[go: nahoru, domu]

1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37#ifndef __LUSTRE_LU_OBJECT_H
38#define __LUSTRE_LU_OBJECT_H
39
40#include <stdarg.h>
41#include "../../include/linux/libcfs/libcfs.h"
42#include "lustre/lustre_idl.h"
43#include "lu_ref.h"
44
45struct seq_file;
46struct proc_dir_entry;
47struct lustre_cfg;
48struct lprocfs_stats;
49
50/** \defgroup lu lu
51 * lu_* data-types represent server-side entities shared by data and meta-data
52 * stacks.
53 *
54 * Design goals:
55 *
56 * -# support for layering.
57 *
58 *     Server side object is split into layers, one per device in the
59 *     corresponding device stack. Individual layer is represented by struct
60 *     lu_object. Compound layered object --- by struct lu_object_header. Most
61 *     interface functions take lu_object as an argument and operate on the
62 *     whole compound object. This decision was made due to the following
63 *     reasons:
64 *
65 *	- it's envisaged that lu_object will be used much more often than
66 *	lu_object_header;
67 *
68 *	- we want lower (non-top) layers to be able to initiate operations
69 *	on the whole object.
70 *
71 *     Generic code supports layering more complex than simple stacking, e.g.,
72 *     it is possible that at some layer object "spawns" multiple sub-objects
73 *     on the lower layer.
74 *
75 * -# fid-based identification.
76 *
77 *     Compound object is uniquely identified by its fid. Objects are indexed
78 *     by their fids (hash table is used for index).
79 *
80 * -# caching and life-cycle management.
81 *
82 *     Object's life-time is controlled by reference counting. When reference
83 *     count drops to 0, object is returned to cache. Cached objects still
84 *     retain their identity (i.e., fid), and can be recovered from cache.
85 *
86 *     Objects are kept in the global LRU list, and lu_site_purge() function
87 *     can be used to reclaim given number of unused objects from the tail of
88 *     the LRU.
89 *
90 * -# avoiding recursion.
91 *
92 *     Generic code tries to replace recursion through layers by iterations
93 *     where possible. Additionally to the end of reducing stack consumption,
94 *     data, when practically possible, are allocated through lu_context_key
95 *     interface rather than on stack.
96 * @{
97 */
98
99struct lu_site;
100struct lu_object;
101struct lu_device;
102struct lu_object_header;
103struct lu_context;
104struct lu_env;
105
106/**
107 * Operations common for data and meta-data devices.
108 */
109struct lu_device_operations {
110	/**
111	 * Allocate object for the given device (without lower-layer
112	 * parts). This is called by lu_object_operations::loo_object_init()
113	 * from the parent layer, and should setup at least lu_object::lo_dev
114	 * and lu_object::lo_ops fields of resulting lu_object.
115	 *
116	 * Object creation protocol.
117	 *
118	 * Due to design goal of avoiding recursion, object creation (see
119	 * lu_object_alloc()) is somewhat involved:
120	 *
121	 *  - first, lu_device_operations::ldo_object_alloc() method of the
122	 *  top-level device in the stack is called. It should allocate top
123	 *  level object (including lu_object_header), but without any
124	 *  lower-layer sub-object(s).
125	 *
126	 *  - then lu_object_alloc() sets fid in the header of newly created
127	 *  object.
128	 *
129	 *  - then lu_object_operations::loo_object_init() is called. It has
130	 *  to allocate lower-layer object(s). To do this,
131	 *  lu_object_operations::loo_object_init() calls ldo_object_alloc()
132	 *  of the lower-layer device(s).
133	 *
134	 *  - for all new objects allocated by
135	 *  lu_object_operations::loo_object_init() (and inserted into object
136	 *  stack), lu_object_operations::loo_object_init() is called again
137	 *  repeatedly, until no new objects are created.
138	 *
139	 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
140	 *			     result->lo_ops != NULL);
141	 */
142	struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
143					      const struct lu_object_header *h,
144					      struct lu_device *d);
145	/**
146	 * process config specific for device.
147	 */
148	int (*ldo_process_config)(const struct lu_env *env,
149				  struct lu_device *, struct lustre_cfg *);
150	int (*ldo_recovery_complete)(const struct lu_env *,
151				     struct lu_device *);
152
153	/**
154	 * initialize local objects for device. this method called after layer has
155	 * been initialized (after LCFG_SETUP stage) and before it starts serving
156	 * user requests.
157	 */
158
159	int (*ldo_prepare)(const struct lu_env *,
160			   struct lu_device *parent,
161			   struct lu_device *dev);
162
163};
164
165/**
166 * For lu_object_conf flags
167 */
168typedef enum {
169	/* This is a new object to be allocated, or the file
170	 * corresponding to the object does not exists. */
171	LOC_F_NEW	= 0x00000001,
172} loc_flags_t;
173
174/**
175 * Object configuration, describing particulars of object being created. On
176 * server this is not used, as server objects are full identified by fid. On
177 * client configuration contains struct lustre_md.
178 */
179struct lu_object_conf {
180	/**
181	 * Some hints for obj find and alloc.
182	 */
183	loc_flags_t     loc_flags;
184};
185
186/**
187 * Type of "printer" function used by lu_object_operations::loo_object_print()
188 * method.
