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>
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,
172 * Object configuration, describing particulars of object being created. On
173 * server this is not used, as server objects are full identified by fid. On
174 * client configuration contains struct lustre_md.
176 struct lu_object_conf {
178 * Some hints for obj find and alloc.
180 loc_flags_t loc_flags;
184 * Type of "printer" function used by lu_object_operations::loo_object_print()
187 * Printer function is needed to provide some flexibility in (semi-)debugging
188 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
190 typedef int (*lu_printer_t)(const struct lu_env *env,
191 void *cookie, const char *format, ...)
192 __attribute__ ((format (printf, 3, 4)));
195 * Operations specific for particular lu_object.
197 struct lu_object_operations {
200 * Allocate lower-layer parts of the object by calling
201 * lu_device_operations::ldo_object_alloc() of the corresponding
204 * This method is called once for each object inserted into object
205 * stack. It's responsibility of this method to insert lower-layer
206 * object(s) it create into appropriate places of object stack.
208 int (*loo_object_init)(const struct lu_env *env,
210 const struct lu_object_conf *conf);
212 * Called (in top-to-bottom order) during object allocation after all
213 * layers were allocated and initialized. Can be used to perform
214 * initialization depending on lower layers.
216 int (*loo_object_start)(const struct lu_env *env,
217 struct lu_object *o);
219 * Called before lu_object_operations::loo_object_free() to signal
220 * that object is being destroyed. Dual to
221 * lu_object_operations::loo_object_init().
223 void (*loo_object_delete)(const struct lu_env *env,
224 struct lu_object *o);
226 * Dual to lu_device_operations::ldo_object_alloc(). Called when
227 * object is removed from memory.
229 void (*loo_object_free)(const struct lu_env *env,
230 struct lu_object *o);
232 * Called when last active reference to the object is released (and
233 * object returns to the cache). This method is optional.
235 void (*loo_object_release)(const struct lu_env *env,
236 struct lu_object *o);
238 * Optional debugging helper. Print given object.
240 int (*loo_object_print)(const struct lu_env *env, void *cookie,
241 lu_printer_t p, const struct lu_object *o);
243 * Optional debugging method. Returns true iff method is internally
246 int (*loo_object_invariant)(const struct lu_object *o);
252 struct lu_device_type;
255 * Device: a layer in the server side abstraction stacking.
259 * reference count. This is incremented, in particular, on each object
260 * created at this layer.
262 * \todo XXX which means that atomic_t is probably too small.
266 * Pointer to device type. Never modified once set.
268 struct lu_device_type *ld_type;
270 * Operation vector for this device.
272 const struct lu_device_operations *ld_ops;
274 * Stack this device belongs to.
276 struct lu_site *ld_site;
277 struct proc_dir_entry *ld_proc_entry;
279 /** \todo XXX: temporary back pointer into obd. */
280 struct obd_device *ld_obd;
282 * A list of references to this object, for debugging.
284 struct lu_ref ld_reference;
286 * Link the device to the site.
288 struct list_head ld_linkage;
291 struct lu_device_type_operations;
294 * Tag bits for device type. They are used to distinguish certain groups of
298 /** this is meta-data device */
299 LU_DEVICE_MD = (1 << 0),
300 /** this is data device */
301 LU_DEVICE_DT = (1 << 1),
302 /** data device in the client stack */
303 LU_DEVICE_CL = (1 << 2)
309 struct lu_device_type {
311 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
315 * Name of this class. Unique system-wide. Never modified once set.
319 * Operations for this type.
321 const struct lu_device_type_operations *ldt_ops;
323 * \todo XXX: temporary pointer to associated obd_type.
325 struct obd_type *ldt_obd_type;
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.
477 enum lu_object_header_attr {
478 LOHA_EXISTS = 1 << 0,
479 LOHA_REMOTE = 1 << 1,
480 LOHA_HAS_AGENT_ENTRY = 1 << 2,
482 * UNIX file type is stored in S_IFMT bits.
484 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
485 LOHA_FT_END = 017 << 12, /**< S_IFMT */
489 * "Compound" object, consisting of multiple layers.
491 * Compound object with given fid is unique with given lu_site.
