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.sun.com/software/products/lustre/docs/GPLv2.pdf
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
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef __LUSTRE_LU_OBJECT_H
38 #define __LUSTRE_LU_OBJECT_H
41 #include <libcfs/libcfs.h>
42 #include <lustre/lustre_idl.h>
46 struct proc_dir_entry;
51 * lu_* data-types represent server-side entities shared by data and meta-data
56 * -# support for layering.
58 * Server side object is split into layers, one per device in the
59 * corresponding device stack. Individual layer is represented by struct
60 * lu_object. Compound layered object --- by struct lu_object_header. Most
61 * interface functions take lu_object as an argument and operate on the
62 * whole compound object. This decision was made due to the following
65 * - it's envisaged that lu_object will be used much more often than
68 * - we want lower (non-top) layers to be able to initiate operations
69 * on the whole object.
71 * Generic code supports layering more complex than simple stacking, e.g.,
72 * it is possible that at some layer object "spawns" multiple sub-objects
75 * -# fid-based identification.
77 * Compound object is uniquely identified by its fid. Objects are indexed
78 * by their fids (hash table is used for index).
80 * -# caching and life-cycle management.
82 * Object's life-time is controlled by reference counting. When reference
83 * count drops to 0, object is returned to cache. Cached objects still
84 * retain their identity (i.e., fid), and can be recovered from cache.
86 * Objects are kept in the global LRU list, and lu_site_purge() function
87 * can be used to reclaim given number of unused objects from the tail of
90 * -# avoiding recursion.
92 * Generic code tries to replace recursion through layers by iterations
93 * where possible. Additionally to the end of reducing stack consumption,
94 * data, when practically possible, are allocated through lu_context_key
95 * interface rather than on stack.
102 struct lu_object_header;
107 * Operations common for data and meta-data devices.
109 struct lu_device_operations {
111 * Allocate object for the given device (without lower-layer
112 * parts). This is called by lu_object_operations::loo_object_init()
113 * from the parent layer, and should setup at least lu_object::lo_dev
114 * and lu_object::lo_ops fields of resulting lu_object.
116 * Object creation protocol.
118 * Due to design goal of avoiding recursion, object creation (see
119 * lu_object_alloc()) is somewhat involved:
121 * - first, lu_device_operations::ldo_object_alloc() method of the
122 * top-level device in the stack is called. It should allocate top
123 * level object (including lu_object_header), but without any
124 * lower-layer sub-object(s).
126 * - then lu_object_alloc() sets fid in the header of newly created
129 * - then lu_object_operations::loo_object_init() is called. It has
130 * to allocate lower-layer object(s). To do this,
131 * lu_object_operations::loo_object_init() calls ldo_object_alloc()
132 * of the lower-layer device(s).
134 * - for all new objects allocated by
135 * lu_object_operations::loo_object_init() (and inserted into object
136 * stack), lu_object_operations::loo_object_init() is called again
137 * repeatedly, until no new objects are created.
139 * \post ergo(!IS_ERR(result), result->lo_dev == d &&
140 * result->lo_ops != NULL);
142 struct lu_object *(*ldo_object_alloc)(const struct lu_env *env,
143 const struct lu_object_header *h,
144 struct lu_device *d);
146 * process config specific for device.
148 int (*ldo_process_config)(const struct lu_env *env,
149 struct lu_device *, struct lustre_cfg *);
150 int (*ldo_recovery_complete)(const struct lu_env *,
154 * initialize local objects for device. this method called after layer has
155 * been initialized (after LCFG_SETUP stage) and before it starts serving
159 int (*ldo_prepare)(const struct lu_env *,
160 struct lu_device *parent,
161 struct lu_device *dev);
166 * For lu_object_conf flags
169 /* This is a new object to be allocated, or the file
170 * corresponding to the object does not exists. */
171 LOC_F_NEW = 0x00000001,
173 /* When find a dying object, just return -EAGAIN at once instead of
174 * blocking the thread. */
175 LOC_F_NOWAIT = 0x00000002,
179 * Object configuration, describing particulars of object being created. On
180 * server this is not used, as server objects are full identified by fid. On
181 * client configuration contains struct lustre_md.
183 struct lu_object_conf {
185 * Some hints for obj find and alloc.
187 loc_flags_t loc_flags;
191 * Type of "printer" function used by lu_object_operations::loo_object_print()
194 * Printer function is needed to provide some flexibility in (semi-)debugging
195 * output: possible implementations: printk, CDEBUG, sysfs/seq_file
197 typedef int (*lu_printer_t)(const struct lu_env *env,
198 void *cookie, const char *format, ...)
199 __attribute__ ((format (printf, 3, 4)));
202 * Operations specific for particular lu_object.