189 *
190 * Printer function is needed to provide some flexibility in (semi-)debugging
191 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
192 */
193typedef int (*lu_printer_t)(const struct lu_env *env,
194			    void *cookie, const char *format, ...)
195	__attribute__ ((format (printf, 3, 4)));
196
197/**
198 * Operations specific for particular lu_object.
199 */
200struct lu_object_operations {
201
202	/**
203	 * Allocate lower-layer parts of the object by calling
204	 * lu_device_operations::ldo_object_alloc() of the corresponding
205	 * underlying device.
206	 *
207	 * This method is called once for each object inserted into object
208	 * stack. It's responsibility of this method to insert lower-layer
209	 * object(s) it create into appropriate places of object stack.
210	 */
211	int (*loo_object_init)(const struct lu_env *env,
212			       struct lu_object *o,
213			       const struct lu_object_conf *conf);
214	/**
215	 * Called (in top-to-bottom order) during object allocation after all
216	 * layers were allocated and initialized. Can be used to perform
217	 * initialization depending on lower layers.
218	 */
219	int (*loo_object_start)(const struct lu_env *env,
220				struct lu_object *o);
221	/**
222	 * Called before lu_object_operations::loo_object_free() to signal
223	 * that object is being destroyed. Dual to
224	 * lu_object_operations::loo_object_init().
225	 */
226	void (*loo_object_delete)(const struct lu_env *env,
227				  struct lu_object *o);
228	/**
229	 * Dual to lu_device_operations::ldo_object_alloc(). Called when
230	 * object is removed from memory.
231	 */
232	void (*loo_object_free)(const struct lu_env *env,
233				struct lu_object *o);
234	/**
235	 * Called when last active reference to the object is released (and
236	 * object returns to the cache). This method is optional.
237	 */
238	void (*loo_object_release)(const struct lu_env *env,
239				   struct lu_object *o);
240	/**
241	 * Optional debugging helper. Print given object.
242	 */
243	int (*loo_object_print)(const struct lu_env *env, void *cookie,
244				lu_printer_t p, const struct lu_object *o);
245	/**
246	 * Optional debugging method. Returns true iff method is internally
247	 * consistent.
248	 */
249	int (*loo_object_invariant)(const struct lu_object *o);
250};
251
252/**
253 * Type of lu_device.
254 */
255struct lu_device_type;
256
257/**
258 * Device: a layer in the server side abstraction stacking.
259 */
260struct lu_device {
261	/**
262	 * reference count. This is incremented, in particular, on each object
263	 * created at this layer.
264	 *
265	 * \todo XXX which means that atomic_t is probably too small.
266	 */
267	atomic_t		       ld_ref;
268	/**
269	 * Pointer to device type. Never modified once set.
270	 */
271	struct lu_device_type       *ld_type;
272	/**
273	 * Operation vector for this device.
274	 */
275	const struct lu_device_operations *ld_ops;
276	/**
277	 * Stack this device belongs to.
278	 */
279	struct lu_site		    *ld_site;
280	struct proc_dir_entry	     *ld_proc_entry;
281
282	/** \todo XXX: temporary back pointer into obd. */
283	struct obd_device		 *ld_obd;
284	/**
285	 * A list of references to this object, for debugging.
286	 */
287	struct lu_ref		      ld_reference;
288	/**
289	 * Link the device to the site.
290	 **/
291	struct list_head			 ld_linkage;
292};
293
294struct lu_device_type_operations;
295
296/**
297 * Tag bits for device type. They are used to distinguish certain groups of
298 * device types.
299 */
300enum lu_device_tag {
301	/** this is meta-data device */
302	LU_DEVICE_MD = (1 << 0),
303	/** this is data device */
304	LU_DEVICE_DT = (1 << 1),
305	/** data device in the client stack */
306	LU_DEVICE_CL = (1 << 2)
307};
308
309/**
310 * Type of device.
311 */
312struct lu_device_type {
313	/**
314	 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
315	 */
316	__u32				   ldt_tags;
317	/**
318	 * Name of this class. Unique system-wide. Never modified once set.
319	 */
320	char				   *ldt_name;
321	/**
322	 * Operations for this type.
323	 */
324	const struct lu_device_type_operations *ldt_ops;
325	/**
326	 * \todo XXX: temporary pointer to associated obd_type.
327	 */
328	struct obd_type			*ldt_obd_type;
329	/**
330	 * \todo XXX: temporary: context tags used by obd_*() calls.
331	 */
332	__u32				   ldt_ctx_tags;
333	/**
334	 * Number of existing device type instances.
335	 */
336	unsigned				ldt_device_nr;
337	/**
338	 * Linkage into a global list of all device types.
339	 *
340	 * \see lu_device_types.
341	 */
342	struct list_head			      ldt_linkage;
343};
344
345/**
346 * Operations on a device type.
347 */
348struct lu_device_type_operations {
349	/**
350	 * Allocate new device.
351	 */
352	struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
353					       struct lu_device_type *t,
354					       struct lustre_cfg *lcfg);
355	/**
356	 * Free device. Dual to
357	 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
358	 * the next device in the stack.