493 * Note, that object does *not* necessary correspond to the real object in the
494 * persistent storage: object is an anchor for locking and method calling, so
495 * it is created for things like not-yet-existing child created by mkdir or
496 * create calls. lu_object_operations::loo_exists() can be used to check
497 * whether object is backed by persistent storage entity.
499 struct lu_object_header {
501 * Fid, uniquely identifying this object.
503 struct lu_fid loh_fid;
505 * Object flags from enum lu_object_header_flags. Set and checked
508 unsigned long loh_flags;
510 * Object reference count. Protected by lu_site::ls_guard.
514 * Common object attributes, cached for efficiency. From enum
515 * lu_object_header_attr.
519 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
521 struct hlist_node loh_hash;
523 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
525 struct list_head loh_lru;
527 * Linkage into list of layers. Never modified once set (except lately
528 * during object destruction). No locking is necessary.
530 struct list_head loh_layers;
532 * A list of references to this object, for debugging.
534 struct lu_ref loh_reference;
544 LU_SS_CACHE_DEATH_RACE,
550 * lu_site is a "compartment" within which objects are unique, and LRU
551 * discipline is maintained.
553 * lu_site exists so that multiple layered stacks can co-exist in the same
556 * lu_site has the same relation to lu_device as lu_object_header to
563 struct cfs_hash *ls_obj_hash;
565 * index of bucket on hash table while purging
567 unsigned int ls_purge_start;
569 * Top-level device for this stack.
571 struct lu_device *ls_top_dev;
573 * Bottom-level device for this stack
575 struct lu_device *ls_bottom_dev;
577 * Linkage into global list of sites.
579 struct list_head ls_linkage;
581 * List for lu device for this site, protected
584 struct list_head ls_ld_linkage;
585 spinlock_t ls_ld_lock;
587 * Lock to serialize site purge.
589 struct mutex ls_purge_mutex;
593 struct lprocfs_stats *ls_stats;
595 * XXX: a hack! fld has to find md_site via site, remove when possible
597 struct seq_server_site *ld_seq_site;
599 * Pointer to the lu_target for this site.
601 struct lu_target *ls_tgt;
604 * Number of objects in lsb_lru_lists - used for shrinking
606 struct percpu_counter ls_lru_len_counter;
610 lu_site_wq_from_fid(struct lu_site *site, struct lu_fid *fid);
612 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
614 return s->ld_seq_site;
618 * Constructors/destructors.
622 int lu_site_init (struct lu_site *s, struct lu_device *d);
623 void lu_site_fini (struct lu_site *s);
624 int lu_site_init_finish (struct lu_site *s);
625 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
626 void lu_device_get (struct lu_device *d);
627 void lu_device_put (struct lu_device *d);
628 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
629 void lu_device_fini (struct lu_device *d);
630 int lu_object_header_init(struct lu_object_header *h);
631 void lu_object_header_fini(struct lu_object_header *h);
632 int lu_object_init (struct lu_object *o,
633 struct lu_object_header *h, struct lu_device *d);
634 void lu_object_fini (struct lu_object *o);
635 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
636 void lu_object_add (struct lu_object *before, struct lu_object *o);
638 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
639 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
642 * Helpers to initialize and finalize device types.
645 int lu_device_type_init(struct lu_device_type *ldt);
646 void lu_device_type_fini(struct lu_device_type *ldt);
651 * Caching and reference counting.
656 * Acquire additional reference to the given object. This function is used to
657 * attain additional reference. To acquire initial reference use
660 static inline void lu_object_get(struct lu_object *o)
662 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
663 atomic_inc(&o->lo_header->loh_ref);
667 * Return true if object will not be cached after last reference to it is
670 static inline int lu_object_is_dying(const struct lu_object_header *h)
672 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
676 * Return true if object is initialized.
