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,
472 enum lu_object_header_attr {
473 LOHA_EXISTS = 1 << 0,
474 LOHA_REMOTE = 1 << 1,
475 LOHA_HAS_AGENT_ENTRY = 1 << 2,
477 * UNIX file type is stored in S_IFMT bits.
479 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
480 LOHA_FT_END = 017 << 12, /**< S_IFMT */
484 * "Compound" object, consisting of multiple layers.
486 * Compound object with given fid is unique with given lu_site.
488 * Note, that object does *not* necessary correspond to the real object in the
489 * persistent storage: object is an anchor for locking and method calling, so
490 * it is created for things like not-yet-existing child created by mkdir or
491 * create calls. lu_object_operations::loo_exists() can be used to check
492 * whether object is backed by persistent storage entity.
494 struct lu_object_header {
496 * Fid, uniquely identifying this object.
498 struct lu_fid loh_fid;
500 * Object flags from enum lu_object_header_flags. Set and checked
503 unsigned long loh_flags;
505 * Object reference count. Protected by lu_site::ls_guard.
509 * Common object attributes, cached for efficiency. From enum
510 * lu_object_header_attr.
514 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
516 struct hlist_node loh_hash;
518 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
520 struct list_head loh_lru;
522 * Linkage into list of layers. Never modified once set (except lately
523 * during object destruction). No locking is necessary.
525 struct list_head loh_layers;
527 * A list of references to this object, for debugging.
529 struct lu_ref loh_reference;
539 LU_SS_CACHE_DEATH_RACE,
545 * lu_site is a "compartment" within which objects are unique, and LRU
546 * discipline is maintained.
548 * lu_site exists so that multiple layered stacks can co-exist in the same
551 * lu_site has the same relation to lu_device as lu_object_header to
558 struct cfs_hash *ls_obj_hash;
560 * index of bucket on hash table while purging
562 unsigned int ls_purge_start;
564 * Top-level device for this stack.
566 struct lu_device *ls_top_dev;
568 * Bottom-level device for this stack
570 struct lu_device *ls_bottom_dev;
572 * Linkage into global list of sites.
574 struct list_head ls_linkage;
576 * List for lu device for this site, protected
579 struct list_head ls_ld_linkage;
580 spinlock_t ls_ld_lock;
582 * Lock to serialize site purge.
584 struct mutex ls_purge_mutex;
588 struct lprocfs_stats *ls_stats;
590 * XXX: a hack! fld has to find md_site via site, remove when possible
592 struct seq_server_site *ld_seq_site;
594 * Pointer to the lu_target for this site.
596 struct lu_target *ls_tgt;
599 * Number of objects in lsb_lru_lists - used for shrinking
601 struct percpu_counter ls_lru_len_counter;
605 lu_site_wq_from_fid(struct lu_site *site, struct lu_fid *fid);
607 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
609 return s->ld_seq_site;
613 * Constructors/destructors.
617 int lu_site_init (struct lu_site *s, struct lu_device *d);
618 void lu_site_fini (struct lu_site *s);
619 int lu_site_init_finish (struct lu_site *s);
620 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
621 void lu_device_get (struct lu_device *d);
622 void lu_device_put (struct lu_device *d);
623 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
624 void lu_device_fini (struct lu_device *d);
625 int lu_object_header_init(struct lu_object_header *h);
626 void lu_object_header_fini(struct lu_object_header *h);
627 int lu_object_init (struct lu_object *o,
628 struct lu_object_header *h, struct lu_device *d);
629 void lu_object_fini (struct lu_object *o);
630 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
631 void lu_object_add (struct lu_object *before, struct lu_object *o);
633 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
634 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
637 * Helpers to initialize and finalize device types.
640 int lu_device_type_init(struct lu_device_type *ldt);
641 void lu_device_type_fini(struct lu_device_type *ldt);
646 * Caching and reference counting.