359	 */
360	struct lu_device *(*ldto_device_free)(const struct lu_env *,
361					      struct lu_device *);
362
363	/**
364	 * Initialize the devices after allocation
365	 */
366	int  (*ldto_device_init)(const struct lu_env *env,
367				 struct lu_device *, const char *,
368				 struct lu_device *);
369	/**
370	 * Finalize device. Dual to
371	 * lu_device_type_operations::ldto_device_init(). Returns pointer to
372	 * the next device in the stack.
373	 */
374	struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
375					      struct lu_device *);
376	/**
377	 * Initialize device type. This is called on module load.
378	 */
379	int  (*ldto_init)(struct lu_device_type *t);
380	/**
381	 * Finalize device type. Dual to
382	 * lu_device_type_operations::ldto_init(). Called on module unload.
383	 */
384	void (*ldto_fini)(struct lu_device_type *t);
385	/**
386	 * Called when the first device is created.
387	 */
388	void (*ldto_start)(struct lu_device_type *t);
389	/**
390	 * Called when number of devices drops to 0.
391	 */
392	void (*ldto_stop)(struct lu_device_type *t);
393};
394
395static inline int lu_device_is_md(const struct lu_device *d)
396{
397	return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
398}
399
400/**
401 * Common object attributes.
402 */
403struct lu_attr {
404	/** size in bytes */
405	__u64	  la_size;
406	/** modification time in seconds since Epoch */
407	s64	  la_mtime;
408	/** access time in seconds since Epoch */
409	s64	  la_atime;
410	/** change time in seconds since Epoch */
411	s64	  la_ctime;
412	/** 512-byte blocks allocated to object */
413	__u64	  la_blocks;
414	/** permission bits and file type */
415	__u32	  la_mode;
416	/** owner id */
417	__u32	  la_uid;
418	/** group id */
419	__u32	  la_gid;
420	/** object flags */
421	__u32	  la_flags;
422	/** number of persistent references to this object */
423	__u32	  la_nlink;
424	/** blk bits of the object*/
425	__u32	  la_blkbits;
426	/** blk size of the object*/
427	__u32	  la_blksize;
428	/** real device */
429	__u32	  la_rdev;
430	/**
431	 * valid bits
432	 *
433	 * \see enum la_valid
434	 */
435	__u64	  la_valid;
436};
437
438/** Bit-mask of valid attributes */
439enum la_valid {
440	LA_ATIME = 1 << 0,
441	LA_MTIME = 1 << 1,
442	LA_CTIME = 1 << 2,
443	LA_SIZE  = 1 << 3,
444	LA_MODE  = 1 << 4,
445	LA_UID   = 1 << 5,
446	LA_GID   = 1 << 6,
447	LA_BLOCKS = 1 << 7,
448	LA_TYPE   = 1 << 8,
449	LA_FLAGS  = 1 << 9,
450	LA_NLINK  = 1 << 10,
451	LA_RDEV   = 1 << 11,
452	LA_BLKSIZE = 1 << 12,
453	LA_KILL_SUID = 1 << 13,
454	LA_KILL_SGID = 1 << 14,
455};
456
457/**
458 * Layer in the layered object.
459 */
460struct lu_object {
461	/**
462	 * Header for this object.
463	 */
464	struct lu_object_header	   *lo_header;
465	/**
466	 * Device for this layer.
467	 */
468	struct lu_device		  *lo_dev;
469	/**
470	 * Operations for this object.
471	 */
472	const struct lu_object_operations *lo_ops;
473	/**
474	 * Linkage into list of all layers.
475	 */
476	struct list_head			 lo_linkage;
477	/**
478	 * Link to the device, for debugging.
479	 */
480	struct lu_ref_link                 lo_dev_ref;
481};
482
483enum lu_object_header_flags {
484	/**
485	 * Don't keep this object in cache. Object will be destroyed as soon
486	 * as last reference to it is released. This flag cannot be cleared
487	 * once set.
488	 */
489	LU_OBJECT_HEARD_BANSHEE = 0,
490	/**
491	 * Mark this object has already been taken out of cache.
492	 */
493	LU_OBJECT_UNHASHED = 1
494};
495
496enum lu_object_header_attr {
497	LOHA_EXISTS   = 1 << 0,
498	LOHA_REMOTE   = 1 << 1,
499	/**
500	 * UNIX file type is stored in S_IFMT bits.
501	 */
502	LOHA_FT_START = 001 << 12, /**< S_IFIFO */
503	LOHA_FT_END   = 017 << 12, /**< S_IFMT */
504};
505
506/**
507 * "Compound" object, consisting of multiple layers.
508 *
509 * Compound object with given fid is unique with given lu_site.
510 *
511 * Note, that object does *not* necessary correspond to the real object in the
512 * persistent storage: object is an anchor for locking and method calling, so
513 * it is created for things like not-yet-existing child created by mkdir or
514 * create calls. lu_object_operations::loo_exists() can be used to check
515 * whether object is backed by persistent storage entity.
516 */
517struct lu_object_header {
518	/**
519	 * Fid, uniquely identifying this object.
520	 */
521	struct lu_fid		loh_fid;
522	/**
523	 * Object flags from enum lu_object_header_flags. Set and checked
524	 * atomically.
525	 */
526	unsigned long	  loh_flags;
527	/**
528	 * Object reference count. Protected by lu_site::ls_guard.