678 static inline int lu_object_is_inited(const struct lu_object_header *h)
680 return test_bit(LU_OBJECT_INITED, &h->loh_flags);
683 void lu_object_put(const struct lu_env *env, struct lu_object *o);
684 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
685 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
686 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
689 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
692 return lu_site_purge_objects(env, s, nr, 1);
695 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
696 lu_printer_t printer);
697 struct lu_object *lu_object_find(const struct lu_env *env,
698 struct lu_device *dev, const struct lu_fid *f,
699 const struct lu_object_conf *conf);
700 struct lu_object *lu_object_find_at(const struct lu_env *env,
701 struct lu_device *dev,
702 const struct lu_fid *f,
703 const struct lu_object_conf *conf);
704 struct lu_object *lu_object_find_slice(const struct lu_env *env,
705 struct lu_device *dev,
706 const struct lu_fid *f,
707 const struct lu_object_conf *conf);
716 * First (topmost) sub-object of given compound object
718 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
720 LASSERT(!list_empty(&h->loh_layers));
721 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
725 * Next sub-object in the layering
727 static inline struct lu_object *lu_object_next(const struct lu_object *o)
729 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
733 * Pointer to the fid of this object.
735 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
737 return &o->lo_header->loh_fid;
741 * return device operations vector for this object
743 static const inline struct lu_device_operations *
744 lu_object_ops(const struct lu_object *o)
746 return o->lo_dev->ld_ops;
750 * Given a compound object, find its slice, corresponding to the device type
753 struct lu_object *lu_object_locate(struct lu_object_header *h,
754 const struct lu_device_type *dtype);
757 * Printer function emitting messages through libcfs_debug_msg().
759 int lu_cdebug_printer(const struct lu_env *env,
760 void *cookie, const char *format, ...);
763 * Print object description followed by a user-supplied message.
765 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
767 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
768 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
769 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
770 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
775 * Print short object description followed by a user-supplied message.
777 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
779 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
780 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
781 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
782 (object)->lo_header); \
783 lu_cdebug_printer(env, &msgdata, "\n"); \
784 CDEBUG(mask, format , ## __VA_ARGS__); \
788 void lu_object_print (const struct lu_env *env, void *cookie,
789 lu_printer_t printer, const struct lu_object *o);
790 void lu_object_header_print(const struct lu_env *env, void *cookie,
791 lu_printer_t printer,
792 const struct lu_object_header *hdr);
795 * Check object consistency.
797 int lu_object_invariant(const struct lu_object *o);
801 * Check whether object exists, no matter on local or remote storage.
802 * Note: LOHA_EXISTS will be set once some one created the object,
803 * and it does not needs to be committed to storage.
805 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
808 * Check whether object on the remote storage.
810 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
813 * Check whether the object as agent entry on current target
815 #define lu_object_has_agent_entry(o) \
816 unlikely((o)->lo_header->loh_attr & LOHA_HAS_AGENT_ENTRY)
818 static inline void lu_object_set_agent_entry(struct lu_object *o)
820 o->lo_header->loh_attr |= LOHA_HAS_AGENT_ENTRY;
823 static inline void lu_object_clear_agent_entry(struct lu_object *o)
825 o->lo_header->loh_attr &= ~LOHA_HAS_AGENT_ENTRY;
828 static inline int lu_object_assert_exists(const struct lu_object *o)
830 return lu_object_exists(o);
833 static inline int lu_object_assert_not_exists(const struct lu_object *o)
835 return !lu_object_exists(o);
839 * Attr of this object.
841 static inline __u32 lu_object_attr(const struct lu_object *o)
843 LASSERT(lu_object_exists(o) != 0);
845 return o->lo_header->loh_attr & S_IFMT;
848 static inline void lu_object_ref_add(struct lu_object *o,
852 lu_ref_add(&o->lo_header->loh_reference, scope, source);
855 static inline void lu_object_ref_add_at(struct lu_object *o,
856 struct lu_ref_link *link,
860 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
863 static inline void lu_object_ref_del(struct lu_object *o,
864 const char *scope, const void *source)
866 lu_ref_del(&o->lo_header->loh_reference, scope, source);
869 static inline void lu_object_ref_del_at(struct lu_object *o,
870 struct lu_ref_link *link,
871 const char *scope, const void *source)
873 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
876 /** input params, should be filled out by mdt */
880 /** count in bytes */
881 unsigned int rp_count;
882 /** number of pages */
883 unsigned int rp_npages;
884 /** requested attr */
886 /** pointers to pages */
887 struct page **rp_pages;
890 enum lu_xattr_flags {
891 LU_XATTR_REPLACE = (1 << 0),
892 LU_XATTR_CREATE = (1 << 1),
893 LU_XATTR_MERGE = (1 << 2),
894 LU_XATTR_SPLIT = (1 << 3),
902 /** For lu_context health-checks */
903 enum lu_context_state {
912 * lu_context. Execution context for lu_object methods. Currently associated
915 * All lu_object methods, except device and device type methods (called during
916 * system initialization and shutdown) are executed "within" some
917 * lu_context. This means, that pointer to some "current" lu_context is passed
918 * as an argument to all methods.
