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 [sun.com URL with a
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) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 * Top-level entry points into osd module
39 * Author: Wang Di <wangdi@clusterfs.com>
40 * Author: Nikita Danilov <nikita@clusterfs.com>
43 #ifndef __LINUX_LUSTRE_IAM_H__
44 #define __LINUX_LUSTRE_IAM_H__
46 #include <linux/module.h>
47 #include <asm/unaligned.h>
49 #include "osd_dynlocks.h"
54 #define CLASSERT(cond) do {switch(42) {case (cond): case 0: break;}} while (0)
57 #define ergo(a, b) (!(a) || (b))
58 /* logical equivalence */
59 #define equi(a, b) (!!(a) == !!(b))
63 * Maximal number of non-leaf levels in htree. In the stock ldiskfs this
67 * XXX reduced back to 2 to make per-node locking work.
69 DX_MAX_TREE_HEIGHT = 5,
71 * Scratch keys used by generic code for temporaries.
75 * [0] reserved for assertions and as a staging area for
76 * record keys immediately used for key comparisons.
78 * [1] reserved for record key, stored during iteration over
79 * node records (see dx_node_check()).
81 * [2] reserved for leaf node operations.
83 * [3] reserved for index operations.
85 * [4] reserved for path->ip_ikey_target
90 * Maximal format name length.
100 * Various debugging levels.
105 * Following macros are defined in config.h and are tunable through
106 * appropriate configure switches (indicated below).
110 * Compile basic assertions in. You want this most of the time.
112 * --{enable,disable}-ldiskfs-assert (on by default).
114 #define LDISKFS_ASSERT (1)
117 * Compile heavier correctness checks in. You want this during development
120 * --{enable,disable}-ldiskfs-correctness (off by default).
122 #define LDISKFS_CORRECTNESS (1)
125 * Compile heavy invariant checking in. You want this early during development
126 * or when chasing a bug.
128 * --{enable,disable}-ldiskfs-invariant (off by default).
130 #define LDISKFS_INVARIANT (1)
133 #if defined(LDISKFS_ASSERT)
134 #define LDISKFS_ASSERT_ON (1)
136 #define LDISKFS_ASSERT_ON (0)
139 #if defined(LDISKFS_CORRECTNESS)
140 #define LDISKFS_CORRECTNESS_ON (1)
142 #define LDISKFS_CORRECTNESS_ON (0)
145 #if defined(LDISKFS_INVARIANT)
146 #define LDISKFS_INVARIANT_ON (1)
148 #define LDISKFS_INVARIANT_ON (0)
152 #if LDISKFS_ASSERT_ON
153 #define assert(test) J_ASSERT(test)
155 #define assert(test) ((void)(test))
159 #if LDISKFS_CORRECTNESS_ON
160 #define assert_corr(test) J_ASSERT(test)
161 #define do_corr(exp) exp
163 #define assert_corr(test) do {;} while (0)
164 #define do_corr(exp) do {;} while (0)
167 #if LDISKFS_INVARIANT_ON
168 #define assert_inv(test) J_ASSERT(test)
170 #define assert_inv(test) do {;} while (0)
174 * Entry within index tree node. Consists of a key immediately followed
175 * (without padding) by a pointer to the child node.
177 * Both key and pointer are of variable size, hence incomplete type.
181 struct iam_entry_compat {
187 * Incomplete type used to refer to keys in iam container.
189 * As key size can be different from container to container, iam has to use
190 * incomplete type. Clients cast pointer to iam_key to real key type and back.
195 * Incomplete type use to refer to the records stored in iam containers.
200 * Key in index node. Possibly compressed. Fixed size.
205 * Scalar type into which certain iam_key's can be uniquely mapped. Used to
206 * support interfaces like readdir(), where iteration over index has to be
209 typedef __u32 iam_ptr_t;
212 * Index node traversed during tree lookup.