529	 */
530	atomic_t	   loh_ref;
531	/**
532	 * Common object attributes, cached for efficiency. From enum
533	 * lu_object_header_attr.
534	 */
535	__u32		  loh_attr;
536	/**
537	 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
538	 */
539	struct hlist_node       loh_hash;
540	/**
541	 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
542	 */
543	struct list_head	     loh_lru;
544	/**
545	 * Linkage into list of layers. Never modified once set (except lately
546	 * during object destruction). No locking is necessary.
547	 */
548	struct list_head	     loh_layers;
549	/**
550	 * A list of references to this object, for debugging.
551	 */
552	struct lu_ref	  loh_reference;
553};
554
555struct fld;
556
557struct lu_site_bkt_data {
558	/**
559	 * number of busy object on this bucket
560	 */
561	long		      lsb_busy;
562	/**
563	 * LRU list, updated on each access to object. Protected by
564	 * bucket lock of lu_site::ls_obj_hash.
565	 *
566	 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
567	 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
568	 * of list_for_each_entry_safe_reverse()).
569	 */
570	struct list_head		lsb_lru;
571	/**
572	 * Wait-queue signaled when an object in this site is ultimately
573	 * destroyed (lu_object_free()). It is used by lu_object_find() to
574	 * wait before re-trying when object in the process of destruction is
575	 * found in the hash table.
576	 *
577	 * \see htable_lookup().
578	 */
579	wait_queue_head_t	       lsb_marche_funebre;
580};
581
582enum {
583	LU_SS_CREATED	 = 0,
584	LU_SS_CACHE_HIT,
585	LU_SS_CACHE_MISS,
586	LU_SS_CACHE_RACE,
587	LU_SS_CACHE_DEATH_RACE,
588	LU_SS_LRU_PURGED,
589	LU_SS_LAST_STAT
590};
591
592/**
593 * lu_site is a "compartment" within which objects are unique, and LRU
594 * discipline is maintained.
595 *
596 * lu_site exists so that multiple layered stacks can co-exist in the same
597 * address space.
598 *
599 * lu_site has the same relation to lu_device as lu_object_header to
600 * lu_object.
601 */
602struct lu_site {
603	/**
604	 * objects hash table
605	 */
606	struct cfs_hash	       *ls_obj_hash;
607	/**
608	 * index of bucket on hash table while purging
609	 */
610	int		       ls_purge_start;
611	/**
612	 * Top-level device for this stack.
613	 */
614	struct lu_device	 *ls_top_dev;
615	/**
616	 * Bottom-level device for this stack
617	 */
618	struct lu_device	*ls_bottom_dev;
619	/**
620	 * Linkage into global list of sites.
621	 */
622	struct list_head		ls_linkage;
623	/**
624	 * List for lu device for this site, protected
625	 * by ls_ld_lock.
626	 **/
627	struct list_head		ls_ld_linkage;
628	spinlock_t		ls_ld_lock;
629
630	/**
631	 * lu_site stats
632	 */
633	struct lprocfs_stats	*ls_stats;
634	/**
635	 * XXX: a hack! fld has to find md_site via site, remove when possible
636	 */
637	struct seq_server_site	*ld_seq_site;
638};
639
640static inline struct lu_site_bkt_data *
641lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
642{
643	struct cfs_hash_bd bd;
644
645	cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
646	return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
647}
648
649static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
650{
651	return s->ld_seq_site;
652}
653
654/** \name ctors
655 * Constructors/destructors.
656 * @{
657 */
658
659int  lu_site_init	 (struct lu_site *s, struct lu_device *d);
660void lu_site_fini	 (struct lu_site *s);
661int  lu_site_init_finish  (struct lu_site *s);
662void lu_stack_fini	(const struct lu_env *env, struct lu_device *top);
663void lu_device_get	(struct lu_device *d);
664void lu_device_put	(struct lu_device *d);
665int  lu_device_init       (struct lu_device *d, struct lu_device_type *t);
666void lu_device_fini       (struct lu_device *d);
667int  lu_object_header_init(struct lu_object_header *h);
668void lu_object_header_fini(struct lu_object_header *h);
669int  lu_object_init       (struct lu_object *o,
670			   struct lu_object_header *h, struct lu_device *d);
671void lu_object_fini       (struct lu_object *o);
672void lu_object_add_top    (struct lu_object_header *h, struct lu_object *o);
673void lu_object_add	(struct lu_object *before, struct lu_object *o);
674
675void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
676void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
677
678/**
679 * Helpers to initialize and finalize device types.
680 */
681
682int  lu_device_type_init(struct lu_device_type *ldt);
683void lu_device_type_fini(struct lu_device_type *ldt);
684void lu_types_stop(void);
685
686/** @} ctors */
687
688/** \name caching
689 * Caching and reference counting.
690 * @{
691 */
692
693/**
694 * Acquire additional reference to the given object. This function is used to
695 * attain additional reference. To acquire initial reference use
696 * lu_object_find().
697 */
698static inline void lu_object_get(struct lu_object *o)
699{
700	LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
701	atomic_inc(&o->lo_header->loh_ref);
702}
703
704/**
705 * Return true of object will not be cached after last reference to it is
706 * released.