920 * All service ptlrpc threads create lu_context as part of their
921 * initialization. It is possible to create "stand-alone" context for other
922 * execution environments (like system calls).
924 * lu_object methods mainly use lu_context through lu_context_key interface
925 * that allows each layer to associate arbitrary pieces of data with each
926 * context (see pthread_key_create(3) for similar interface).
928 * On a client, lu_context is bound to a thread, see cl_env_get().
930 * \see lu_context_key
934 * lu_context is used on the client side too. Yet we don't want to
935 * allocate values of server-side keys for the client contexts and
938 * To achieve this, set of tags in introduced. Contexts and keys are
939 * marked with tags. Key value are created only for context whose set
940 * of tags has non-empty intersection with one for key. Tags are taken
941 * from enum lu_context_tag.
944 enum lu_context_state lc_state;
946 * Pointer to the home service thread. NULL for other execution
949 struct ptlrpc_thread *lc_thread;
951 * Pointer to an array with key values. Internal implementation
956 * Linkage into a list of all remembered contexts. Only
957 * `non-transient' contexts, i.e., ones created for service threads
960 struct list_head lc_remember;
962 * Version counter used to skip calls to lu_context_refill() when no
963 * keys were registered.
973 * lu_context_key interface. Similar to pthread_key.
976 enum lu_context_tag {
978 * Thread on md server
980 LCT_MD_THREAD = 1 << 0,
982 * Thread on dt server
984 LCT_DT_THREAD = 1 << 1,
988 LCT_CL_THREAD = 1 << 3,
990 * A per-request session on a server, and a per-system-call session on
993 LCT_SESSION = 1 << 4,
995 * A per-request data on OSP device
997 LCT_OSP_THREAD = 1 << 5,
1001 LCT_MG_THREAD = 1 << 6,
1003 * Context for local operations
1007 * session for server thread
1009 LCT_SERVER_SESSION = 1 << 8,
1011 * Set when at least one of keys, having values in this context has
1012 * non-NULL lu_context_key::lct_exit() method. This is used to
1013 * optimize lu_context_exit() call.
1015 LCT_HAS_EXIT = 1 << 28,
1017 * Don't add references for modules creating key values in that context.
1018 * This is only for contexts used internally by lu_object framework.
1020 LCT_NOREF = 1 << 29,
1022 * Key is being prepared for retiring, don't create new values for it.
1024 LCT_QUIESCENT = 1 << 30,
1026 * Context should be remembered.
1028 LCT_REMEMBER = 1 << 31,
1030 * Contexts usable in cache shrinker thread.
1032 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1036 * Key. Represents per-context value slot.
1038 * Keys are usually registered when module owning the key is initialized, and
1039 * de-registered when module is unloaded. Once key is registered, all new
1040 * contexts with matching tags, will get key value. "Old" contexts, already
1041 * initialized at the time of key registration, can be forced to get key value
1042 * by calling lu_context_refill().
1044 * Every key value is counted in lu_context_key::lct_used and acquires a
1045 * reference on an owning module. This means, that all key values have to be
1046 * destroyed before module can be unloaded. This is usually achieved by
1047 * stopping threads started by the module, that created contexts in their
1048 * entry functions. Situation is complicated by the threads shared by multiple
1049 * modules, like ptlrpcd daemon on a client. To work around this problem,
1050 * contexts, created in such threads, are `remembered' (see
1051 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1052 * for unloading it does the following:
1054 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1055 * preventing new key values from being allocated in the new contexts,
1058 * - scans a list of remembered contexts, destroying values of module
1059 * keys, thus releasing references to the module.