204 struct lu_object_operations {
207 * Allocate lower-layer parts of the object by calling
208 * lu_device_operations::ldo_object_alloc() of the corresponding
211 * This method is called once for each object inserted into object
212 * stack. It's responsibility of this method to insert lower-layer
213 * object(s) it create into appropriate places of object stack.
215 int (*loo_object_init)(const struct lu_env *env,
217 const struct lu_object_conf *conf);
219 * Called (in top-to-bottom order) during object allocation after all
220 * layers were allocated and initialized. Can be used to perform
221 * initialization depending on lower layers.
223 int (*loo_object_start)(const struct lu_env *env,
224 struct lu_object *o);
226 * Called before lu_object_operations::loo_object_free() to signal
227 * that object is being destroyed. Dual to
228 * lu_object_operations::loo_object_init().
230 void (*loo_object_delete)(const struct lu_env *env,
231 struct lu_object *o);
233 * Dual to lu_device_operations::ldo_object_alloc(). Called when
234 * object is removed from memory.
236 void (*loo_object_free)(const struct lu_env *env,
237 struct lu_object *o);
239 * Called when last active reference to the object is released (and
240 * object returns to the cache). This method is optional.
242 void (*loo_object_release)(const struct lu_env *env,
243 struct lu_object *o);
245 * Optional debugging helper. Print given object.
247 int (*loo_object_print)(const struct lu_env *env, void *cookie,
248 lu_printer_t p, const struct lu_object *o);
250 * Optional debugging method. Returns true iff method is internally
253 int (*loo_object_invariant)(const struct lu_object *o);
259 struct lu_device_type;
262 * Device: a layer in the server side abstraction stacking.
266 * reference count. This is incremented, in particular, on each object
267 * created at this layer.
269 * \todo XXX which means that atomic_t is probably too small.
273 * Pointer to device type. Never modified once set.
275 struct lu_device_type *ld_type;
277 * Operation vector for this device.
279 const struct lu_device_operations *ld_ops;
281 * Stack this device belongs to.
283 struct lu_site *ld_site;
284 struct proc_dir_entry *ld_proc_entry;
286 /** \todo XXX: temporary back pointer into obd. */
287 struct obd_device *ld_obd;
289 * A list of references to this object, for debugging.
291 struct lu_ref ld_reference;
293 * Link the device to the site.
295 struct list_head ld_linkage;
298 struct lu_device_type_operations;
301 * Tag bits for device type. They are used to distinguish certain groups of
305 /** this is meta-data device */
306 LU_DEVICE_MD = (1 << 0),
307 /** this is data device */
308 LU_DEVICE_DT = (1 << 1),
309 /** data device in the client stack */
310 LU_DEVICE_CL = (1 << 2)
316 struct lu_device_type {
318 * Tag bits. Taken from enum lu_device_tag. Never modified once set.
322 * Name of this class. Unique system-wide. Never modified once set.
326 * Operations for this type.
328 const struct lu_device_type_operations *ldt_ops;
330 * \todo XXX: temporary pointer to associated obd_type.
332 struct obd_type *ldt_obd_type;
334 * \todo XXX: temporary: context tags used by obd_*() calls.
338 * Number of existing device type instances.
340 atomic_t ldt_device_nr;
342 * Linkage into a global list of all device types.
344 * \see lu_device_types.
346 struct list_head ldt_linkage;
350 * Operations on a device type.
352 struct lu_device_type_operations {
354 * Allocate new device.
356 struct lu_device *(*ldto_device_alloc)(const struct lu_env *env,
357 struct lu_device_type *t,
358 struct lustre_cfg *lcfg);
360 * Free device. Dual to
361 * lu_device_type_operations::ldto_device_alloc(). Returns pointer to
362 * the next device in the stack.
364 struct lu_device *(*ldto_device_free)(const struct lu_env *,
368 * Initialize the devices after allocation
370 int (*ldto_device_init)(const struct lu_env *env,
371 struct lu_device *, const char *,
374 * Finalize device. Dual to
375 * lu_device_type_operations::ldto_device_init(). Returns pointer to
376 * the next device in the stack.
378 struct lu_device *(*ldto_device_fini)(const struct lu_env *env,
381 * Initialize device type. This is called on module load.
383 int (*ldto_init)(struct lu_device_type *t);
385 * Finalize device type. Dual to
386 * lu_device_type_operations::ldto_init(). Called on module unload.
388 void (*ldto_fini)(struct lu_device_type *t);
390 * Called when the first device is created.
392 void (*ldto_start)(struct lu_device_type *t);
394 * Called when number of devices drops to 0.
396 void (*ldto_stop)(struct lu_device_type *t);
399 static inline int lu_device_is_md(const struct lu_device *d)
401 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_MD);
405 * Common object attributes.