707 */
708static inline int lu_object_is_dying(const struct lu_object_header *h)
709{
710	return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
711}
712
713void lu_object_put(const struct lu_env *env, struct lu_object *o);
714void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
715void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
716
717int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
718
719void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
720		   lu_printer_t printer);
721struct lu_object *lu_object_find(const struct lu_env *env,
722				 struct lu_device *dev, const struct lu_fid *f,
723				 const struct lu_object_conf *conf);
724struct lu_object *lu_object_find_at(const struct lu_env *env,
725				    struct lu_device *dev,
726				    const struct lu_fid *f,
727				    const struct lu_object_conf *conf);
728struct lu_object *lu_object_find_slice(const struct lu_env *env,
729				       struct lu_device *dev,
730				       const struct lu_fid *f,
731				       const struct lu_object_conf *conf);
732/** @} caching */
733
734/** \name helpers
735 * Helpers.
736 * @{
737 */
738
739/**
740 * First (topmost) sub-object of given compound object
741 */
742static inline struct lu_object *lu_object_top(struct lu_object_header *h)
743{
744	LASSERT(!list_empty(&h->loh_layers));
745	return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
746}
747
748/**
749 * Next sub-object in the layering
750 */
751static inline struct lu_object *lu_object_next(const struct lu_object *o)
752{
753	return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
754}
755
756/**
757 * Pointer to the fid of this object.
758 */
759static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
760{
761	return &o->lo_header->loh_fid;
762}
763
764/**
765 * return device operations vector for this object
766 */
767static const inline struct lu_device_operations *
768lu_object_ops(const struct lu_object *o)
769{
770	return o->lo_dev->ld_ops;
771}
772
773/**
774 * Given a compound object, find its slice, corresponding to the device type
775 * \a dtype.
776 */
777struct lu_object *lu_object_locate(struct lu_object_header *h,
778				   const struct lu_device_type *dtype);
779
780/**
781 * Printer function emitting messages through libcfs_debug_msg().
782 */
783int lu_cdebug_printer(const struct lu_env *env,
784		      void *cookie, const char *format, ...);
785
786/**
787 * Print object description followed by a user-supplied message.
788 */
789#define LU_OBJECT_DEBUG(mask, env, object, format, ...)		   \
790do {								      \
791	LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL);		  \
792									  \
793	if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) {		     \
794		lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
795		CDEBUG(mask, format , ## __VA_ARGS__);		    \
796	}								 \
797} while (0)
798
799/**
800 * Print short object description followed by a user-supplied message.
801 */
802#define LU_OBJECT_HEADER(mask, env, object, format, ...)		\
803do {								    \
804	LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL);		\
805									\
806	if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) {		   \
807		lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
808				       (object)->lo_header);	    \
809		lu_cdebug_printer(env, &msgdata, "\n");		 \
810		CDEBUG(mask, format , ## __VA_ARGS__);		  \
811	}							       \
812} while (0)
813
814void lu_object_print       (const struct lu_env *env, void *cookie,
815			    lu_printer_t printer, const struct lu_object *o);
816void lu_object_header_print(const struct lu_env *env, void *cookie,
817			    lu_printer_t printer,
818			    const struct lu_object_header *hdr);
819
820/**
821 * Check object consistency.
822 */
823int lu_object_invariant(const struct lu_object *o);
824
825
826/**
827 * Check whether object exists, no matter on local or remote storage.
828 * Note: LOHA_EXISTS will be set once some one created the object,
829 * and it does not needs to be committed to storage.
830 */
831#define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
832
833/**
834 * Check whether object on the remote storage.
835 */
836#define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
837
838static inline int lu_object_assert_exists(const struct lu_object *o)
839{
840	return lu_object_exists(o);
841}
842
843static inline int lu_object_assert_not_exists(const struct lu_object *o)
844{
845	return !lu_object_exists(o);
846}
847
848/**
849 * Attr of this object.
850 */
851static inline __u32 lu_object_attr(const struct lu_object *o)
852{
853	LASSERT(lu_object_exists(o) != 0);
854	return o->lo_header->loh_attr;
855}
856
857static inline void lu_object_ref_add(struct lu_object *o,
858				     const char *scope,
859				     const void *source)
860{
861	lu_ref_add(&o->lo_header->loh_reference, scope, source);
862}
863
864static inline void lu_object_ref_add_at(struct lu_object *o,
865					struct lu_ref_link *link,
866					const char *scope,
867					const void *source)
868{
869	lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
870}
871
872static inline void lu_object_ref_del(struct lu_object *o,
873				     const char *scope, const void *source)
874{
875	lu_ref_del(&o->lo_header->loh_reference, scope, source);
876}
877
878static inline void lu_object_ref_del_at(struct lu_object *o,
879					struct lu_ref_link *link,
880					const char *scope, const void *source)
881{
882	lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
883}
884
885/** input params, should be filled out by mdt */
886struct lu_rdpg {
887	/** hash */
888	__u64		   rp_hash;
889	/** count in bytes */
890	unsigned int	    rp_count;
891	/** number of pages */
892	unsigned int	    rp_npages;
893	/** requested attr */
894	__u32		   rp_attrs;
895	/** pointers to pages */
896	struct page	   **rp_pages;
897};
898
899enum lu_xattr_flags {
900	LU_XATTR_REPLACE = (1 << 0),
901	LU_XATTR_CREATE  = (1 << 1)
902};
903
904/** @} helpers */
905
906/** \name lu_context
907 * @{ */
908
909/** For lu_context health-checks */
910enum lu_context_state {
911	LCS_INITIALIZED = 1,
912	LCS_ENTERED,
913	LCS_LEFT,
914	LCS_FINALIZED
915};
916
917/**
918 * lu_context. Execution context for lu_object methods. Currently associated
919 * with thread.