215 struct buffer_head *bh; /* buffer holding node data */
216 struct iam_entry *entries; /* array of entries */
217 struct iam_entry *at; /* target entry, found by binary search */
218 iam_ptr_t leaf; /* (logical) offset of child node found by
220 iam_ptr_t curidx; /* (logical) offset of this node. Used to
221 * per-node locking to detect concurrent
223 unsigned int at_shifted:1; /* The "at" entry has moved to next
224 * because of shrinking index node
225 * for recycling empty leaf node. */
229 * Opaque entry in the leaf node.
234 struct iam_container;
237 /* leaf node reached by tree lookup */
239 struct iam_path *il_path;
240 struct buffer_head *il_bh;
241 struct iam_lentry *il_entries;
242 struct iam_lentry *il_at;
244 * Lock on a leaf node.
246 struct dynlock_handle *il_lock;
247 iam_ptr_t il_curidx; /* logical offset of leaf node. */
252 * Return values of ->lookup() operation from struct iam_leaf_operations.
256 * lookup found a record with the key requested
258 IAM_LOOKUP_EXACT = 0,
260 * lookup positioned leaf on some record
266 IAM_LOOKUP_EMPTY = 2,
268 * lookup positioned leaf before first record
270 IAM_LOOKUP_BEFORE = 3,
272 * Found hash may have a continuation in the next leaf.
274 IAM_LOOKUP_LAST = 0x100
278 * Format-specific container operations. These are called by generic iam code.
280 struct iam_operations {
282 * Returns pointer (in the same sense as pointer in index entry) to
285 __u32 (*id_root_ptr)(struct iam_container *c);
288 * Check validity and consistency of index node.
290 int (*id_node_check)(struct iam_path *path, struct iam_frame *frame);
292 * Copy some data from node header into frame. This is called when
293 * new node is loaded into frame.
295 int (*id_node_load)(struct iam_path *path, struct iam_frame *frame);
297 * Initialize new node (stored in @bh) that is going to be added into
300 int (*id_node_init)(struct iam_container *c,
301 struct buffer_head *bh, int root);
302 int (*id_node_read)(struct iam_container *c, iam_ptr_t ptr,
303 handle_t *h, struct buffer_head **bh);
305 * Key comparison functions. Returns -1, 0, +1.
307 int (*id_ikeycmp)(const struct iam_container *c,
308 const struct iam_ikey *k1,
309 const struct iam_ikey *k2);
311 * Modify root node when tree height increases.
313 struct iam_entry *(*id_root_inc)(struct iam_container *c,
314 struct iam_path *path,
315 struct iam_frame *frame);
317 struct iam_path_descr *(*id_ipd_alloc)(const struct iam_container *c,
319 void (*id_ipd_free)(struct iam_path_descr *ipd);
323 char id_name[DX_FMT_NAME_LEN];
327 * Another format-specific operation vector, consisting of methods to access
328 * leaf nodes. This is separated from struct iam_operations, because it is
329 * assumed that there will be many formats with different format of leaf
330 * nodes, yes the same struct iam_operations.
332 struct iam_leaf_operations {
338 * initialize just loaded leaf node.
340 int (*init)(struct iam_leaf *p);
344 void (*init_new)(struct iam_container *c, struct buffer_head *bh);
348 void (*fini)(struct iam_leaf *l);
350 * returns true iff leaf is positioned at the last entry.
352 int (*at_end)(const struct iam_leaf *l);
353 /* position leaf at the first entry */
354 void (*start)(struct iam_leaf *l);
355 /* more leaf to the next entry. */
356 void (*next)(struct iam_leaf *l);
358 * return key of current leaf record. This method may return
359 * either pointer to the key stored in node, or copy key into
360 * @k buffer supplied by caller and return pointer to this
361 * buffer. The latter approach is used when keys in nodes are
362 * not stored in plain form (e.g., htree doesn't store keys at
365 * Caller should assume that returned pointer is only valid
366 * while leaf node is pinned and locked.