410 /** modification time in seconds since Epoch */
412 /** access time in seconds since Epoch */
414 /** change time in seconds since Epoch */
416 /** 512-byte blocks allocated to object */
418 /** permission bits and file type */
426 /** number of persistent references to this object */
428 /** blk bits of the object*/
430 /** blk size of the object*/
442 static inline void lu_attr_cpu_to_le(struct lu_attr *dst_attr,
443 struct lu_attr *src_attr)
445 dst_attr->la_size = cpu_to_le64(src_attr->la_size);
446 dst_attr->la_mtime = cpu_to_le64(src_attr->la_mtime);
447 dst_attr->la_atime = cpu_to_le64(src_attr->la_atime);
448 dst_attr->la_ctime = cpu_to_le64(src_attr->la_ctime);
449 dst_attr->la_blocks = cpu_to_le64(src_attr->la_blocks);
450 dst_attr->la_mode = cpu_to_le32(src_attr->la_mode);
451 dst_attr->la_uid = cpu_to_le32(src_attr->la_uid);
452 dst_attr->la_gid = cpu_to_le32(src_attr->la_gid);
453 dst_attr->la_flags = cpu_to_le32(src_attr->la_flags);
454 dst_attr->la_nlink = cpu_to_le32(src_attr->la_nlink);
455 dst_attr->la_blkbits = cpu_to_le32(src_attr->la_blkbits);
456 dst_attr->la_blksize = cpu_to_le32(src_attr->la_blksize);
457 dst_attr->la_rdev = cpu_to_le32(src_attr->la_rdev);
458 dst_attr->la_valid = cpu_to_le64(src_attr->la_valid);
461 static inline void lu_attr_le_to_cpu(struct lu_attr *dst_attr,
462 struct lu_attr *src_attr)
464 dst_attr->la_size = le64_to_cpu(src_attr->la_size);
465 dst_attr->la_mtime = le64_to_cpu(src_attr->la_mtime);
466 dst_attr->la_atime = le64_to_cpu(src_attr->la_atime);
467 dst_attr->la_ctime = le64_to_cpu(src_attr->la_ctime);
468 dst_attr->la_blocks = le64_to_cpu(src_attr->la_blocks);
469 dst_attr->la_mode = le32_to_cpu(src_attr->la_mode);
470 dst_attr->la_uid = le32_to_cpu(src_attr->la_uid);
471 dst_attr->la_gid = le32_to_cpu(src_attr->la_gid);
472 dst_attr->la_flags = le32_to_cpu(src_attr->la_flags);
473 dst_attr->la_nlink = le32_to_cpu(src_attr->la_nlink);
474 dst_attr->la_blkbits = le32_to_cpu(src_attr->la_blkbits);
475 dst_attr->la_blksize = le32_to_cpu(src_attr->la_blksize);
476 dst_attr->la_rdev = le32_to_cpu(src_attr->la_rdev);
477 dst_attr->la_valid = le64_to_cpu(src_attr->la_valid);
480 /** Bit-mask of valid attributes */
494 LA_BLKSIZE = 1 << 12,
495 LA_KILL_SUID = 1 << 13,
496 LA_KILL_SGID = 1 << 14,
500 * Layer in the layered object.
504 * Header for this object.
506 struct lu_object_header *lo_header;
508 * Device for this layer.
510 struct lu_device *lo_dev;
512 * Operations for this object.
514 const struct lu_object_operations *lo_ops;
516 * Linkage into list of all layers.
518 struct list_head lo_linkage;
520 * Link to the device, for debugging.
522 struct lu_ref_link lo_dev_ref;
525 enum lu_object_header_flags {
527 * Don't keep this object in cache. Object will be destroyed as soon
528 * as last reference to it is released. This flag cannot be cleared
531 LU_OBJECT_HEARD_BANSHEE = 0,
533 * Mark this object has already been taken out of cache.
535 LU_OBJECT_UNHASHED = 1,
538 enum lu_object_header_attr {
539 LOHA_EXISTS = 1 << 0,
540 LOHA_REMOTE = 1 << 1,
542 * UNIX file type is stored in S_IFMT bits.
544 LOHA_FT_START = 001 << 12, /**< S_IFIFO */
545 LOHA_FT_END = 017 << 12, /**< S_IFMT */
549 * "Compound" object, consisting of multiple layers.
551 * Compound object with given fid is unique with given lu_site.
553 * Note, that object does *not* necessary correspond to the real object in the
554 * persistent storage: object is an anchor for locking and method calling, so
555 * it is created for things like not-yet-existing child created by mkdir or
556 * create calls. lu_object_operations::loo_exists() can be used to check
557 * whether object is backed by persistent storage entity.
559 struct lu_object_header {
561 * Fid, uniquely identifying this object.
563 struct lu_fid loh_fid;
565 * Object flags from enum lu_object_header_flags. Set and checked
568 unsigned long loh_flags;
570 * Object reference count. Protected by lu_site::ls_guard.
