * Linkage into list of all layers.
450 struct list_head lo_linkage;
452 * Link to the device, for debugging.
454 struct lu_ref_link lo_dev_ref;
457 enum lu_object_header_flags {
459 * Don't keep this object in cache. Object will be destroyed as soon
460 * as last reference to it is released. This flag cannot be cleared
463 LU_OBJECT_HEARD_BANSHEE = 0,
465 * Mark this object has already been taken out of cache.
467 LU_OBJECT_UNHASHED = 1,
469 * Object is initialized, when object is found in cache, it may not be
470 * intialized yet, the object allocator will initialize it.
472 LU_OBJECT_INITED = 2,
474 * Object is being purged, so mustn't be returned by
477 LU_OBJECT_PURGING = 3,
480 enum lu_object_header_attr {
481 LOHA_EXISTS = 1 << 0,
482 LOHA_REMOTE = 1 << 1,
483 LOHA_HAS_AGENT_ENTRY = 1 << 2,
485 * UNIX file type is stored in S_IFMT bits.
487 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
488 LOHA_FT_END = 017 << 12, /**< S_IFMT */
492 * "Compound" object, consisting of multiple layers.
494 * Compound object with given fid is unique with given lu_site.
496 * Note, that object does *not* necessary correspond to the real object in the
497 * persistent storage: object is an anchor for locking and method calling, so
498 * it is created for things like not-yet-existing child created by mkdir or
499 * create calls. lu_object_operations::loo_exists() can be used to check
500 * whether object is backed by persistent storage entity.
502 struct lu_object_header {
504 * Fid, uniquely identifying this object.
506 struct lu_fid loh_fid;
508 * Object flags from enum lu_object_header_flags. Set and checked
511 unsigned long loh_flags;
513 * Object reference count. Protected by lu_site::ls_guard.
517 * Common object attributes, cached for efficiency. From enum
518 * lu_object_header_attr.
522 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
524 struct hlist_node loh_hash;
526 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
528 struct list_head loh_lru;
530 * Linkage into list of layers. Never modified once set (except lately
531 * during object destruction). No locking is necessary.
533 struct list_head loh_layers;
535 * A list of references to this object, for debugging.
537 struct lu_ref loh_reference;
547 LU_SS_CACHE_DEATH_RACE,
553 * lu_site is a "compartment" within which objects are unique, and LRU
554 * discipline is maintained.
556 * lu_site exists so that multiple layered stacks can co-exist in the same
559 * lu_site has the same relation to lu_device as lu_object_header to
566 struct cfs_hash *ls_obj_hash;
568 * buckets for summary data
570 struct lu_site_bkt_data *ls_bkts;
574 * index of bucket on hash table while purging
576 unsigned int ls_purge_start;
578 * Top-level device for this stack.
580 struct lu_device *ls_top_dev;
582 * Bottom-level device for this stack
584 struct lu_device *ls_bottom_dev;
586 * Linkage into global list of sites.
588 struct list_head ls_linkage;
590 * List for lu device for this site, protected
593 struct list_head ls_ld_linkage;
594 spinlock_t ls_ld_lock;
596 * Lock to serialize site purge.
598 struct mutex ls_purge_mutex;
602 struct lprocfs_stats *ls_stats;
604 * XXX: a hack! fld has to find md_site via site, remove when possible
606 struct seq_server_site *ld_seq_site;
608 * Pointer to the lu_target for this site.
610 struct lu_target *ls_tgt;
613 * Number of objects in lsb_lru_lists - used for shrinking
615 struct percpu_counter ls_lru_len_counter;
619 lu_site_wq_from_fid(struct lu_site *site, struct lu_fid *fid);
621 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
623 return s->ld_seq_site;
627 * Constructors/destructors.