368 struct iam_ikey *(*ikey)(const struct iam_leaf *l, struct iam_ikey *k);
369 struct iam_key *(*key)(const struct iam_leaf *l);
370 /* return pointer to entry body. Pointer is valid while
371 corresponding leaf node is locked and pinned. */
372 struct iam_rec *(*rec)(const struct iam_leaf *l);
374 void (*key_set)(struct iam_leaf *l, const struct iam_key *k);
375 void (*rec_set)(struct iam_leaf *l, const struct iam_rec *r);
376 void (*rec_get)(const struct iam_leaf *l, struct iam_rec *r);
378 int (*key_cmp)(const struct iam_leaf *l, const struct iam_key *k);
379 int (*key_eq)(const struct iam_leaf *l, const struct iam_key *k);
381 int (*rec_eq)(const struct iam_leaf *l, const struct iam_rec *r);
383 int (*key_size)(const struct iam_leaf *l);
385 * Search leaf @l for a record with key @k or for a place
386 * where such record is to be inserted.
388 * Scratch keys from @path can be used.
390 int (*lookup)(struct iam_leaf *l, const struct iam_key *k);
391 int (*ilookup)(struct iam_leaf *l, const struct iam_ikey *ik);
393 int (*can_add)(const struct iam_leaf *l,
394 const struct iam_key *k, const struct iam_rec *r);
398 void (*rec_add)(struct iam_leaf *l,
399 const struct iam_key *k, const struct iam_rec *r);
401 * remove rec for a leaf
403 void (*rec_del)(struct iam_leaf *l, int shift);
405 * split leaf node, moving some entries into @bh (the latter currently
406 * is assumed to be empty).
408 void (*split)(struct iam_leaf *l, struct buffer_head **bh,
413 int (*leaf_empty)(struct iam_leaf *l);
417 * Parameters, describing a flavor of iam container.
421 * Size of a key in this container, in bytes.
425 * Size of a key in index nodes, in bytes.
429 * Size of a pointer to the next level (stored in index nodes), in
434 * Size of a record (stored in leaf nodes), in bytes.
438 * Size of unused (by iam) space at the beginning of every non-root
439 * node, in bytes. Used for compatibility with ldiskfs.
443 * Size of unused (by iam) space at the beginning of root node, in
444 * bytes. Used for compatibility with ldiskfs.
448 struct iam_operations *id_ops;
449 struct iam_leaf_operations *id_leaf_ops;
453 IAM_IDLE_HEADER_MAGIC = 0x7903,
457 * Header structure to record idle blocks.
459 struct iam_idle_head {
461 __le16 iih_count; /* how many idle blocks in this head */
462 __le32 iih_next; /* next head for idle blocks */
467 * An instance of iam container.
469 struct iam_container {
471 * Underlying flat file. IO against this object is issued to
474 struct inode *ic_object;
478 struct buffer_head *ic_root_bh;
482 struct iam_descr *ic_descr;
484 * read-write lock protecting index consistency.
486 struct rw_semaphore ic_sem;
487 struct dynlock ic_tree_lock;
491 struct semaphore ic_idle_sem;
495 struct buffer_head *ic_idle_bh;
496 unsigned int ic_idle_failed:1; /* Idle block mechanism failed */
500 * description-specific part of iam_path. This is usually embedded into larger
503 struct iam_path_descr {
505 * Scratch-pad area for temporary keys.
507 struct iam_ikey *ipd_key_scratch[DX_SCRATCH_KEYS];
511 * Structure to keep track of a path drilled through htree.
517 struct iam_container *ip_container;
519 * Number of index levels minus one.
523 * Nodes that top-to-bottom traversal passed through.
525 struct iam_frame ip_frames[DX_MAX_TREE_HEIGHT];
527 * Last filled frame in ->ip_frames. Refers to the 'twig' node (one
528 * immediately above leaf).
530 struct iam_frame *ip_frame;
532 * Leaf node: a child of ->ip_frame.
534 struct iam_leaf ip_leaf;
538 const struct iam_key *ip_key_target;
539 const struct iam_ikey *ip_ikey_target;
541 * Description-specific data.
543 struct iam_path_descr *ip_data;
546 struct ldiskfs_dx_hash_info;
549 * Helper structure for legacy htrees.