574 * Common object attributes, cached for efficiency. From enum
575 * lu_object_header_attr.
579 * Linkage into per-site hash table. Protected by lu_site::ls_guard.
581 struct hlist_node loh_hash;
583 * Linkage into per-site LRU list. Protected by lu_site::ls_guard.
585 struct list_head loh_lru;
587 * Linkage into list of layers. Never modified once set (except lately
588 * during object destruction). No locking is necessary.
590 struct list_head loh_layers;
592 * A list of references to this object, for debugging.
594 struct lu_ref loh_reference;
599 struct lu_site_bkt_data {
601 * number of object in this bucket on the lsb_lru list.
605 * LRU list, updated on each access to object. Protected by
606 * bucket lock of lu_site::ls_obj_hash.
608 * "Cold" end of LRU is lu_site::ls_lru.next. Accessed object are
609 * moved to the lu_site::ls_lru.prev (this is due to the non-existence
610 * of list_for_each_entry_safe_reverse()).
612 struct list_head lsb_lru;
614 * Wait-queue signaled when an object in this site is ultimately
615 * destroyed (lu_object_free()). It is used by lu_object_find() to
616 * wait before re-trying when object in the process of destruction is
617 * found in the hash table.
619 * \see htable_lookup().
621 wait_queue_head_t lsb_marche_funebre;
629 LU_SS_CACHE_DEATH_RACE,
635 * lu_site is a "compartment" within which objects are unique, and LRU
636 * discipline is maintained.
638 * lu_site exists so that multiple layered stacks can co-exist in the same
641 * lu_site has the same relation to lu_device as lu_object_header to
648 struct cfs_hash *ls_obj_hash;
650 * index of bucket on hash table while purging
652 unsigned int ls_purge_start;
654 * Top-level device for this stack.
656 struct lu_device *ls_top_dev;
658 * Bottom-level device for this stack
660 struct lu_device *ls_bottom_dev;
662 * Linkage into global list of sites.
664 struct list_head ls_linkage;
666 * List for lu device for this site, protected
669 struct list_head ls_ld_linkage;
670 spinlock_t ls_ld_lock;
672 * Lock to serialize site purge.
674 struct mutex ls_purge_mutex;
678 struct lprocfs_stats *ls_stats;
680 * XXX: a hack! fld has to find md_site via site, remove when possible
682 struct seq_server_site *ld_seq_site;
684 * Pointer to the lu_target for this site.
686 struct lu_target *ls_tgt;
689 static inline struct lu_site_bkt_data *
690 lu_site_bkt_from_fid(struct lu_site *site, struct lu_fid *fid)
692 struct cfs_hash_bd bd;
694 cfs_hash_bd_get(site->ls_obj_hash, fid, &bd);
695 return cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
698 static inline struct seq_server_site *lu_site2seq(const struct lu_site *s)
700 return s->ld_seq_site;
704 * Constructors/destructors.
708 int lu_site_init (struct lu_site *s, struct lu_device *d);
709 void lu_site_fini (struct lu_site *s);
710 int lu_site_init_finish (struct lu_site *s);
711 void lu_stack_fini (const struct lu_env *env, struct lu_device *top);
712 void lu_device_get (struct lu_device *d);
713 void lu_device_put (struct lu_device *d);
714 int lu_device_init (struct lu_device *d, struct lu_device_type *t);
715 void lu_device_fini (struct lu_device *d);
716 int lu_object_header_init(struct lu_object_header *h);
717 void lu_object_header_fini(struct lu_object_header *h);
718 int lu_object_init (struct lu_object *o,
719 struct lu_object_header *h, struct lu_device *d);
720 void lu_object_fini (struct lu_object *o);
721 void lu_object_add_top (struct lu_object_header *h, struct lu_object *o);
722 void lu_object_add (struct lu_object *before, struct lu_object *o);
724 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d);
725 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d);
728 * Helpers to initialize and finalize device types.
731 int lu_device_type_init(struct lu_device_type *ldt);
732 void lu_device_type_fini(struct lu_device_type *ldt);
737 * Caching and reference counting.