651 * Acquire additional reference to the given object. This function is used to
652 * attain additional reference. To acquire initial reference use
655 static inline void lu_object_get(struct lu_object *o)
657 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
658 atomic_inc(&o->lo_header->loh_ref);
662 * Return true if object will not be cached after last reference to it is
665 static inline int lu_object_is_dying(const struct lu_object_header *h)
667 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
670 void lu_object_put(const struct lu_env *env, struct lu_object *o);
671 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
672 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
673 int lu_site_purge_objects(const struct lu_env *env, struct lu_site *s, int nr,
676 static inline int lu_site_purge(const struct lu_env *env, struct lu_site *s,
679 return lu_site_purge_objects(env, s, nr, 1);
682 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
683 lu_printer_t printer);
684 struct lu_object *lu_object_find(const struct lu_env *env,
685 struct lu_device *dev, const struct lu_fid *f,
686 const struct lu_object_conf *conf);
687 struct lu_object *lu_object_find_at(const struct lu_env *env,
688 struct lu_device *dev,
689 const struct lu_fid *f,
690 const struct lu_object_conf *conf);
691 struct lu_object *lu_object_find_slice(const struct lu_env *env,
692 struct lu_device *dev,
693 const struct lu_fid *f,
694 const struct lu_object_conf *conf);
703 * First (topmost) sub-object of given compound object
705 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
707 LASSERT(!list_empty(&h->loh_layers));
708 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
712 * Next sub-object in the layering
714 static inline struct lu_object *lu_object_next(const struct lu_object *o)
716 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
720 * Pointer to the fid of this object.
722 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
724 return &o->lo_header->loh_fid;
728 * return device operations vector for this object
730 static const inline struct lu_device_operations *
731 lu_object_ops(const struct lu_object *o)
733 return o->lo_dev->ld_ops;
737 * Given a compound object, find its slice, corresponding to the device type
740 struct lu_object *lu_object_locate(struct lu_object_header *h,
741 const struct lu_device_type *dtype);
744 * Printer function emitting messages through libcfs_debug_msg().
746 int lu_cdebug_printer(const struct lu_env *env,
747 void *cookie, const char *format, ...);
750 * Print object description followed by a user-supplied message.
752 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
754 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
755 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
756 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
757 CDEBUG(mask, format "\n", ## __VA_ARGS__); \
762 * Print short object description followed by a user-supplied message.
764 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
766 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
767 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
768 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
769 (object)->lo_header); \
770 lu_cdebug_printer(env, &msgdata, "\n"); \
771 CDEBUG(mask, format , ## __VA_ARGS__); \
775 void lu_object_print (const struct lu_env *env, void *cookie,
776 lu_printer_t printer, const struct lu_object *o);
777 void lu_object_header_print(const struct lu_env *env, void *cookie,
778 lu_printer_t printer,
779 const struct lu_object_header *hdr);
782 * Check object consistency.
784 int lu_object_invariant(const struct lu_object *o);
788 * Check whether object exists, no matter on local or remote storage.
789 * Note: LOHA_EXISTS will be set once some one created the object,
790 * and it does not needs to be committed to storage.
792 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
795 * Check whether object on the remote storage.
797 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
800 * Check whether the object as agent entry on current target
802 #define lu_object_has_agent_entry(o) \
803 unlikely((o)->lo_header->loh_attr & LOHA_HAS_AGENT_ENTRY)
805 static inline void lu_object_set_agent_entry(struct lu_object *o)
807 o->lo_header->loh_attr |= LOHA_HAS_AGENT_ENTRY;
810 static inline void lu_object_clear_agent_entry(struct lu_object *o)
812 o->lo_header->loh_attr &= ~LOHA_HAS_AGENT_ENTRY;
815 static inline int lu_object_assert_exists(const struct lu_object *o)
817 return lu_object_exists(o);
820 static inline int lu_object_assert_not_exists(const struct lu_object *o)
822 return !lu_object_exists(o);
826 * Attr of this object.