551 struct iam_path_compat {
552 struct iam_path ipc_path;
553 struct iam_container ipc_container;
554 __u32 ipc_scratch[DX_SCRATCH_KEYS];
555 struct ldiskfs_dx_hash_info *ipc_hinfo;
556 struct qstr *ipc_qstr;
557 struct iam_path_descr ipc_descr;
558 struct ldiskfs_dx_hash_info ipc_hinfo_area;
561 #define const_max(p, q) ((p > q) ? p : q)
564 DX_MAX_IKEY_SIZE = 32, /* be generous */
566 * Hack to avoid dynamic allocation and freeing of ipd.
568 DX_IPD_MAX_SIZE = const_max(sizeof(struct iam_path_compat),
569 DX_MAX_IKEY_SIZE * DX_SCRATCH_KEYS +
570 sizeof(struct iam_path_descr))
574 * iam cursor (iterator) api.
578 * States of iterator state machine.
583 /* iterator is above particular record in the container */
585 /* iterator is positioned before record */
590 * Flags controlling iterator functionality.
594 * this iterator will move (iam_it_next() will be called on it)
596 IAM_IT_MOVE = (1 << 0),
598 * tree can be updated through this iterator.
600 IAM_IT_WRITE = (1 << 1)
606 * Immediately after call to iam_it_init() iterator is in "detached"
607 * (IAM_IT_DETACHED) state: it is associated with given parent container, but
608 * doesn't point to any particular record in this container.
610 * After successful call to iam_it_get() and until corresponding call to
611 * iam_it_put() iterator is in one of "active" states: IAM_IT_ATTACHED or
614 * Active iterator can move through records in a container (provided
615 * IAM_IT_MOVE permission) in a key order, can get record and key values as it
616 * passes over them, and can modify container (provided IAM_IT_WRITE
619 * Iteration may reach the end of container, at which point iterator switches
620 * into IAM_IT_DETACHED state.
622 * Concurrency: iterators are supposed to be local to thread. Interfaces below
623 * do no internal serialization of access to the iterator fields.
625 * When in non-detached state, iterator keeps some container nodes pinned in
626 * memory and locked (that locking may be implemented at the container
627 * granularity though). In particular, clients may assume that pointers to
628 * records and keys obtained through iterator interface as valid until
629 * iterator is detached (except that they may be invalidated by sub-sequent
630 * operations done through the same iterator).
633 struct iam_iterator {
635 * iterator flags, taken from enum iam_it_flags.
638 enum iam_it_state ii_state;
640 * path to the record. Valid in IAM_IT_ATTACHED, and IAM_IT_SKEWED
643 struct iam_path ii_path;
646 void iam_path_init(struct iam_path *path, struct iam_container *c,
647 struct iam_path_descr *pd);
648 void iam_path_fini(struct iam_path *path);
649 void iam_path_release(struct iam_path *path);
651 void iam_path_compat_init(struct iam_path_compat *path, struct inode *inode);
652 void iam_path_compat_fini(struct iam_path_compat *path);
654 struct iam_path_descr *iam_ipd_alloc(void *area, int keysize);
655 void iam_ipd_free(struct iam_path_descr *ipd);
657 int iam_it_init(struct iam_iterator *it, struct iam_container *c, __u32 flags,
658 struct iam_path_descr *pd);
659 void iam_it_fini(struct iam_iterator *it);
660 int iam_it_get(struct iam_iterator *it, const struct iam_key *k);
661 int iam_it_get_at(struct iam_iterator *it, const struct iam_key *k);
662 void iam_it_dup(struct iam_iterator *dst, const struct iam_iterator *src);
663 void iam_it_put(struct iam_iterator *it);
664 int iam_it_next(struct iam_iterator *it);
665 struct iam_rec *iam_it_rec_get(const struct iam_iterator *it);
666 int iam_it_rec_set(handle_t *h,
667 struct iam_iterator *it, const struct iam_rec *r);
668 struct iam_key *iam_it_key_get(const struct iam_iterator *it);
669 int iam_it_key_size(const struct iam_iterator *it);
670 int iam_it_rec_insert(handle_t *h, struct iam_iterator *it,
671 const struct iam_key *k, const struct iam_rec *r);
672 int iam_it_rec_delete(handle_t *h, struct iam_iterator *it);
674 typedef __u64 iam_pos_t;
676 iam_pos_t iam_it_store(const struct iam_iterator *it);
677 int iam_it_load(struct iam_iterator *it, iam_pos_t pos);
679 int iam_lookup(struct iam_container *c, const struct iam_key *k,
680 struct iam_rec *r, struct iam_path_descr *pd);
681 int iam_delete(handle_t *h, struct iam_container *c, const struct iam_key *k,
682 struct iam_path_descr *pd);
683 int iam_update(handle_t *h, struct iam_container *c, const struct iam_key *k,
684 const struct iam_rec *r, struct iam_path_descr *pd);
685 int iam_insert(handle_t *handle, struct iam_container *c,
686 const struct iam_key *k,
687 const struct iam_rec *r, struct iam_path_descr *pd);
689 * Initialize container @c.