742 * Acquire additional reference to the given object. This function is used to
743 * attain additional reference. To acquire initial reference use
746 static inline void lu_object_get(struct lu_object *o)
748 LASSERT(atomic_read(&o->lo_header->loh_ref) > 0);
749 atomic_inc(&o->lo_header->loh_ref);
753 * Return true of object will not be cached after last reference to it is
756 static inline int lu_object_is_dying(const struct lu_object_header *h)
758 return test_bit(LU_OBJECT_HEARD_BANSHEE, &h->loh_flags);
761 void lu_object_put(const struct lu_env *env, struct lu_object *o);
762 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o);
763 void lu_object_unhash(const struct lu_env *env, struct lu_object *o);
765 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr);
767 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
768 lu_printer_t printer);
769 struct lu_object *lu_object_find(const struct lu_env *env,
770 struct lu_device *dev, const struct lu_fid *f,
771 const struct lu_object_conf *conf);
772 struct lu_object *lu_object_find_at(const struct lu_env *env,
773 struct lu_device *dev,
774 const struct lu_fid *f,
775 const struct lu_object_conf *conf);
776 struct lu_object *lu_object_find_slice(const struct lu_env *env,
777 struct lu_device *dev,
778 const struct lu_fid *f,
779 const struct lu_object_conf *conf);
788 * First (topmost) sub-object of given compound object
790 static inline struct lu_object *lu_object_top(struct lu_object_header *h)
792 LASSERT(!list_empty(&h->loh_layers));
793 return container_of0(h->loh_layers.next, struct lu_object, lo_linkage);
797 * Next sub-object in the layering
799 static inline struct lu_object *lu_object_next(const struct lu_object *o)
801 return container_of0(o->lo_linkage.next, struct lu_object, lo_linkage);
805 * Pointer to the fid of this object.
807 static inline const struct lu_fid *lu_object_fid(const struct lu_object *o)
809 return &o->lo_header->loh_fid;
813 * return device operations vector for this object
815 static const inline struct lu_device_operations *
816 lu_object_ops(const struct lu_object *o)
818 return o->lo_dev->ld_ops;
822 * Given a compound object, find its slice, corresponding to the device type
825 struct lu_object *lu_object_locate(struct lu_object_header *h,
826 const struct lu_device_type *dtype);
829 * Printer function emitting messages through libcfs_debug_msg().
831 int lu_cdebug_printer(const struct lu_env *env,
832 void *cookie, const char *format, ...);
835 * Print object description followed by a user-supplied message.
837 #define LU_OBJECT_DEBUG(mask, env, object, format, ...) \
839 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
840 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
841 lu_object_print(env, &msgdata, lu_cdebug_printer, object);\
842 CDEBUG(mask, format , ## __VA_ARGS__); \
847 * Print short object description followed by a user-supplied message.
849 #define LU_OBJECT_HEADER(mask, env, object, format, ...) \
851 if (cfs_cdebug_show(mask, DEBUG_SUBSYSTEM)) { \
852 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, NULL); \
853 lu_object_header_print(env, &msgdata, lu_cdebug_printer,\
854 (object)->lo_header); \
855 lu_cdebug_printer(env, &msgdata, "\n"); \
856 CDEBUG(mask, format , ## __VA_ARGS__); \
860 void lu_object_print (const struct lu_env *env, void *cookie,
861 lu_printer_t printer, const struct lu_object *o);
862 void lu_object_header_print(const struct lu_env *env, void *cookie,
863 lu_printer_t printer,
864 const struct lu_object_header *hdr);
867 * Check object consistency.
869 int lu_object_invariant(const struct lu_object *o);
873 * Check whether object exists, no matter on local or remote storage.
874 * Note: LOHA_EXISTS will be set once some one created the object,
875 * and it does not needs to be committed to storage.
877 #define lu_object_exists(o) ((o)->lo_header->loh_attr & LOHA_EXISTS)
880 * Check whether object on the remote storage.
882 #define lu_object_remote(o) unlikely((o)->lo_header->loh_attr & LOHA_REMOTE)
884 static inline int lu_object_assert_exists(const struct lu_object *o)
886 return lu_object_exists(o);
889 static inline int lu_object_assert_not_exists(const struct lu_object *o)
891 return !lu_object_exists(o);
895 * Attr of this object.
897 static inline __u32 lu_object_attr(const struct lu_object *o)
899 LASSERT(lu_object_exists(o) != 0);
900 return o->lo_header->loh_attr;
903 static inline void lu_object_ref_add(struct lu_object *o,
907 lu_ref_add(&o->lo_header->loh_reference, scope, source);
910 static inline void lu_object_ref_add_at(struct lu_object *o,
911 struct lu_ref_link *link,
915 lu_ref_add_at(&o->lo_header->loh_reference, link, scope, source);
918 static inline void lu_object_ref_del(struct lu_object *o,
919 const char *scope, const void *source)
921 lu_ref_del(&o->lo_header->loh_reference, scope, source);
924 static inline void lu_object_ref_del_at(struct lu_object *o,
925 struct lu_ref_link *link,
926 const char *scope, const void *source)
928 lu_ref_del_at(&o->lo_header->loh_reference, link, scope, source);
931 /** input params, should be filled out by mdt */
935 /** count in bytes */
936 unsigned int rp_count;
937 /** number of pages */
938 unsigned int rp_npages;
939 /** requested attr */
941 /** pointers to pages */
942 struct page **rp_pages;
945 enum lu_xattr_flags {
946 LU_XATTR_REPLACE = (1 << 0),
947 LU_XATTR_CREATE = (1 << 1)
955 /** For lu_context health-checks */
956 enum lu_context_state {
964 * lu_context. Execution context for lu_object methods. Currently associated
967 * All lu_object methods, except device and device type methods (called during
968 * system initialization and shutdown) are executed "within" some
969 * lu_context. This means, that pointer to some "current" lu_context is passed
970 * as an argument to all methods.