631 int lu_site_init (struct lu_site *s, struct lu_device *d);
632 void lu_site_fini (struct lu_site *s);
633 int lu_site_init_finish (struct lu_site *s);
634 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
635 void lu_device_get (struct lu_device *d);
636 void lu_device_put (struct lu_device *d);
637 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
638 void lu_device_fini (struct lu_device *d);
639 int lu_object_header_init(struct lu_object_header *h);
640 void lu_object_header_fini(struct lu_object_header *h);
641 int lu_object_init (struct lu_object *o,
642 struct lu_object_header *h, struct lu_device *d);
643 void lu_object_fini (struct lu_object *o);
644 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
645 void lu_object_add (struct lu_object *before, struct lu_object *o);
647 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
648 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
651 * Helpers to initialize and finalize device types.
654 int lu_device_type_init(struct lu_device_type *ldt);
655 void lu_device_type_fini(struct lu_device_type *ldt);
660 * Caching and reference counting.
665 * Acquire additional reference to the given object. This function is used to
666 * attain additional reference. To acquire initial reference use
669 static inline void lu_object_get(struct lu_object *o)
671 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
672 atomic_inc(&o->lo_header->loh_ref);
676 * Return true if object will not be cached after last reference to it is
679 static inline int lu_object_is_dying(const struct lu_object_header *h)
681 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
685 * Return true if object is initialized.
687 static inline int lu_object_is_inited(const struct lu_object_header *h)
689 return test_bit(LU_OBJECT_INITED, &h->loh_flags);
692 void lu_object_put(const struct lu_env *env, struct lu_object *o);
693 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
694 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
695 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
698 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
701 return lu_site_purge_objects(env, s, nr, 1);
704 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
705 lu_printer_t printer);
706 struct lu_object *lu_object_find(const struct lu_env *env,
707 struct lu_device *dev, const struct lu_fid *f,
708 const struct lu_object_conf *conf);
709 struct lu_object *lu_object_find_at(const struct lu_env *env,
710 struct lu_device *dev,
711 const struct lu_fid *f,
712 const struct lu_object_conf *conf);
713 struct lu_object *lu_object_find_slice(const struct lu_env *env,
714 struct lu_device *dev,
715 const struct lu_fid *f,
716 const struct lu_object_conf *conf);
725 * First (topmost) sub-object of given compound object
727 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
729 LASSERT(!list_empty(&h->loh_layers));
730 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
734 * Next sub-object in the layering
736 static inline struct lu_object *lu_object_next(const struct lu_object *o)
738 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
742 * Pointer to the fid of this object.
744 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
746 return &o->lo_header->loh_fid;
750 * return device operations vector for this object
752 static const inline struct lu_device_operations *
753 lu_object_ops(const struct lu_object *o)
755 return o->lo_dev->ld_ops;
759 * Given a compound object, find its slice, corresponding to the device type
762 struct lu_object *lu_object_locate(struct lu_object_header *h,
763 const struct lu_device_type *dtype);
766 * Printer function emitting messages through libcfs_debug_msg().
768 int lu_cdebug_printer(const struct lu_env *env,
769 void *cookie, const char *format, ...);
772 * Print object description followed by a user-supplied message.
774 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
776 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
777 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
778 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
779 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
784 * Print short object description followed by a user-supplied message.
786 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
788 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
789 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
790 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
791 (object)->lo_header); \
792 lu_cdebug_printer(env, &msgdata, "\n"); \
793 CDEBUG(mask, format , ## __VA_ARGS__); \
797 void lu_object_print (const struct lu_env *env, void *cookie,
798 lu_printer_t printer, const struct lu_object *o);
799 void lu_object_header_print(const struct lu_env *env, void *cookie,
800 lu_printer_t printer,
801 const struct lu_object_header *hdr);
804 * Check object consistency.
806 int lu_object_invariant(const struct lu_object *o);
810 * Check whether object exists, no matter on local or remote storage.
811 * Note: LOHA_EXISTS will be set once some one created the object,
812 * and it does not needs to be committed to storage.
814 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
817 * Check whether object on the remote storage.