920 *
921 * All lu_object methods, except device and device type methods (called during
922 * system initialization and shutdown) are executed "within" some
923 * lu_context. This means, that pointer to some "current" lu_context is passed
924 * as an argument to all methods.
925 *
926 * All service ptlrpc threads create lu_context as part of their
927 * initialization. It is possible to create "stand-alone" context for other
928 * execution environments (like system calls).
929 *
930 * lu_object methods mainly use lu_context through lu_context_key interface
931 * that allows each layer to associate arbitrary pieces of data with each
932 * context (see pthread_key_create(3) for similar interface).
933 *
934 * On a client, lu_context is bound to a thread, see cl_env_get().
935 *
936 * \see lu_context_key
937 */
938struct lu_context {
939	/**
940	 * lu_context is used on the client side too. Yet we don't want to
941	 * allocate values of server-side keys for the client contexts and
942	 * vice versa.
943	 *
944	 * To achieve this, set of tags in introduced. Contexts and keys are
945	 * marked with tags. Key value are created only for context whose set
946	 * of tags has non-empty intersection with one for key. Tags are taken
947	 * from enum lu_context_tag.
948	 */
949	__u32		  lc_tags;
950	enum lu_context_state  lc_state;
951	/**
952	 * Pointer to the home service thread. NULL for other execution
953	 * contexts.
954	 */
955	struct ptlrpc_thread  *lc_thread;
956	/**
957	 * Pointer to an array with key values. Internal implementation
958	 * detail.
959	 */
960	void		 **lc_value;
961	/**
962	 * Linkage into a list of all remembered contexts. Only
963	 * `non-transient' contexts, i.e., ones created for service threads
964	 * are placed here.
965	 */
966	struct list_head	     lc_remember;
967	/**
968	 * Version counter used to skip calls to lu_context_refill() when no
969	 * keys were registered.
970	 */
971	unsigned	       lc_version;
972	/**
973	 * Debugging cookie.
974	 */
975	unsigned	       lc_cookie;
976};
977
978/**
979 * lu_context_key interface. Similar to pthread_key.
980 */
981
982enum lu_context_tag {
983	/**
984	 * Thread on md server
985	 */
986	LCT_MD_THREAD = 1 << 0,
987	/**
988	 * Thread on dt server
989	 */
990	LCT_DT_THREAD = 1 << 1,
991	/**
992	 * Context for transaction handle
993	 */
994	LCT_TX_HANDLE = 1 << 2,
995	/**
996	 * Thread on client
997	 */
998	LCT_CL_THREAD = 1 << 3,
999	/**
1000	 * A per-request session on a server, and a per-system-call session on
1001	 * a client.
1002	 */
1003	LCT_SESSION   = 1 << 4,
1004	/**
1005	 * A per-request data on OSP device
1006	 */
1007	LCT_OSP_THREAD = 1 << 5,
1008	/**
1009	 * MGS device thread
1010	 */
1011	LCT_MG_THREAD = 1 << 6,
1012	/**
1013	 * Context for local operations
1014	 */
1015	LCT_LOCAL = 1 << 7,
1016	/**
1017	 * Set when at least one of keys, having values in this context has
1018	 * non-NULL lu_context_key::lct_exit() method. This is used to
1019	 * optimize lu_context_exit() call.
1020	 */
1021	LCT_HAS_EXIT  = 1 << 28,
1022	/**
1023	 * Don't add references for modules creating key values in that context.
1024	 * This is only for contexts used internally by lu_object framework.
1025	 */
1026	LCT_NOREF     = 1 << 29,
1027	/**
1028	 * Key is being prepared for retiring, don't create new values for it.
1029	 */
1030	LCT_QUIESCENT = 1 << 30,
1031	/**
1032	 * Context should be remembered.
1033	 */
1034	LCT_REMEMBER  = 1 << 31,
1035	/**
1036	 * Contexts usable in cache shrinker thread.
1037	 */
1038	LCT_SHRINKER  = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1039};
1040
1041/**
1042 * Key. Represents per-context value slot.
1043 *
1044 * Keys are usually registered when module owning the key is initialized, and
1045 * de-registered when module is unloaded. Once key is registered, all new
1046 * contexts with matching tags, will get key value. "Old" contexts, already
1047 * initialized at the time of key registration, can be forced to get key value
1048 * by calling lu_context_refill().
1049 *
1050 * Every key value is counted in lu_context_key::lct_used and acquires a
1051 * reference on an owning module. This means, that all key values have to be
1052 * destroyed before module can be unloaded. This is usually achieved by
1053 * stopping threads started by the module, that created contexts in their
1054 * entry functions. Situation is complicated by the threads shared by multiple
1055 * modules, like ptlrpcd daemon on a client. To work around this problem,
1056 * contexts, created in such threads, are `remembered' (see
1057 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1058 * for unloading it does the following:
1059 *
1060 *     - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1061 *       preventing new key values from being allocated in the new contexts,
1062 *       and
1063 *
1064 *     - scans a list of remembered contexts, destroying values of module
1065 *       keys, thus releasing references to the module.