972 * All service ptlrpc threads create lu_context as part of their
973 * initialization. It is possible to create "stand-alone" context for other
974 * execution environments (like system calls).
976 * lu_object methods mainly use lu_context through lu_context_key interface
977 * that allows each layer to associate arbitrary pieces of data with each
978 * context (see pthread_key_create(3) for similar interface).
980 * On a client, lu_context is bound to a thread, see cl_env_get().
982 * \see lu_context_key
986 * lu_context is used on the client side too. Yet we don't want to
987 * allocate values of server-side keys for the client contexts and
990 * To achieve this, set of tags in introduced. Contexts and keys are
991 * marked with tags. Key value are created only for context whose set
992 * of tags has non-empty intersection with one for key. Tags are taken
993 * from enum lu_context_tag.
996 enum lu_context_state lc_state;
998 * Pointer to the home service thread. NULL for other execution
1001 struct ptlrpc_thread *lc_thread;
1003 * Pointer to an array with key values. Internal implementation
1008 * Linkage into a list of all remembered contexts. Only
1009 * `non-transient' contexts, i.e., ones created for service threads
1012 struct list_head lc_remember;
1014 * Version counter used to skip calls to lu_context_refill() when no
1015 * keys were registered.
1017 unsigned lc_version;
1025 * lu_context_key interface. Similar to pthread_key.
1028 enum lu_context_tag {
1030 * Thread on md server
1032 LCT_MD_THREAD = 1 << 0,
1034 * Thread on dt server
1036 LCT_DT_THREAD = 1 << 1,
1038 * Context for transaction handle
1040 LCT_TX_HANDLE = 1 << 2,
1044 LCT_CL_THREAD = 1 << 3,
1046 * A per-request session on a server, and a per-system-call session on
1049 LCT_SESSION = 1 << 4,
1051 * A per-request data on OSP device
1053 LCT_OSP_THREAD = 1 << 5,
1057 LCT_MG_THREAD = 1 << 6,
1059 * Context for local operations
1063 * session for server thread
1065 LCT_SERVER_SESSION = 1 << 8,
1067 * Set when at least one of keys, having values in this context has
1068 * non-NULL lu_context_key::lct_exit() method. This is used to
1069 * optimize lu_context_exit() call.
1071 LCT_HAS_EXIT = 1 << 28,
1073 * Don't add references for modules creating key values in that context.
1074 * This is only for contexts used internally by lu_object framework.
1076 LCT_NOREF = 1 << 29,
1078 * Key is being prepared for retiring, don't create new values for it.
1080 LCT_QUIESCENT = 1 << 30,
1082 * Context should be remembered.
1084 LCT_REMEMBER = 1 << 31,
1086 * Contexts usable in cache shrinker thread.
1088 LCT_SHRINKER = LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD|LCT_NOREF
1092 * Key. Represents per-context value slot.
1094 * Keys are usually registered when module owning the key is initialized, and
1095 * de-registered when module is unloaded. Once key is registered, all new
1096 * contexts with matching tags, will get key value. "Old" contexts, already
1097 * initialized at the time of key registration, can be forced to get key value
1098 * by calling lu_context_refill().
1100 * Every key value is counted in lu_context_key::lct_used and acquires a
1101 * reference on an owning module. This means, that all key values have to be
1102 * destroyed before module can be unloaded. This is usually achieved by
1103 * stopping threads started by the module, that created contexts in their
1104 * entry functions. Situation is complicated by the threads shared by multiple
1105 * modules, like ptlrpcd daemon on a client. To work around this problem,
1106 * contexts, created in such threads, are `remembered' (see
1107 * LCT_REMEMBER)---i.e., added into a global list. When module is preparing
1108 * for unloading it does the following:
1110 * - marks its keys as `quiescent' (lu_context_tag::LCT_QUIESCENT)
1111 * preventing new key values from being allocated in the new contexts,
1114 * - scans a list of remembered contexts, destroying values of module
1115 * keys, thus releasing references to the module.
1117 * This is done by lu_context_key_quiesce(). If module is re-activated
1118 * before key has been de-registered, lu_context_key_revive() call clears
1119 * `quiescent' marker.
1121 * lu_context code doesn't provide any internal synchronization for these
1122 * activities---it's assumed that startup (including threads start-up) and
1123 * shutdown are serialized by some external means.