819 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
822 * Check whether the object as agent entry on current target
824 #define lu_object_has_agent_entry(o) \
825 unlikely((o)->lo_header->loh_attr & LOHA_HAS_AGENT_ENTRY)
827 static inline void lu_object_set_agent_entry(struct lu_object *o)
829 o->lo_header->loh_attr |= LOHA_HAS_AGENT_ENTRY;
832 static inline void lu_object_clear_agent_entry(struct lu_object *o)
834 o->lo_header->loh_attr &= ~LOHA_HAS_AGENT_ENTRY;
837 static inline int lu_object_assert_exists(const struct lu_object *o)
839 return lu_object_exists(o);
842 static inline int lu_object_assert_not_exists(const struct lu_object *o)
844 return !lu_object_exists(o);
848 * Attr of this object.
850 static inline __u32 lu_object_attr(const struct lu_object *o)
852 LASSERT(lu_object_exists(o) != 0);
854 return o->lo_header->loh_attr & S_IFMT;
857 static inline void lu_object_ref_add(struct lu_object *o,
861 lu_ref_add(&o->lo_header->loh_reference, scope, source);
864 static inline void lu_object_ref_add_at(struct lu_object *o,
865 struct lu_ref_link *link,
869 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
872 static inline void lu_object_ref_del(struct lu_object *o,
873 const char *scope, const void *source)
875 lu_ref_del(&o->lo_header->loh_reference, scope, source);
878 static inline void lu_object_ref_del_at(struct lu_object *o,
879 struct lu_ref_link *link,
880 const char *scope, const void *source)
882 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
885 /** input params, should be filled out by mdt */
889 /** count in bytes */
890 unsigned int rp_count;
891 /** number of pages */
892 unsigned int rp_npages;
893 /** requested attr */
895 /** pointers to pages */
896 struct page **rp_pages;
899 enum lu_xattr_flags {
900 LU_XATTR_REPLACE = (1 << 0),
901 LU_XATTR_CREATE = (1 << 1),
902 LU_XATTR_MERGE = (1 << 2),
903 LU_XATTR_SPLIT = (1 << 3),
911 /** For lu_context health-checks */
912 enum lu_context_state {
921 * lu_context. Execution context for lu_object methods. Currently associated
924 * All lu_object methods, except device and device type methods (called during
925 * system initialization and shutdown) are executed "within" some
926 * lu_context. This means, that pointer to some "current" lu_context is passed
927 * as an argument to all methods.
929 * All service ptlrpc threads create lu_context as part of their
930 * initialization. It is possible to create "stand-alone" context for other
931 * execution environments (like system calls).
933 * lu_object methods mainly use lu_context through lu_context_key interface
934 * that allows each layer to associate arbitrary pieces of data with each
935 * context (see pthread_key_create(3) for similar interface).
937 * On a client, lu_context is bound to a thread, see cl_env_get().
939 * \see lu_context_key
943 * lu_context is used on the client side too. Yet we don't want to
944 * allocate values of server-side keys for the client contexts and
947 * To achieve this, set of tags in introduced. Contexts and keys are
948 * marked with tags. Key value are created only for context whose set
949 * of tags has non-empty intersection with one for key. Tags are taken
950 * from enum lu_context_tag.
953 enum lu_context_state lc_state;
955 * Pointer to the home service thread. NULL for other execution
958 struct ptlrpc_thread *lc_thread;
960 * Pointer to an array with key values. Internal implementation
965 * Linkage into a list of all remembered contexts. Only
966 * `non-transient' contexts, i.e., ones created for service threads
969 struct list_head lc_remember;
971 * Version counter used to skip calls to lu_context_refill() when no
972 * keys were registered.
982 * lu_context_key interface. Similar to pthread_key.
985 enum lu_context_tag {
987 * Thread on md server
989 LCT_MD_THREAD = 1 << 0,
991 * Thread on dt server
993 LCT_DT_THREAD = 1 << 1,
997 LCT_CL_THREAD = 1 << 3,
999 * A per-request session on a server, and a per-system-call session on
1002 LCT_SESSION = 1 << 4,
1004 * A per-request data on OSP device
1006 LCT_OSP_THREAD = 1 << 5,
1010 LCT_MG_THREAD = 1 << 6,
1012 * Context for local operations
1016 * session for server thread
1018 LCT_SERVER_SESSION = 1 << 8,
1020 * Set when at least one of keys, having values in this context has
1021 * non-NULL lu_context_key::lct_exit() method. This is used to
1022 * optimize lu_context_exit() call.