1066 *
1067 * This is done by lu_context_key_quiesce(). If module is re-activated
1068 * before key has been de-registered, lu_context_key_revive() call clears
1069 * `quiescent' marker.
1070 *
1071 * lu_context code doesn't provide any internal synchronization for these
1072 * activities---it's assumed that startup (including threads start-up) and
1073 * shutdown are serialized by some external means.
1074 *
1075 * \see lu_context
1076 */
1077struct lu_context_key {
1078	/**
1079	 * Set of tags for which values of this key are to be instantiated.
1080	 */
1081	__u32 lct_tags;
1082	/**
1083	 * Value constructor. This is called when new value is created for a
1084	 * context. Returns pointer to new value of error pointer.
1085	 */
1086	void  *(*lct_init)(const struct lu_context *ctx,
1087			   struct lu_context_key *key);
1088	/**
1089	 * Value destructor. Called when context with previously allocated
1090	 * value of this slot is destroyed. \a data is a value that was returned
1091	 * by a matching call to lu_context_key::lct_init().
1092	 */
1093	void   (*lct_fini)(const struct lu_context *ctx,
1094			   struct lu_context_key *key, void *data);
1095	/**
1096	 * Optional method called on lu_context_exit() for all allocated
1097	 * keys. Can be used by debugging code checking that locks are
1098	 * released, etc.
1099	 */
1100	void   (*lct_exit)(const struct lu_context *ctx,
1101			   struct lu_context_key *key, void *data);
1102	/**
1103	 * Internal implementation detail: index within lu_context::lc_value[]
1104	 * reserved for this key.
1105	 */
1106	int      lct_index;
1107	/**
1108	 * Internal implementation detail: number of values created for this
1109	 * key.
1110	 */
1111	atomic_t lct_used;
1112	/**
1113	 * Internal implementation detail: module for this key.
1114	 */
1115	struct module *lct_owner;
1116	/**
1117	 * References to this key. For debugging.
1118	 */
1119	struct lu_ref  lct_reference;
1120};
1121
1122#define LU_KEY_INIT(mod, type)				    \
1123	static void* mod##_key_init(const struct lu_context *ctx, \
1124				    struct lu_context_key *key)   \
1125	{							 \
1126		type *value;				      \
1127								  \
1128		CLASSERT(PAGE_CACHE_SIZE >= sizeof (*value));       \
1129								  \
1130		OBD_ALLOC_PTR(value);			     \
1131		if (value == NULL)				\
1132			value = ERR_PTR(-ENOMEM);		 \
1133								  \
1134		return value;				     \
1135	}							 \
1136	struct __##mod##__dummy_init {;} /* semicolon catcher */
1137
1138#define LU_KEY_FINI(mod, type)					      \
1139	static void mod##_key_fini(const struct lu_context *ctx,	    \
1140				    struct lu_context_key *key, void* data) \
1141	{								   \
1142		type *info = data;					  \
1143									    \
1144		OBD_FREE_PTR(info);					 \
1145	}								   \
1146	struct __##mod##__dummy_fini {;} /* semicolon catcher */
1147
1148#define LU_KEY_INIT_FINI(mod, type)   \
1149	LU_KEY_INIT(mod, type);	\
1150	LU_KEY_FINI(mod, type)
1151
1152#define LU_CONTEXT_KEY_DEFINE(mod, tags)		\
1153	struct lu_context_key mod##_thread_key = {      \
1154		.lct_tags = tags,		       \
1155		.lct_init = mod##_key_init,	     \
1156		.lct_fini = mod##_key_fini	      \
1157	}
1158
1159#define LU_CONTEXT_KEY_INIT(key)			\
1160do {						    \
1161	(key)->lct_owner = THIS_MODULE;		 \
1162} while (0)
1163
1164int   lu_context_key_register(struct lu_context_key *key);
1165void  lu_context_key_degister(struct lu_context_key *key);
1166void *lu_context_key_get     (const struct lu_context *ctx,
1167			       const struct lu_context_key *key);
1168void  lu_context_key_quiesce (struct lu_context_key *key);
1169void  lu_context_key_revive  (struct lu_context_key *key);
1170
1171
1172/*
1173 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1174 * owning module.