1127 struct lu_context_key {
1129 * Set of tags for which values of this key are to be instantiated.
1133 * Value constructor. This is called when new value is created for a
1134 * context. Returns pointer to new value of error pointer.
1136 void *(*lct_init)(const struct lu_context *ctx,
1137 struct lu_context_key *key);
1139 * Value destructor. Called when context with previously allocated
1140 * value of this slot is destroyed. \a data is a value that was returned
1141 * by a matching call to lu_context_key::lct_init().
1143 void (*lct_fini)(const struct lu_context *ctx,
1144 struct lu_context_key *key, void *data);
1146 * Optional method called on lu_context_exit() for all allocated
1147 * keys. Can be used by debugging code checking that locks are
1150 void (*lct_exit)(const struct lu_context *ctx,
1151 struct lu_context_key *key, void *data);
1153 * Internal implementation detail: index within lu_context::lc_value[]
1154 * reserved for this key.
1158 * Internal implementation detail: number of values created for this
1163 * Internal implementation detail: module for this key.
1165 struct module *lct_owner;
1167 * References to this key. For debugging.
1169 struct lu_ref lct_reference;
1172 #define LU_KEY_INIT(mod, type) \
1173 static void* mod##_key_init(const struct lu_context *ctx, \
1174 struct lu_context_key *key) \
1178 CLASSERT(PAGE_CACHE_SIZE >= sizeof (*value)); \
1180 OBD_ALLOC_PTR(value); \
1181 if (value == NULL) \
1182 value = ERR_PTR(-ENOMEM); \
1186 struct __##mod##__dummy_init {;} /* semicolon catcher */
1188 #define LU_KEY_FINI(mod, type) \
1189 static void mod##_key_fini(const struct lu_context *ctx, \
1190 struct lu_context_key *key, void* data) \
1192 type *info = data; \
1194 OBD_FREE_PTR(info); \
1196 struct __##mod##__dummy_fini {;} /* semicolon catcher */
1198 #define LU_KEY_INIT_FINI(mod, type) \
1199 LU_KEY_INIT(mod,type); \
1200 LU_KEY_FINI(mod,type)
1202 #define LU_CONTEXT_KEY_DEFINE(mod, tags) \
1203 struct lu_context_key mod##_thread_key = { \
1205 .lct_init = mod##_key_init, \
1206 .lct_fini = mod##_key_fini \
1209 #define LU_CONTEXT_KEY_INIT(key) \
1211 (key)->lct_owner = THIS_MODULE; \
1214 int lu_context_key_register(struct lu_context_key *key);
1215 void lu_context_key_degister(struct lu_context_key *key);
1216 void *lu_context_key_get (const struct lu_context *ctx,
1217 const struct lu_context_key *key);
1218 void lu_context_key_quiesce (struct lu_context_key *key);
1219 void lu_context_key_revive (struct lu_context_key *key);
1223 * LU_KEY_INIT_GENERIC() has to be a macro to correctly determine an
1227 #define LU_KEY_INIT_GENERIC(mod) \
1228 static void mod##_key_init_generic(struct lu_context_key *k, ...) \
1230 struct lu_context_key *key = k; \
1233 va_start(args, k); \
1235 LU_CONTEXT_KEY_INIT(key); \
1236 key = va_arg(args, struct lu_context_key *); \
1237 } while (key != NULL); \
1241 #define LU_TYPE_INIT(mod, ...) \
1242 LU_KEY_INIT_GENERIC(mod) \
1243 static int mod##_type_init(struct lu_device_type *t) \
1245 mod##_key_init_generic(__VA_ARGS__, NULL); \
1246 return lu_context_key_register_many(__VA_ARGS__, NULL); \
1248 struct __##mod##_dummy_type_init {;}
1250 #define LU_TYPE_FINI(mod, ...) \
1251 static void mod##_type_fini(struct lu_device_type *t) \
1253 lu_context_key_degister_many(__VA_ARGS__, NULL); \
1255 struct __##mod##_dummy_type_fini {;}
1257 #define LU_TYPE_START(mod, ...) \
1258 static void mod##_type_start(struct lu_device_type *t) \
1260 lu_context_key_revive_many(__VA_ARGS__, NULL); \
1262 struct __##mod##_dummy_type_start {;}
1264 #define LU_TYPE_STOP(mod, ...) \
1265 static void mod##_type_stop(struct lu_device_type *t) \
1267 lu_context_key_quiesce_many(__VA_ARGS__, NULL); \
1269 struct __##mod##_dummy_type_stop {;}
1273 #define LU_TYPE_INIT_FINI(mod, ...) \
1274 LU_TYPE_INIT(mod, __VA_ARGS__); \
1275 LU_TYPE_FINI(mod, __VA_ARGS__); \
1276 LU_TYPE_START(mod, __VA_ARGS__); \
1277 LU_TYPE_STOP(mod, __VA_ARGS__)
1279 int lu_context_init (struct lu_context *ctx, __u32 tags);
1280 void lu_context_fini (struct lu_context *ctx);
1281 void lu_context_enter (struct lu_context *ctx);
1282 void lu_context_exit (struct lu_context *ctx);
1283 int lu_context_refill(struct lu_context *ctx);
1286 * Helper functions to operate on multiple keys. These are used by the default
1287 * device type operations, defined by LU_TYPE_INIT_FINI().