828 static inline __u32 lu_object_attr(const struct lu_object *o)
830 LASSERT(lu_object_exists(o) != 0);
832 return o->lo_header->loh_attr & S_IFMT;
835 static inline void lu_object_ref_add(struct lu_object *o,
839 lu_ref_add(&o->lo_header->loh_reference, scope, source);
842 static inline void lu_object_ref_add_at(struct lu_object *o,
843 struct lu_ref_link *link,
847 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
850 static inline void lu_object_ref_del(struct lu_object *o,
851 const char *scope, const void *source)
853 lu_ref_del(&o->lo_header->loh_reference, scope, source);
856 static inline void lu_object_ref_del_at(struct lu_object *o,
857 struct lu_ref_link *link,
858 const char *scope, const void *source)
860 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
863 /** input params, should be filled out by mdt */
867 /** count in bytes */
868 unsigned int rp_count;
869 /** number of pages */
870 unsigned int rp_npages;
871 /** requested attr */
873 /** pointers to pages */
874 struct page **rp_pages;
877 enum lu_xattr_flags {
878 LU_XATTR_REPLACE = (1 << 0),
879 LU_XATTR_CREATE = (1 << 1),
880 LU_XATTR_MERGE = (1 << 2),
881 LU_XATTR_SPLIT = (1 << 3),
889 /** For lu_context health-checks */
890 enum lu_context_state {
899 * lu_context. Execution context for lu_object methods. Currently associated
902 * All lu_object methods, except device and device type methods (called during
903 * system initialization and shutdown) are executed "within" some
904 * lu_context. This means, that pointer to some "current" lu_context is passed
905 * as an argument to all methods.
907 * All service ptlrpc threads create lu_context as part of their
908 * initialization. It is possible to create "stand-alone" context for other
909 * execution environments (like system calls).
911 * lu_object methods mainly use lu_context through lu_context_key interface
912 * that allows each layer to associate arbitrary pieces of data with each
913 * context (see pthread_key_create(3) for similar interface).
915 * On a client, lu_context is bound to a thread, see cl_env_get().
917 * \see lu_context_key
921 * lu_context is used on the client side too. Yet we don't want to
922 * allocate values of server-side keys for the client contexts and
925 * To achieve this, set of tags in introduced. Contexts and keys are
926 * marked with tags. Key value are created only for context whose set
927 * of tags has non-empty intersection with one for key. Tags are taken
928 * from enum lu_context_tag.
931 enum lu_context_state lc_state;
933 * Pointer to the home service thread. NULL for other execution
936 struct ptlrpc_thread *lc_thread;
938 * Pointer to an array with key values. Internal implementation
943 * Linkage into a list of all remembered contexts. Only
944 * `non-transient' contexts, i.e., ones created for service threads
947 struct list_head lc_remember;
949 * Version counter used to skip calls to lu_context_refill() when no
950 * keys were registered.
960 * lu_context_key interface. Similar to pthread_key.
963 enum lu_context_tag {
965 * Thread on md server
967 LCT_MD_THREAD = 1 << 0,
969 * Thread on dt server
971 LCT_DT_THREAD = 1 << 1,
975 LCT_CL_THREAD = 1 << 3,
977 * A per-request session on a server, and a per-system-call session on
980 LCT_SESSION = 1 << 4,
982 * A per-request data on OSP device
984 LCT_OSP_THREAD = 1 << 5,
988 LCT_MG_THREAD = 1 << 6,
990 * Context for local operations
994 * session for server thread
996 LCT_SERVER_SESSION = 1 << 8,
998 * Set when at least one of keys, having values in this context has
999 * non-NULL lu_context_key::lct_exit() method. This is used to
1000 * optimize lu_context_exit() call.
1002 LCT_HAS_EXIT = 1 << 28,
1004 * Don't add references for modules creating key values in that context.
1005 * This is only for contexts used internally by lu_object framework.
1007 LCT_NOREF = 1 << 29,
1009 * Key is being prepared for retiring, don't create new values for it.
1011 LCT_QUIESCENT = 1 << 30,
1013 * Context should be remembered.
1015 LCT_REMEMBER = 1 << 31,
1017 * Contexts usable in cache shrinker thread.
1019 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1023 * Key. Represents per-context value slot.
1025 * Keys are usually registered when module owning the key is initialized, and
1026 * de-registered when module is unloaded. Once key is registered, all new
1027 * contexts with matching tags, will get key value. "Old" contexts, already
1028 * initialized at the time of key registration, can be forced to get key value
1029 * by calling lu_context_refill().
1031 * Every key value is counted in lu_context_key::lct_used and acquires a
1032 * reference on an owning module. This means, that all key values have to be
1033 * destroyed before module can be unloaded. This is usually achieved by
1034 * stopping threads started by the module, that created contexts in their
1035 * entry functions. Situation is complicated by the threads shared by multiple
1036 * modules, like ptlrpcd daemon on a client. To work around this problem,
1037 * contexts, created in such threads, are `remembered' (see
1038 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1039 * for unloading it does the following:
1041 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1042 * preventing new key values from being allocated in the new contexts,
1045 * - scans a list of remembered contexts, destroying values of module
1046 * keys, thus releasing references to the module.