1024 LCT_HAS_EXIT = 1 << 28,
1026 * Don't add references for modules creating key values in that context.
1027 * This is only for contexts used internally by lu_object framework.
1029 LCT_NOREF = 1 << 29,
1031 * Key is being prepared for retiring, don't create new values for it.
1033 LCT_QUIESCENT = 1 << 30,
1035 * Context should be remembered.
1037 LCT_REMEMBER = 1 << 31,
1039 * Contexts usable in cache shrinker thread.
1041 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1045 * Key. Represents per-context value slot.
1047 * Keys are usually registered when module owning the key is initialized, and
1048 * de-registered when module is unloaded. Once key is registered, all new
1049 * contexts with matching tags, will get key value. "Old" contexts, already
1050 * initialized at the time of key registration, can be forced to get key value
1051 * by calling lu_context_refill().
1053 * Every key value is counted in lu_context_key::lct_used and acquires a
1054 * reference on an owning module. This means, that all key values have to be
1055 * destroyed before module can be unloaded. This is usually achieved by
1056 * stopping threads started by the module, that created contexts in their
1057 * entry functions. Situation is complicated by the threads shared by multiple
1058 * modules, like ptlrpcd daemon on a client. To work around this problem,
1059 * contexts, created in such threads, are `remembered' (see
1060 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1061 * for unloading it does the following:
1063 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1064 * preventing new key values from being allocated in the new contexts,
1067 * - scans a list of remembered contexts, destroying values of module
1068 * keys, thus releasing references to the module.
1070 * This is done by lu_context_key_quiesce(). If module is re-activated
1071 * before key has been de-registered, lu_context_key_revive() call clears
1072 * `quiescent' marker.
1074 * lu_context code doesn't provide any internal synchronization for these
1075 * activities---it's assumed that startup (including threads start-up) and
1076 * shutdown are serialized by some external means.
1080 struct lu_context_key {
1082 * Set of tags for which values of this key are to be instantiated.
1086 * Value constructor. This is called when new value is created for a
1087 * context. Returns pointer to new value of error pointer.
1089 void *(*lct_init)(const struct lu_context *ctx,
1090 struct lu_context_key *key);
1092 * Value destructor. Called when context with previously allocated
1093 * value of this slot is destroyed. \a data is a value that was returned
1094 * by a matching call to lu_context_key::lct_init().
1096 void (*lct_fini)(const struct lu_context *ctx,
1097 struct lu_context_key *key, void *data);
1099 * Optional method called on lu_context_exit() for all allocated
1100 * keys. Can be used by debugging code checking that locks are
1103 void (*lct_exit)(const struct lu_context *ctx,
1104 struct lu_context_key *key, void *data);
1106 * Internal implementation detail: index within lu_context::lc_value[]
1107 * reserved for this key.
1111 * Internal implementation detail: number of values created for this
1116 * Internal implementation detail: module for this key.
1118 struct module *lct_owner;
1120 * References to this key. For debugging.