691 int iam_container_init(struct iam_container *c,
692 struct iam_descr *descr, struct inode *inode);
694 * Finalize container @c, release all resources.
696 void iam_container_fini(struct iam_container *c);
699 * Determine container format.
701 int iam_container_setup(struct iam_container *c);
703 static inline struct iam_descr *iam_container_descr(struct iam_container *c)
708 static inline struct iam_descr *iam_path_descr(const struct iam_path *p)
710 return p->ip_container->ic_descr;
713 static inline struct inode *iam_path_obj(struct iam_path *p)
715 return p->ip_container->ic_object;
718 static inline void iam_ikeycpy(const struct iam_container *c,
719 struct iam_ikey *k1, const struct iam_ikey *k2)
721 memcpy(k1, k2, c->ic_descr->id_ikey_size);
724 static inline size_t iam_entry_size(struct iam_path *p)
726 return iam_path_descr(p)->id_ikey_size + iam_path_descr(p)->id_ptr_size;
729 static inline struct iam_entry *iam_entry_shift(struct iam_path *p,
730 struct iam_entry *entry,
734 return e + shift * iam_entry_size(p);
737 static inline struct iam_ikey *dx_get_ikey(struct iam_path *p,
738 struct iam_entry *entry,
739 struct iam_ikey *key)
741 return memcpy(key, entry, iam_path_descr(p)->id_ikey_size);
744 static inline struct iam_ikey *iam_ikey_at(struct iam_path *p,
745 struct iam_entry *entry)
747 return (struct iam_ikey *)entry;
750 static inline ptrdiff_t iam_entry_diff(struct iam_path *p,
751 struct iam_entry *e1,
752 struct iam_entry *e2)
756 diff = (void *)e1 - (void *)e2;
757 assert_corr(diff / iam_entry_size(p) * iam_entry_size(p) == diff);
758 return diff / iam_entry_size(p);
762 * Helper for the frequent case, where key was already placed into @k1 by
765 static inline void iam_ikeycpy0(const struct iam_container *c,
766 struct iam_ikey *k1, const struct iam_ikey *k2)
769 iam_ikeycpy(c, k1, k2);
772 static inline int iam_ikeycmp(const struct iam_container *c,
773 const struct iam_ikey *k1,
774 const struct iam_ikey *k2)
776 return c->ic_descr->id_ops->id_ikeycmp(c, k1, k2);
779 static inline void *iam_entry_off(struct iam_entry *entry, size_t off)
781 return (void *)((char *)entry + off);
788 static inline struct iam_path *iam_leaf_path(const struct iam_leaf *leaf)
790 return leaf->il_path;
793 static inline struct iam_container *
794 iam_leaf_container(const struct iam_leaf *leaf)
796 return iam_leaf_path(leaf)->ip_container;
799 static inline struct iam_descr *iam_leaf_descr(const struct iam_leaf *leaf)
801 return iam_leaf_container(leaf)->ic_descr;
804 static inline struct iam_leaf_operations *
805 iam_leaf_ops(const struct iam_leaf *leaf)
807 return iam_leaf_descr(leaf)->id_leaf_ops;
810 static inline void iam_reccpy(const struct iam_leaf *leaf,
811 struct iam_rec *rec_dst)
813 iam_leaf_ops(leaf)->rec_get(leaf, rec_dst);
816 /*XXX These stuff put here, just because they are used by iam.c */
817 static inline unsigned dx_get_block(struct iam_path *p, struct iam_entry *entry)
821 addr = iam_entry_off(entry, iam_path_descr(p)->id_ikey_size);
822 return le32_to_cpu(get_unaligned(addr));
825 static inline void dx_set_block(struct iam_path *p,
826 struct iam_entry *entry, unsigned value)
830 addr = iam_entry_off(entry, iam_path_descr(p)->id_ikey_size);
831 put_unaligned(cpu_to_le32(value), addr);