1048 * This is done by lu_context_key_quiesce(). If module is re-activated
1049 * before key has been de-registered, lu_context_key_revive() call clears
1050 * `quiescent' marker.
1052 * lu_context code doesn't provide any internal synchronization for these
1053 * activities---it's assumed that startup (including threads start-up) and
1054 * shutdown are serialized by some external means.
1058 struct lu_context_key {
1060 * Set of tags for which values of this key are to be instantiated.
1064 * Value constructor. This is called when new value is created for a
1065 * context. Returns pointer to new value of error pointer.
1067 void *(*lct_init)(const struct lu_context *ctx,
1068 struct lu_context_key *key);
1070 * Value destructor. Called when context with previously allocated
1071 * value of this slot is destroyed. \a data is a value that was returned
1072 * by a matching call to lu_context_key::lct_init().
1074 void (*lct_fini)(const struct lu_context *ctx,
1075 struct lu_context_key *key, void *data);
1077 * Optional method called on lu_context_exit() for all allocated
1078 * keys. Can be used by debugging code checking that locks are
1081 void (*lct_exit)(const struct lu_context *ctx,
1082 struct lu_context_key *key, void *data);
1084 * Internal implementation detail: index within lu_context::lc_value[]
1085 * reserved for this key.
1089 * Internal implementation detail: number of values created for this
1094 * Internal implementation detail: module for this key.
1096 struct module *lct_owner;
1098 * References to this key. For debugging.
1100 struct lu_ref lct_reference;
1103 #define LU_KEY_INIT(mod, type) \
1104 static void *mod##_key_init(const struct lu_context *ctx, \
1105 struct lu_context_key *key) \
1109 CLASSERT(PAGE_SIZE >= sizeof(*value)); \
1111 OBD_ALLOC_PTR(value); \
1112 if (value == NULL) \
1113 value = ERR_PTR(-ENOMEM); \
1117 struct __##mod##__dummy_init { ; } /* semicolon catcher */
1119 #define LU_KEY_FINI(mod, type) \
1120 static void mod##_key_fini(const struct lu_context *ctx, \
1121 struct lu_context_key *key, void* data) \
1123 type *info = data; \
1125 OBD_FREE_PTR(info); \
1127 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1129 #define LU_KEY_INIT_FINI(mod, type) \
1130 LU_KEY_INIT(mod,type); \
1131 LU_KEY_FINI(mod,type)
1133 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1134 struct lu_context_key mod##_thread_key = { \
1136 .lct_init = mod##_key_init, \
1137 .lct_fini = mod##_key_fini \
1140 #define LU_CONTEXT_KEY_INIT(key) \
1142 (key)->lct_owner = THIS_MODULE; \
1145 int lu_context_key_register(struct lu_context_key *key);
1146 void lu_context_key_degister(struct lu_context_key *key);
1147 void *lu_context_key_get (const struct lu_context *ctx,
1148 const struct lu_context_key *key);
1149 void lu_context_key_quiesce (struct lu_context_key *key);
1150 void lu_context_key_revive (struct lu_context_key *key);
1154 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1158 #define LU_KEY_INIT_GENERIC(mod) \
1159 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1161 struct lu_context_key *key = k; \
1164 va_start(args, k); \
1166 LU_CONTEXT_KEY_INIT(key); \
1167 key = va_arg(args, struct lu_context_key *); \
1168 } while (key != NULL); \
1172 #define LU_TYPE_INIT(mod, ...) \
1173 LU_KEY_INIT_GENERIC(mod) \
1174 static int mod##_type_init(struct lu_device_type *t) \
1176 mod##_key_init_generic(__VA_ARGS__, NULL); \
1177 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1179 struct __##mod##_dummy_type_init {;}
1181 #define LU_TYPE_FINI(mod, ...) \
1182 static void mod##_type_fini(struct lu_device_type *t) \
1184 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1186 struct __##mod##_dummy_type_fini {;}
1188 #define LU_TYPE_START(mod, ...) \
1189 static void mod##_type_start(struct lu_device_type *t) \
1191 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1193 struct __##mod##_dummy_type_start {;}
1195 #define LU_TYPE_STOP(mod, ...) \
1196 static void mod##_type_stop(struct lu_device_type *t) \
1198 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1200 struct __##mod##_dummy_type_stop {;}
1204 #define LU_TYPE_INIT_FINI(mod, ...) \
1205 LU_TYPE_INIT(mod, __VA_ARGS__); \
1206 LU_TYPE_FINI(mod, __VA_ARGS__); \
1207 LU_TYPE_START(mod, __VA_ARGS__); \
1208 LU_TYPE_STOP(mod, __VA_ARGS__)
1210 int lu_context_init (struct lu_context *ctx, __u32 tags);
1211 void lu_context_fini (struct lu_context *ctx);
1212 void lu_context_enter (struct lu_context *ctx);
1213 void lu_context_exit (struct lu_context *ctx);
1214 int lu_context_refill(struct lu_context *ctx);
1217 * Helper functions to operate on multiple keys. These are used by the default
1218 * device type operations, defined by LU_TYPE_INIT_FINI().