1061 * This is done by lu_context_key_quiesce(). If module is re-activated
1062 * before key has been de-registered, lu_context_key_revive() call clears
1063 * `quiescent' marker.
1065 * lu_context code doesn't provide any internal synchronization for these
1066 * activities---it's assumed that startup (including threads start-up) and
1067 * shutdown are serialized by some external means.
1071 struct lu_context_key {
1073 * Set of tags for which values of this key are to be instantiated.
1077 * Value constructor. This is called when new value is created for a
1078 * context. Returns pointer to new value of error pointer.
1080 void *(*lct_init)(const struct lu_context *ctx,
1081 struct lu_context_key *key);
1083 * Value destructor. Called when context with previously allocated
1084 * value of this slot is destroyed. \a data is a value that was returned
1085 * by a matching call to lu_context_key::lct_init().
1087 void (*lct_fini)(const struct lu_context *ctx,
1088 struct lu_context_key *key, void *data);
1090 * Optional method called on lu_context_exit() for all allocated
1091 * keys. Can be used by debugging code checking that locks are
1094 void (*lct_exit)(const struct lu_context *ctx,
1095 struct lu_context_key *key, void *data);
1097 * Internal implementation detail: index within lu_context::lc_value[]
1098 * reserved for this key.
1102 * Internal implementation detail: number of values created for this
1107 * Internal implementation detail: module for this key.
1109 struct module *lct_owner;
1111 * References to this key. For debugging.
1113 struct lu_ref lct_reference;
1116 #define LU_KEY_INIT(mod, type) \
1117 static void *mod##_key_init(const struct lu_context *ctx, \
1118 struct lu_context_key *key) \
1122 CLASSERT(PAGE_SIZE >= sizeof(*value)); \
1124 OBD_ALLOC_PTR(value); \
1125 if (value == NULL) \
1126 value = ERR_PTR(-ENOMEM); \
1130 struct __##mod##__dummy_init { ; } /* semicolon catcher */
1132 #define LU_KEY_FINI(mod, type) \
1133 static void mod##_key_fini(const struct lu_context *ctx, \
1134 struct lu_context_key *key, void* data) \
1136 type *info = data; \
1138 OBD_FREE_PTR(info); \
1140 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1142 #define LU_KEY_INIT_FINI(mod, type) \
1143 LU_KEY_INIT(mod,type); \
1144 LU_KEY_FINI(mod,type)
1146 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1147 struct lu_context_key mod##_thread_key = { \
1149 .lct_init = mod##_key_init, \
1150 .lct_fini = mod##_key_fini \
1153 #define LU_CONTEXT_KEY_INIT(key) \
1155 (key)->lct_owner = THIS_MODULE; \
1158 int lu_context_key_register(struct lu_context_key *key);
1159 void lu_context_key_degister(struct lu_context_key *key);
1160 void *lu_context_key_get (const struct lu_context *ctx,
1161 const struct lu_context_key *key);
1162 void lu_context_key_quiesce (struct lu_context_key *key);
1163 void lu_context_key_revive (struct lu_context_key *key);
1167 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1171 #define LU_KEY_INIT_GENERIC(mod) \
1172 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1174 struct lu_context_key *key = k; \
1177 va_start(args, k); \
1179 LU_CONTEXT_KEY_INIT(key); \
1180 key = va_arg(args, struct lu_context_key *); \
1181 } while (key != NULL); \
1185 #define LU_TYPE_INIT(mod, ...) \
1186 LU_KEY_INIT_GENERIC(mod) \
1187 static int mod##_type_init(struct lu_device_type *t) \
1189 mod##_key_init_generic(__VA_ARGS__, NULL); \
1190 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1192 struct __##mod##_dummy_type_init {;}
1194 #define LU_TYPE_FINI(mod, ...) \
1195 static void mod##_type_fini(struct lu_device_type *t) \
1197 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1199 struct __##mod##_dummy_type_fini {;}
1201 #define LU_TYPE_START(mod, ...) \
1202 static void mod##_type_start(struct lu_device_type *t) \
1204 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1206 struct __##mod##_dummy_type_start {;}
1208 #define LU_TYPE_STOP(mod, ...) \
1209 static void mod##_type_stop(struct lu_device_type *t) \
1211 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1213 struct __##mod##_dummy_type_stop {;}
1217 #define LU_TYPE_INIT_FINI(mod, ...) \
1218 LU_TYPE_INIT(mod, __VA_ARGS__); \
1219 LU_TYPE_FINI(mod, __VA_ARGS__); \
1220 LU_TYPE_START(mod, __VA_ARGS__); \
1221 LU_TYPE_STOP(mod, __VA_ARGS__)
1223 int lu_context_init (struct lu_context *ctx, __u32 tags);
1224 void lu_context_fini (struct lu_context *ctx);
1225 void lu_context_enter (struct lu_context *ctx);
1226 void lu_context_exit (struct lu_context *ctx);
1227 int lu_context_refill(struct lu_context *ctx);
1230 * Helper functions to operate on multiple keys. These are used by the default
1231 * device type operations, defined by LU_TYPE_INIT_FINI().