834 static inline void dx_set_ikey(struct iam_path *p, struct iam_entry *entry,
835 const struct iam_ikey *key)
837 iam_ikeycpy(p->ip_container, iam_entry_off(entry, 0), key);
853 struct dx_countlimit {
859 * dx_root_info is laid out so that if it should somehow get overlaid by a
860 * dirent the two low bits of the hash version will be zero. Therefore, the
861 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
865 struct fake_dirent dot;
867 struct fake_dirent dotdot;
871 __le32 reserved_zero;
873 u8 info_length; /* 8 */
878 struct {} entries[0];
883 struct fake_dirent fake;
884 struct {} entries[0];
888 static inline unsigned dx_get_count(struct iam_entry *entries)
890 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
893 static inline unsigned dx_get_limit(struct iam_entry *entries)
895 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
898 static inline void dx_set_count(struct iam_entry *entries, unsigned value)
900 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
903 static inline unsigned dx_node_limit(struct iam_path *p)
905 struct iam_descr *param = iam_path_descr(p);
906 unsigned entry_space = iam_path_obj(p)->i_sb->s_blocksize -
908 return entry_space / (param->id_ikey_size + param->id_ptr_size);
911 static inline unsigned dx_root_limit(struct iam_path *p)
913 struct iam_descr *param = iam_path_descr(p);
914 unsigned limit = iam_path_obj(p)->i_sb->s_blocksize -
916 limit /= (param->id_ikey_size + param->id_ptr_size);
917 if (limit == dx_node_limit(p))
923 static inline struct iam_entry *dx_get_entries(struct iam_path *path,
924 void *data, int root)
926 struct iam_descr *param = iam_path_descr(path);
927 return data + (root ? param->id_root_gap : param->id_node_gap);
931 static inline struct iam_entry *dx_node_get_entries(struct iam_path *path,
932 struct iam_frame *frame)
934 return dx_get_entries(path,
935 frame->bh->b_data, frame == path->ip_frames);
938 static inline struct iam_ikey *iam_path_ikey(const struct iam_path *path,
941 LASSERT(0 <= nr && nr < ARRAY_SIZE(path->ip_data->ipd_key_scratch));
942 return path->ip_data->ipd_key_scratch[nr];
945 static inline int iam_leaf_is_locked(const struct iam_leaf *leaf)
949 result = dynlock_is_locked(&iam_leaf_container(leaf)->ic_tree_lock,
956 static inline int iam_frame_is_locked(struct iam_path *path,
957 const struct iam_frame *frame)
961 result = dynlock_is_locked(&path->ip_container->ic_tree_lock,
968 int dx_lookup_lock(struct iam_path *path,
969 struct dynlock_handle **dl, enum dynlock_type lt);
971 void dx_insert_block(struct iam_path *path, struct iam_frame *frame,
972 u32 hash, u32 block);
973 int dx_index_is_compat(struct iam_path *path);
975 int ldiskfs_htree_next_block(struct inode *dir, __u32 hash,
976 struct iam_path *path, __u32 *start_hash);
978 int split_index_node(handle_t *handle, struct iam_path *path,
979 struct dynlock_handle **lh);
980 struct ldiskfs_dir_entry_2 *split_entry(struct inode *dir,
981 struct ldiskfs_dir_entry_2 *de,
982 unsigned long ino, mode_t mode,
983 const char *name, int namelen);
984 struct ldiskfs_dir_entry_2 *find_insertion_point(struct inode *dir,
985 struct buffer_head *bh,
986 const char *name, int namelen);
987 struct ldiskfs_dir_entry_2 *move_entries(struct inode *dir,
988 struct ldiskfs_dx_hash_info *hinfo,
989 struct buffer_head **bh1,
990 struct buffer_head **bh2,