1122 struct lu_ref lct_reference;
1125 #define LU_KEY_INIT(mod, type) \
1126 static void *mod##_key_init(const struct lu_context *ctx, \
1127 struct lu_context_key *key) \
1131 BUILD_BUG_ON(PAGE_SIZE < sizeof(*value)); \
1133 OBD_ALLOC_PTR(value); \
1134 if (value == NULL) \
1135 value = ERR_PTR(-ENOMEM); \
1139 struct __##mod##__dummy_init { ; } /* semicolon catcher */
1141 #define LU_KEY_FINI(mod, type) \
1142 static void mod##_key_fini(const struct lu_context *ctx, \
1143 struct lu_context_key *key, void* data) \
1145 type *info = data; \
1147 OBD_FREE_PTR(info); \
1149 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1151 #define LU_KEY_INIT_FINI(mod, type) \
1152 LU_KEY_INIT(mod,type); \
1153 LU_KEY_FINI(mod,type)
1155 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1156 struct lu_context_key mod##_thread_key = { \
1158 .lct_init = mod##_key_init, \
1159 .lct_fini = mod##_key_fini \
1162 #define LU_CONTEXT_KEY_INIT(key) \
1164 (key)->lct_owner = THIS_MODULE; \
1167 int lu_context_key_register(struct lu_context_key *key);
1168 void lu_context_key_degister(struct lu_context_key *key);
1169 void *lu_context_key_get (const struct lu_context *ctx,
1170 const struct lu_context_key *key);
1171 void lu_context_key_quiesce (struct lu_context_key *key);
1172 void lu_context_key_revive (struct lu_context_key *key);
1176 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1180 #define LU_KEY_INIT_GENERIC(mod) \
1181 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1183 struct lu_context_key *key = k; \
1186 va_start(args, k); \
1188 LU_CONTEXT_KEY_INIT(key); \
1189 key = va_arg(args, struct lu_context_key *); \
1190 } while (key != NULL); \
1194 #define LU_TYPE_INIT(mod, ...) \
1195 LU_KEY_INIT_GENERIC(mod) \
1196 static int mod##_type_init(struct lu_device_type *t) \
1198 mod##_key_init_generic(__VA_ARGS__, NULL); \
1199 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1201 struct __##mod##_dummy_type_init {;}
1203 #define LU_TYPE_FINI(mod, ...) \
1204 static void mod##_type_fini(struct lu_device_type *t) \
1206 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1208 struct __##mod##_dummy_type_fini {;}
1210 #define LU_TYPE_START(mod, ...) \
1211 static void mod##_type_start(struct lu_device_type *t) \
1213 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1215 struct __##mod##_dummy_type_start {;}
1217 #define LU_TYPE_STOP(mod, ...) \
1218 static void mod##_type_stop(struct lu_device_type *t) \
1220 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1222 struct __##mod##_dummy_type_stop {;}
1226 #define LU_TYPE_INIT_FINI(mod, ...) \
1227 LU_TYPE_INIT(mod, __VA_ARGS__); \
1228 LU_TYPE_FINI(mod, __VA_ARGS__); \
1229 LU_TYPE_START(mod, __VA_ARGS__); \
1230 LU_TYPE_STOP(mod, __VA_ARGS__)
1232 int lu_context_init (struct lu_context *ctx, __u32 tags);
1233 void lu_context_fini (struct lu_context *ctx);
1234 void lu_context_enter (struct lu_context *ctx);
1235 void lu_context_exit (struct lu_context *ctx);
1236 int lu_context_refill(struct lu_context *ctx);
1239 * Helper functions to operate on multiple keys. These are used by the default
1240 * device type operations, defined by LU_TYPE_INIT_FINI().
1243 int lu_context_key_register_many(struct lu_context_key *k, ...);
1244 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1245 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1246 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1249 * update/clear ctx/ses tags.
1251 void lu_context_tags_update(__u32 tags);
1252 void lu_context_tags_clear(__u32 tags);
1253 void lu_session_tags_update(__u32 tags);
1254 void lu_session_tags_clear(__u32 tags);
1261 * "Local" context, used to store data instead of stack.
1263 struct lu_context le_ctx;
1265 * "Session" context for per-request data.