1234 int lu_context_key_register_many(struct lu_context_key *k, ...);
1235 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1236 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1237 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1240 * update/clear ctx/ses tags.
1242 void lu_context_tags_update(__u32 tags);
1243 void lu_context_tags_clear(__u32 tags);
1244 void lu_session_tags_update(__u32 tags);
1245 void lu_session_tags_clear(__u32 tags);
1252 * "Local" context, used to store data instead of stack.
1254 struct lu_context le_ctx;
1256 * "Session" context for per-request data.
1258 struct lu_context *le_ses;
1261 int lu_env_init (struct lu_env *env, __u32 tags);
1262 void lu_env_fini (struct lu_env *env);
1263 int lu_env_refill(struct lu_env *env);
1264 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1266 struct lu_env *lu_env_find(void);
1267 int lu_env_add(struct lu_env *env);
1268 void lu_env_remove(struct lu_env *env);
1270 /** @} lu_context */
1273 * Output site statistical counters into a buffer. Suitable for
1274 * ll_rd_*()-style functions.
1276 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1279 * Common name structure to be passed around for various name related methods.
1282 const char *ln_name;
1286 static inline bool name_is_dot_or_dotdot(const char *name, int namelen)
1288 return name[0] == '.' &&
1289 (namelen == 1 || (namelen == 2 && name[1] == '.'));
1292 static inline bool lu_name_is_dot_or_dotdot(const struct lu_name *lname)
1294 return name_is_dot_or_dotdot(lname->ln_name, lname->ln_namelen);
1297 static inline bool lu_name_is_valid_len(const char *name, size_t name_len)
1299 return name != NULL &&
1301 name_len < INT_MAX &&
1302 strlen(name) == name_len &&
1303 memchr(name, '/', name_len) == NULL;
1307 * Validate names (path components)
1309 * To be valid \a name must be non-empty, '\0' terminated of length \a
1310 * name_len, and not contain '/'. The maximum length of a name (before
1311 * say -ENAMETOOLONG will be returned) is really controlled by llite
1312 * and the server. We only check for something insane coming from bad
1313 * integer handling here.
1315 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1317 return lu_name_is_valid_len(name, name_len) && name[name_len] == '\0';
1320 static inline bool lu_name_is_valid(const struct lu_name *ln)
1322 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1325 #define DNAME "%.*s"
1327 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1328 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1331 * Common buffer structure to be passed around for various xattr_{s,g}et()
1339 #define DLUBUF "(%p %zu)"
1340 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1342 /* read buffer params, should be filled out by out */
1344 /** number of buffers */
1345 unsigned int rb_nbufs;
1346 /** pointers to buffers */
1347 struct lu_buf rb_bufs[];
1351 * One-time initializers, called at obdclass module initialization, not
1356 * Initialization of global lu_* data.
1358 int lu_global_init(void);
1361 * Dual to lu_global_init().