1290 int lu_context_key_register_many(struct lu_context_key *k, ...);
1291 void lu_context_key_degister_many(struct lu_context_key *k, ...);
1292 void lu_context_key_revive_many (struct lu_context_key *k, ...);
1293 void lu_context_key_quiesce_many (struct lu_context_key *k, ...);
1296 * update/clear ctx/ses tags.
1298 void lu_context_tags_update(__u32 tags);
1299 void lu_context_tags_clear(__u32 tags);
1300 void lu_session_tags_update(__u32 tags);
1301 void lu_session_tags_clear(__u32 tags);
1308 * "Local" context, used to store data instead of stack.
1310 struct lu_context le_ctx;
1312 * "Session" context for per-request data.
1314 struct lu_context *le_ses;
1317 int lu_env_init (struct lu_env *env, __u32 tags);
1318 void lu_env_fini (struct lu_env *env);
1319 int lu_env_refill(struct lu_env *env);
1320 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags, __u32 stags);
1322 /** @} lu_context */
1325 * Output site statistical counters into a buffer. Suitable for
1326 * ll_rd_*()-style functions.
1328 int lu_site_stats_seq_print(const struct lu_site *s, struct seq_file *m);
1329 int lu_site_stats_print(const struct lu_site *s, char *page, int count);
1332 * Common name structure to be passed around for various name related methods.
1335 const char *ln_name;
1340 * Validate names (path components)
1342 * To be valid \a name must be non-empty, '\0' terminated of length \a
1343 * name_len, and not contain '/'. The maximum length of a name (before
1344 * say -ENAMETOOLONG will be returned) is really controlled by llite
1345 * and the server. We only check for something insane coming from bad
1346 * integer handling here.
1348 static inline bool lu_name_is_valid_2(const char *name, size_t name_len)
1350 return name != NULL &&
1352 name_len < INT_MAX &&
1353 name[name_len] == '\0' &&
1354 strlen(name) == name_len &&
1355 memchr(name, '/', name_len) == NULL;
1358 static inline bool lu_name_is_valid(const struct lu_name *ln)
1360 return lu_name_is_valid_2(ln->ln_name, ln->ln_namelen);
1363 #define DNAME "%.*s"
1365 (lu_name_is_valid(ln) ? (ln)->ln_namelen : 0), \
1366 (lu_name_is_valid(ln) ? (ln)->ln_name : "")
1369 * Common buffer structure to be passed around for various xattr_{s,g}et()
1377 #define DLUBUF "(%p %zu)"
1378 #define PLUBUF(buf) (buf)->lb_buf, (buf)->lb_len
1380 * One-time initializers, called at obdclass module initialization, not
1385 * Initialization of global lu_* data.
1387 int lu_global_init(void);
1390 * Dual to lu_global_init().
1392 void lu_global_fini(void);
1394 struct lu_kmem_descr {
1395 struct kmem_cache **ckd_cache;
1396 const char *ckd_name;
1397 const size_t ckd_size;
1400 int lu_kmem_init(struct lu_kmem_descr *caches);
1401 void lu_kmem_fini(struct lu_kmem_descr *caches);
1403 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
1404 const struct lu_fid *fid);
1405 struct lu_object *lu_object_anon(const struct lu_env *env,
1406 struct lu_device *dev,
1407 const struct lu_object_conf *conf);
1410 extern struct lu_buf LU_BUF_NULL;
1412 void lu_buf_free(struct lu_buf *buf);
1413 void lu_buf_alloc(struct lu_buf *buf, size_t size);
1414 void lu_buf_realloc(struct lu_buf *buf, size_t size);
1416 int lu_buf_check_and_grow(struct lu_buf *buf, size_t len);
1417 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, size_t len);
1419 extern __u32 lu_context_tags_default;
1420 extern __u32 lu_session_tags_default;
1422 static inline bool lu_device_is_cl(const struct lu_device *d)
1424 return d->ld_type->ldt_tags & LU_DEVICE_CL;
1427 static inline bool lu_object_is_cl(const struct lu_object *o)
1429 return lu_device_is_cl(o->lo_dev);
1433 #endif /* __LUSTRE_LU_OBJECT_H */