1175 */
1176
1177#define LU_KEY_INIT_GENERIC(mod)					\
1178	static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1179	{							       \
1180		struct lu_context_key *key = k;			 \
1181		va_list args;					   \
1182									\
1183		va_start(args, k);				      \
1184		do {						    \
1185			LU_CONTEXT_KEY_INIT(key);		       \
1186			key = va_arg(args, struct lu_context_key *);    \
1187		} while (key != NULL);				  \
1188		va_end(args);					   \
1189	}
1190
1191#define LU_TYPE_INIT(mod, ...)					  \
1192	LU_KEY_INIT_GENERIC(mod)					\
1193	static int mod##_type_init(struct lu_device_type *t)	    \
1194	{							       \
1195		mod##_key_init_generic(__VA_ARGS__, NULL);	      \
1196		return lu_context_key_register_many(__VA_ARGS__, NULL); \
1197	}							       \
1198	struct __##mod##_dummy_type_init {;}
1199
1200#define LU_TYPE_FINI(mod, ...)					  \
1201	static void mod##_type_fini(struct lu_device_type *t)	   \
1202	{							       \
1203		lu_context_key_degister_many(__VA_ARGS__, NULL);	\
1204	}							       \
1205	struct __##mod##_dummy_type_fini {;}
1206
1207#define LU_TYPE_START(mod, ...)				 \
1208	static void mod##_type_start(struct lu_device_type *t)  \
1209	{						       \
1210		lu_context_key_revive_many(__VA_ARGS__, NULL);  \
1211	}						       \
1212	struct __##mod##_dummy_type_start {;}
1213
1214#define LU_TYPE_STOP(mod, ...)				  \
1215	static void mod##_type_stop(struct lu_device_type *t)   \
1216	{						       \
1217		lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1218	}						       \
1219	struct __##mod##_dummy_type_stop {;}
1220
1221
1222
1223#define LU_TYPE_INIT_FINI(mod, ...)	     \
1224	LU_TYPE_INIT(mod, __VA_ARGS__);	 \
1225	LU_TYPE_FINI(mod, __VA_ARGS__);	 \
1226	LU_TYPE_START(mod, __VA_ARGS__);	\
1227	LU_TYPE_STOP(mod, __VA_ARGS__)
1228
1229int   lu_context_init  (struct lu_context *ctx, __u32 tags);
1230void  lu_context_fini  (struct lu_context *ctx);
1231void  lu_context_enter (struct lu_context *ctx);
1232void  lu_context_exit  (struct lu_context *ctx);
1233int   lu_context_refill(struct lu_context *ctx);
1234
1235/*
1236 * Helper functions to operate on multiple keys. These are used by the default
1237 * device type operations, defined by LU_TYPE_INIT_FINI().
1238 */
1239
1240int  lu_context_key_register_many(struct lu_context_key *k, ...);
1241void lu_context_key_degister_many(struct lu_context_key *k, ...);
1242void lu_context_key_revive_many  (struct lu_context_key *k, ...);
1243void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1244
1245/*
1246 * update/clear ctx/ses tags.
1247 */
1248void lu_context_tags_update(__u32 tags);
1249void lu_context_tags_clear(__u32 tags);
1250void lu_session_tags_update(__u32 tags);
1251void lu_session_tags_clear(__u32 tags);
1252
1253/**
1254 * Environment.
1255 */
1256struct lu_env {
1257	/**
1258	 * "Local" context, used to store data instead of stack.
1259	 */
1260	struct lu_context  le_ctx;
1261	/**
1262	 * "Session" context for per-request data.
1263	 */
1264	struct lu_context *le_ses;
1265};
1266
1267int  lu_env_init  (struct lu_env *env, __u32 tags);
1268void lu_env_fini  (struct lu_env *env);
1269int  lu_env_refill(struct lu_env *env);
1270int  lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1271
1272/** @} lu_context */
1273
1274/**
1275 * Output site statistical counters into a buffer. Suitable for
1276 * ll_rd_*()-style functions.
1277 */
1278int lu_site_stats_print(const struct lu_site *s, struct seq_file *m);
1279
1280/**
1281 * Common name structure to be passed around for various name related methods.
1282 */
1283struct lu_name {
1284	const char    *ln_name;
1285	int	    ln_namelen;
1286};
1287
1288/**
1289 * Common buffer structure to be passed around for various xattr_{s,g}et()
1290 * methods.
1291 */
1292struct lu_buf {
1293	void   *lb_buf;
1294	ssize_t lb_len;
1295};
1296
1297#define DLUBUF "(%p %zu)"
1298#define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1299/**
1300 * One-time initializers, called at obdclass module initialization, not
1301 * exported.
1302 */
1303
1304/**
1305 * Initialization of global lu_* data.
1306 */
1307int lu_global_init(void);
1308
1309/**
1310 * Dual to lu_global_init().
1311 */
1312void lu_global_fini(void);
1313
1314struct lu_kmem_descr {
1315	struct kmem_cache **ckd_cache;
1316	const char       *ckd_name;
1317	const size_t      ckd_size;
1318};
1319
1320int  lu_kmem_init(struct lu_kmem_descr *caches);
1321void lu_kmem_fini(struct lu_kmem_descr *caches);
1322
1323void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1324			  const struct lu_fid *fid);
1325struct lu_object *lu_object_anon(const struct lu_env *env,
1326				 struct lu_device *dev,
1327				 const struct lu_object_conf *conf);
1328
1329/** null buffer */
1330extern struct lu_buf LU_BUF_NULL;
1331
1332void lu_buf_free(struct lu_buf *buf);
1333void lu_buf_alloc(struct lu_buf *buf, int size);
1334void lu_buf_realloc(struct lu_buf *buf, int size);
1335
1336int lu_buf_check_and_grow(struct lu_buf *buf, int len);
1337struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len);
1338
1339/** @} lu */
1340#endif /* __LUSTRE_LU_OBJECT_H */
1341