1221 int lu_context_key_register_many(struct lu_context_key *k, ...);
1222 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1223 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1224 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1227 * update/clear ctx/ses tags.
1229 void lu_context_tags_update(__u32 tags);
1230 void lu_context_tags_clear(__u32 tags);
1231 void lu_session_tags_update(__u32 tags);
1232 void lu_session_tags_clear(__u32 tags);
1239 * "Local" context, used to store data instead of stack.
1241 struct lu_context le_ctx;
1243 * "Session" context for per-request data.
1245 struct lu_context *le_ses;
1248 int lu_env_init (struct lu_env *env, __u32 tags);
1249 void lu_env_fini (struct lu_env *env);
1250 int lu_env_refill(struct lu_env *env);
1251 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1253 struct lu_env *lu_env_find(void);
1254 int lu_env_add(struct lu_env *env);
1255 void lu_env_remove(struct lu_env *env);
1257 /** @} lu_context */
1260 * Output site statistical counters into a buffer. Suitable for
1261 * ll_rd_*()-style functions.
1263 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1266 * Common name structure to be passed around for various name related methods.
1269 const char *ln_name;
1273 static inline bool name_is_dot_or_dotdot(const char *name, int namelen)
1275 return name[0] == '.' &&
1276 (namelen == 1 || (namelen == 2 && name[1] == '.'));
1279 static inline bool lu_name_is_dot_or_dotdot(const struct lu_name *lname)
1281 return name_is_dot_or_dotdot(lname->ln_name, lname->ln_namelen);
1284 static inline bool lu_name_is_valid_len(const char *name, size_t name_len)
1286 return name != NULL &&
1288 name_len < INT_MAX &&
1289 strlen(name) == name_len &&
1290 memchr(name, '/', name_len) == NULL;
1294 * Validate names (path components)
1296 * To be valid \a name must be non-empty, '\0' terminated of length \a
1297 * name_len, and not contain '/'. The maximum length of a name (before
1298 * say -ENAMETOOLONG will be returned) is really controlled by llite
1299 * and the server. We only check for something insane coming from bad
1300 * integer handling here.
1302 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1304 return lu_name_is_valid_len(name, name_len) && name[name_len] == '\0';
1307 static inline bool lu_name_is_valid(const struct lu_name *ln)
1309 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1312 #define DNAME "%.*s"
1314 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1315 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1318 * Common buffer structure to be passed around for various xattr_{s,g}et()
1326 #define DLUBUF "(%p %zu)"
1327 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1329 /* read buffer params, should be filled out by out */
1331 /** number of buffers */
1332 unsigned int rb_nbufs;
1333 /** pointers to buffers */
1334 struct lu_buf rb_bufs[];
1338 * One-time initializers, called at obdclass module initialization, not
1343 * Initialization of global lu_* data.
1345 int lu_global_init(void);
1348 * Dual to lu_global_init().