1363 void lu_global_fini(void);
1365 struct lu_kmem_descr {
1366 struct kmem_cache **ckd_cache;
1367 const char *ckd_name;
1368 const size_t ckd_size;
1371 int lu_kmem_init(struct lu_kmem_descr *caches);
1372 void lu_kmem_fini(struct lu_kmem_descr *caches);
1374 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1375 const struct lu_fid *fid);
1376 struct lu_object *lu_object_anon(const struct lu_env *env,
1377 struct lu_device *dev,
1378 const struct lu_object_conf *conf);
1381 extern struct lu_buf LU_BUF_NULL;
1383 void lu_buf_free(struct lu_buf *buf);
1384 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1385 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1387 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1388 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1390 extern __u32 lu_context_tags_default;
1391 extern __u32 lu_session_tags_default;
1393 static inline bool lu_device_is_cl(const struct lu_device *d)
1395 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1398 static inline bool lu_object_is_cl(const struct lu_object *o)
1400 return lu_device_is_cl(o->lo_dev);
1403 /* Generic subset of OSTs */
1405 __u32 *op_array; /* array of index of
1406 * lov_obd->lov_tgts */
1407 unsigned int op_count; /* number of OSTs in the array */
1408 unsigned int op_size; /* allocated size of lp_array */
1409 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
1412 /* round-robin QoS data for LOD/LMV */
1414 spinlock_t lqr_alloc; /* protect allocation index */
1415 __u32 lqr_start_idx; /* start index of new inode */
1416 __u32 lqr_offset_idx;/* aliasing for start_idx */
1417 int lqr_start_count;/* reseed counter */
1418 struct ost_pool lqr_pool; /* round-robin optimized list */
1419 unsigned long lqr_dirty:1; /* recalc round-robin list */
1422 /* QoS data per MDS/OSS */
1424 struct obd_uuid lsq_uuid; /* ptlrpc's c_remote_uuid */
1425 struct list_head lsq_svr_list; /* link to lq_svr_list */
1426 __u64 lsq_bavail; /* total bytes avail on svr */
1427 __u64 lsq_iavail; /* tital inode avail on svr */
1428 __u64 lsq_penalty; /* current penalty */
1429 __u64 lsq_penalty_per_obj; /* penalty decrease
1431 time64_t lsq_used; /* last used time, seconds */
1432 __u32 lsq_tgt_count; /* number of tgts on this svr */
1433 __u32 lsq_id; /* unique svr id */
1436 /* QoS data per MDT/OST */
1438 struct lu_svr_qos *ltq_svr; /* svr info */
1439 __u64 ltq_penalty; /* current penalty */
1440 __u64 ltq_penalty_per_obj; /* penalty decrease
1442 __u64 ltq_weight; /* net weighting */
1443 time64_t ltq_used; /* last used time, seconds */
1444 bool ltq_usable:1; /* usable for striping */
1447 /* target descriptor */
1448 struct lu_tgt_desc {
1450 struct dt_device *ltd_tgt;
1451 struct obd_device *ltd_obd;
1453 struct obd_export *ltd_exp;
1454 struct obd_uuid ltd_uuid;
1457 struct list_head ltd_kill;
1458 struct ptlrpc_thread *ltd_recovery_thread;
1459 struct mutex ltd_fid_mutex;
1460 struct lu_tgt_qos ltd_qos; /* qos info per target */
1461 struct obd_statfs ltd_statfs;
1462 time64_t ltd_statfs_age;
1463 unsigned long ltd_active:1,/* is this target up for requests */
1464 ltd_activate:1,/* should target be activated */
1465 ltd_reap:1, /* should this target be deleted */
1466 ltd_got_update_log:1, /* Already got update log */
1467 ltd_connecting:1; /* target is connecting */
1470 /* QoS data for LOD/LMV */
1472 struct list_head lq_svr_list; /* lu_svr_qos list */
1473 struct rw_semaphore lq_rw_sem;
1474 __u32 lq_active_svr_count;
1475 unsigned int lq_prio_free; /* priority for free space */
1476 unsigned int lq_threshold_rr;/* priority for rr */
1477 struct lu_qos_rr lq_rr; /* round robin qos data */
1478 unsigned long lq_dirty:1, /* recalc qos data */
1479 lq_same_space:1,/* the servers all have approx.
1480 * the same space avail */
1481 lq_reset:1; /* zero current penalties */
1484 void lu_qos_rr_init(struct lu_qos_rr *lqr);
1485 int lqos_add_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1486 int lqos_del_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1487 u64 lu_prandom_u64_max(u64 ep_ro);
1490 #endif /* __LUSTRE_LU_OBJECT_H */