1267 struct lu_context *le_ses;
1270 int lu_env_init (struct lu_env *env, __u32 tags);
1271 void lu_env_fini (struct lu_env *env);
1272 int lu_env_refill(struct lu_env *env);
1273 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1275 static inline void* lu_env_info(const struct lu_env *env,
1276 const struct lu_context_key *key)
1279 info = lu_context_key_get(&env->le_ctx, key);
1281 if (!lu_env_refill((struct lu_env *)env))
1282 info = lu_context_key_get(&env->le_ctx, key);
1288 struct lu_env *lu_env_find(void);
1289 int lu_env_add(struct lu_env *env);
1290 void lu_env_remove(struct lu_env *env);
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;
1308 static inline bool name_is_dot_or_dotdot(const char *name, int namelen)
1310 return name[0] == '.' &&
1311 (namelen == 1 || (namelen == 2 && name[1] == '.'));
1314 static inline bool lu_name_is_dot_or_dotdot(const struct lu_name *lname)
1316 return name_is_dot_or_dotdot(lname->ln_name, lname->ln_namelen);
1319 static inline bool lu_name_is_temp_file(const char *name, int namelen,
1320 bool dot_prefix, int suffixlen)
1325 int len = suffixlen;
1327 if (dot_prefix && name[0] != '.')
1330 if (namelen < dot_prefix + suffixlen + 2 ||
1331 name[namelen - suffixlen - 1] != '.')
1335 lower += islower(name[namelen - len]);
1336 upper += isupper(name[namelen - len]);
1337 digit += isdigit(name[namelen - len]);
1340 /* mktemp() filename suffixes will have a mix of upper- and lower-case
1341 * letters and/or numbers, not all numbers, or all upper or lower-case.
1342 * About 0.07% of randomly-generated names will slip through,
1343 * but this avoids 99.93% of cross-MDT renames for those files.
1345 if (digit >= suffixlen - 2 || upper == suffixlen || lower == suffixlen)
1351 static inline bool lu_name_is_backup_file(const char *name, int namelen,
1355 name[namelen - 2] != '.' && name[namelen - 1] == '~') {
1361 if (namelen > 4 && name[namelen - 4] == '.' &&
1362 (!strncasecmp(name + namelen - 3, "bak", 3) ||
1363 !strncasecmp(name + namelen - 3, "sav", 3))) {
1369 if (namelen > 5 && name[namelen - 5] == '.' &&
1370 !strncasecmp(name + namelen - 4, "orig", 4)) {
1379 static inline bool lu_name_is_valid_len(const char *name, size_t name_len)
1381 return name != NULL &&
1383 name_len < INT_MAX &&
1384 strlen(name) == name_len &&
1385 memchr(name, '/', name_len) == NULL;
1389 * Validate names (path components)
1391 * To be valid \a name must be non-empty, '\0' terminated of length \a
1392 * name_len, and not contain '/'. The maximum length of a name (before
1393 * say -ENAMETOOLONG will be returned) is really controlled by llite
1394 * and the server. We only check for something insane coming from bad
1395 * integer handling here.
1397 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1399 return lu_name_is_valid_len(name, name_len) && name[name_len] == '\0';
1402 static inline bool lu_name_is_valid(const struct lu_name *ln)
1404 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1407 #define DNAME "%.*s"
1409 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1410 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1413 * Common buffer structure to be passed around for various xattr_{s,g}et()
1421 #define DLUBUF "(%p %zu)"
1422 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1424 /* read buffer params, should be filled out by out */
1426 /** number of buffers */
1427 unsigned int rb_nbufs;
1428 /** pointers to buffers */
1429 struct lu_buf rb_bufs[];
1433 * One-time initializers, called at obdclass module initialization, not
1438 * Initialization of global lu_* data.
1440 int lu_global_init(void);
1443 * Dual to lu_global_init().