1350 void lu_global_fini(void);
1352 struct lu_kmem_descr {
1353 struct kmem_cache **ckd_cache;
1354 const char *ckd_name;
1355 const size_t ckd_size;
1358 int lu_kmem_init(struct lu_kmem_descr *caches);
1359 void lu_kmem_fini(struct lu_kmem_descr *caches);
1361 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1362 const struct lu_fid *fid);
1363 struct lu_object *lu_object_anon(const struct lu_env *env,
1364 struct lu_device *dev,
1365 const struct lu_object_conf *conf);
1368 extern struct lu_buf LU_BUF_NULL;
1370 void lu_buf_free(struct lu_buf *buf);
1371 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1372 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1374 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1375 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1377 extern __u32 lu_context_tags_default;
1378 extern __u32 lu_session_tags_default;
1380 static inline bool lu_device_is_cl(const struct lu_device *d)
1382 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1385 static inline bool lu_object_is_cl(const struct lu_object *o)
1387 return lu_device_is_cl(o->lo_dev);
1390 /* Generic subset of OSTs */
1392 __u32 *op_array; /* array of index of
1393 * lov_obd->lov_tgts */
1394 unsigned int op_count; /* number of OSTs in the array */
1395 unsigned int op_size; /* allocated size of lp_array */
1396 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
1399 /* round-robin QoS data for LOD/LMV */
1401 spinlock_t lqr_alloc; /* protect allocation index */
1402 __u32 lqr_start_idx; /* start index of new inode */
1403 __u32 lqr_offset_idx;/* aliasing for start_idx */
1404 int lqr_start_count;/* reseed counter */
1405 struct ost_pool lqr_pool; /* round-robin optimized list */
1406 unsigned long lqr_dirty:1; /* recalc round-robin list */
1409 /* QoS data per MDS/OSS */
1411 struct obd_uuid lsq_uuid; /* ptlrpc's c_remote_uuid */
1412 struct list_head lsq_svr_list; /* link to lq_svr_list */
1413 __u64 lsq_bavail; /* total bytes avail on svr */
1414 __u64 lsq_iavail; /* tital inode avail on svr */
1415 __u64 lsq_penalty; /* current penalty */
1416 __u64 lsq_penalty_per_obj; /* penalty decrease
1418 time64_t lsq_used; /* last used time, seconds */
1419 __u32 lsq_tgt_count; /* number of tgts on this svr */
1420 __u32 lsq_id; /* unique svr id */
1423 /* QoS data per MDT/OST */
1425 struct lu_svr_qos *ltq_svr; /* svr info */
1426 __u64 ltq_penalty; /* current penalty */
1427 __u64 ltq_penalty_per_obj; /* penalty decrease
1429 __u64 ltq_weight; /* net weighting */
1430 time64_t ltq_used; /* last used time, seconds */
1431 bool ltq_usable:1; /* usable for striping */
1434 /* target descriptor */
1435 struct lu_tgt_desc {
1437 struct dt_device *ltd_tgt;
1438 struct obd_device *ltd_obd;
1440 struct obd_export *ltd_exp;
1441 struct obd_uuid ltd_uuid;
1444 struct list_head ltd_kill;
1445 struct ptlrpc_thread *ltd_recovery_thread;
1446 struct mutex ltd_fid_mutex;
1447 struct lu_tgt_qos ltd_qos; /* qos info per target */
1448 struct obd_statfs ltd_statfs;
1449 time64_t ltd_statfs_age;
1450 unsigned long ltd_active:1,/* is this target up for requests */
1451 ltd_activate:1,/* should target be activated */
1452 ltd_reap:1, /* should this target be deleted */
1453 ltd_got_update_log:1, /* Already got update log */
1454 ltd_connecting:1; /* target is connecting */
1457 /* QoS data for LOD/LMV */
1459 struct list_head lq_svr_list; /* lu_svr_qos list */
1460 struct rw_semaphore lq_rw_sem;
1461 __u32 lq_active_svr_count;
1462 unsigned int lq_prio_free; /* priority for free space */
1463 unsigned int lq_threshold_rr;/* priority for rr */
1464 struct lu_qos_rr lq_rr; /* round robin qos data */
1465 unsigned long lq_dirty:1, /* recalc qos data */
1466 lq_same_space:1,/* the servers all have approx.
1467 * the same space avail */
1468 lq_reset:1; /* zero current penalties */
1471 void lu_qos_rr_init(struct lu_qos_rr *lqr);
1472 int lqos_add_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1473 int lqos_del_tgt(struct lu_qos *qos, struct lu_tgt_desc *ltd);
1474 u64 lu_prandom_u64_max(u64 ep_ro);
1477 #endif /* __LUSTRE_LU_OBJECT_H */