1445 void lu_global_fini(void);
1447 struct lu_kmem_descr {
1448 struct kmem_cache **ckd_cache;
1449 const char *ckd_name;
1450 const size_t ckd_size;
1453 int lu_kmem_init(struct lu_kmem_descr *caches);
1454 void lu_kmem_fini(struct lu_kmem_descr *caches);
1456 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1457 const struct lu_fid *fid);
1458 struct lu_object *lu_object_anon(const struct lu_env *env,
1459 struct lu_device *dev,
1460 const struct lu_object_conf *conf);
1463 extern struct lu_buf LU_BUF_NULL;
1465 void lu_buf_free(struct lu_buf *buf);
1466 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1467 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1469 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1470 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1472 extern __u32 lu_context_tags_default;
1473 extern __u32 lu_session_tags_default;
1475 static inline bool lu_device_is_cl(const struct lu_device *d)
1477 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1480 static inline bool lu_object_is_cl(const struct lu_object *o)
1482 return lu_device_is_cl(o->lo_dev);
1485 /* Generic subset of tgts */
1486 struct lu_tgt_pool {
1487 __u32 *op_array; /* array of index of
1488 * lov_obd->lov_tgts */
1489 unsigned int op_count; /* number of tgts in the array */
1490 unsigned int op_size; /* allocated size of op_array */
1491 struct rw_semaphore op_rw_sem; /* to protect lu_tgt_pool use */
1494 /* round-robin QoS data for LOD/LMV */
1496 spinlock_t lqr_alloc; /* protect allocation index */
1497 __u32 lqr_start_idx; /* start index of new inode */
1498 __u32 lqr_offset_idx;/* aliasing for start_idx */
1499 int lqr_start_count;/* reseed counter */
1500 struct lu_tgt_pool lqr_pool; /* round-robin optimized list */
1501 unsigned long lqr_dirty:1; /* recalc round-robin list */
1504 /* QoS data per MDS/OSS */
1506 struct obd_uuid lsq_uuid; /* ptlrpc's c_remote_uuid */
1507 struct list_head lsq_svr_list; /* link to lq_svr_list */
1508 __u64 lsq_bavail; /* total bytes avail on svr */
1509 __u64 lsq_iavail; /* tital inode avail on svr */
1510 __u64 lsq_penalty; /* current penalty */
1511 __u64 lsq_penalty_per_obj; /* penalty decrease
1513 time64_t lsq_used; /* last used time, seconds */
1514 __u32 lsq_tgt_count; /* number of tgts on this svr */
1515 __u32 lsq_id; /* unique svr id */
1518 /* QoS data per MDT/OST */
1520 struct lu_svr_qos *ltq_svr; /* svr info */
1521 __u64 ltq_penalty; /* current penalty */
1522 __u64 ltq_penalty_per_obj; /* penalty decrease
1524 __u64 ltq_weight; /* net weighting */
1525 time64_t ltq_used; /* last used time, seconds */
1526 bool ltq_usable:1; /* usable for striping */
1529 /* target descriptor */
1530 struct lu_tgt_desc {
1532 struct dt_device *ltd_tgt;
1533 struct obd_device *ltd_obd;
1535 struct obd_export *ltd_exp;
1536 struct obd_uuid ltd_uuid;
1539 struct list_head ltd_kill;
1540 struct task_struct *ltd_recovery_task;
1541 struct mutex ltd_fid_mutex;
1542 struct lu_tgt_qos ltd_qos; /* qos info per target */
1543 struct obd_statfs ltd_statfs;
1544 time64_t ltd_statfs_age;
1545 unsigned long ltd_active:1,/* is this target up for requests */
1546 ltd_activate:1,/* should target be activated */
1547 ltd_reap:1, /* should this target be deleted */
1548 ltd_got_update_log:1, /* Already got update log */
1549 ltd_connecting:1; /* target is connecting */
1552 /* number of pointers at 1st level */
1553 #define TGT_PTRS (PAGE_SIZE / sizeof(void *))
1554 /* number of pointers at 2nd level */
1555 #define TGT_PTRS_PER_BLOCK (PAGE_SIZE / sizeof(void *))
1557 struct lu_tgt_desc_idx {
1558 struct lu_tgt_desc *ldi_tgt[TGT_PTRS_PER_BLOCK];
1561 /* QoS data for LOD/LMV */
1563 struct list_head lq_svr_list; /* lu_svr_qos list */
1564 struct rw_semaphore lq_rw_sem;
1565 __u32 lq_active_svr_count;
1566 unsigned int lq_prio_free; /* priority for free space */
1567 unsigned int lq_threshold_rr;/* priority for rr */
1568 struct lu_qos_rr lq_rr; /* round robin qos data */
1569 unsigned long lq_dirty:1, /* recalc qos data */
1570 lq_same_space:1,/* the servers all have approx.
1571 * the same space avail */
1572 lq_reset:1; /* zero current penalties */
1575 struct lu_tgt_descs {
1577 struct lov_desc ltd_lov_desc;
1578 struct lmv_desc ltd_lmv_desc;
1580 /* list of known TGTs */
1581 struct lu_tgt_desc_idx *ltd_tgt_idx[TGT_PTRS];
1582 /* Size of the lu_tgts array, granted to be a power of 2 */
1583 __u32 ltd_tgts_size;
1584 /* bitmap of TGTs available */
1585 struct cfs_bitmap *ltd_tgt_bitmap;
1586 /* TGTs scheduled to be deleted */
1587 __u32 ltd_death_row;
1588 /* Table refcount used for delayed deletion */
1590 /* mutex to serialize concurrent updates to the tgt table */
1591 struct mutex ltd_mutex;
1592 /* read/write semaphore used for array relocation */
1593 struct rw_semaphore ltd_rw_sem;
1595 struct lu_qos ltd_qos;
1596 /* all tgts in a packed array */
1597 struct lu_tgt_pool ltd_tgt_pool;
1598 /* true if tgt is MDT */
1602 #define LTD_TGT(ltd, index) \
1603 (ltd)->ltd_tgt_idx[(index) / \
1604 TGT_PTRS_PER_BLOCK]->ldi_tgt[(index) % TGT_PTRS_PER_BLOCK]
1606 u64 lu_prandom_u64_max(u64 ep_ro);
1607 void lu_qos_rr_init(struct lu_qos_rr *lqr);
1608 int lu_qos_add_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1609 void lu_tgt_qos_weight_calc(struct lu_tgt_desc *tgt);
1611 int lu_tgt_descs_init(struct lu_tgt_descs *ltd, bool is_mdt);
1612 void lu_tgt_descs_fini(struct lu_tgt_descs *ltd);
1613 int ltd_add_tgt(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt);
1614 void ltd_del_tgt(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt);
1615 bool ltd_qos_is_usable(struct lu_tgt_descs *ltd);
1616 int ltd_qos_penalties_calc(struct lu_tgt_descs *ltd);
1617 int ltd_qos_update(struct lu_tgt_descs *ltd, struct lu_tgt_desc *tgt,
1620 static inline struct lu_tgt_desc *ltd_first_tgt(struct lu_tgt_descs *ltd)
1624 index = find_first_bit(ltd->ltd_tgt_bitmap->data,
1625 ltd->ltd_tgt_bitmap->size);
1626 return (index < ltd->ltd_tgt_bitmap->size) ? LTD_TGT(ltd, index) : NULL;
1629 static inline struct lu_tgt_desc *ltd_next_tgt(struct lu_tgt_descs *ltd,
1630 struct lu_tgt_desc *tgt)
1637 index = tgt->ltd_index;
1638 LASSERT(index < ltd->ltd_tgt_bitmap->size);
1639 index = find_next_bit(ltd->ltd_tgt_bitmap->data,
1640 ltd->ltd_tgt_bitmap->size, index + 1);
1641 return (index < ltd->ltd_tgt_bitmap->size) ? LTD_TGT(ltd, index) : NULL;
1644 #define ltd_foreach_tgt(ltd, tgt) \
1645 for (tgt = ltd_first_tgt(ltd); tgt; tgt = ltd_next_tgt(ltd, tgt))
1647 #define ltd_foreach_tgt_safe(ltd, tgt, tmp) \
1648 for (tgt = ltd_first_tgt(ltd), tmp = ltd_next_tgt(ltd, tgt); tgt; \
1649 tgt = tmp, tmp = ltd_next_tgt(ltd, tgt))
1652 #endif /* __LUSTRE_LU_OBJECT_H */