993 extern struct iam_descr iam_htree_compat_param;
995 struct dynlock_handle *dx_lock_htree(struct inode *dir, unsigned long value,
996 enum dynlock_type lt);
997 void dx_unlock_htree(struct inode *dir, struct dynlock_handle *lh);
1002 void iam_container_write_lock(struct iam_container *c);
1003 void iam_container_write_unlock(struct iam_container *c);
1005 void iam_container_read_lock(struct iam_container *c);
1006 void iam_container_read_unlock(struct iam_container *c);
1008 int iam_index_next(struct iam_container *c, struct iam_path *p);
1009 int iam_read_leaf(struct iam_path *p);
1011 int iam_node_read(struct iam_container *c, iam_ptr_t ptr,
1012 handle_t *handle, struct buffer_head **bh);
1013 int iam_lvar_create(struct inode *obj,
1014 int keysize, int ptrsize, int recsize, handle_t *handle);
1017 #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
1021 #define dxtrace(command) command
1023 #define dxtrace(command)
1026 #define BH_DXLock 25
1027 #define DX_DEBUG (0)
1029 static struct iam_lock_stats {
1030 unsigned dls_bh_lock;
1031 unsigned dls_bh_busy;
1032 unsigned dls_bh_again;
1033 unsigned dls_bh_full_again;
1034 } iam_lock_stats = { 0, };
1035 #define DX_DEVAL(x) x
1040 static inline void iam_lock_bh(struct buffer_head volatile *bh)
1042 DX_DEVAL(iam_lock_stats.dls_bh_lock++);
1044 while (test_and_set_bit(BH_DXLock, &bh->b_state)) {
1045 DX_DEVAL(iam_lock_stats.dls_bh_busy++);
1046 while (test_bit(BH_DXLock, &bh->b_state))
1052 static inline void iam_unlock_bh(struct buffer_head *bh)
1055 smp_mb__before_clear_bit();
1056 clear_bit(BH_DXLock, &bh->b_state);
1061 void iam_insert_key(struct iam_path *path, struct iam_frame *frame,
1062 const struct iam_ikey *key, iam_ptr_t ptr);
1064 void iam_insert_key_lock(struct iam_path *path, struct iam_frame *frame,
1065 const struct iam_ikey *key, iam_ptr_t ptr);
1068 int iam_leaf_at_end(const struct iam_leaf *l);
1069 void iam_leaf_next(struct iam_leaf *folio);
1070 int iam_leaf_can_add(const struct iam_leaf *l,
1071 const struct iam_key *k, const struct iam_rec *r);
1073 struct iam_path *iam_leaf_path(const struct iam_leaf *leaf);
1074 struct iam_container *iam_leaf_container(const struct iam_leaf *leaf);
1075 struct iam_descr *iam_leaf_descr(const struct iam_leaf *leaf);
1076 struct iam_leaf_operations *iam_leaf_ops(const struct iam_leaf *leaf);
1079 int iam_node_read(struct iam_container *c, iam_ptr_t ptr,
1080 handle_t *h, struct buffer_head **bh);
1087 * Method called to recognize container format. Should return true iff
1088 * container @c conforms to this format. This method may do IO to read
1091 * If container is recognized, this method sets operation vectors
1092 * ->id_ops and ->id_leaf_ops in container description (c->ic_descr),
1093 * and fills other description fields.
1095 int (*if_guess)(struct iam_container *c);
1097 * Linkage into global list of container formats.
1099 cfs_list_t if_linkage;
1102 void iam_format_register(struct iam_format *fmt);
1103 int iam_root_limit(int rootgap, int blocksize, int size);
1105 void iam_lfix_format_init(void);
1106 void iam_lvar_format_init(void);
1107 void iam_htree_format_init(void);
1109 int iam_lfix_create(struct inode *obj,
1110 int keysize, int ptrsize, int recsize, handle_t *handle);
1111 struct iam_private_info;
1113 void ldiskfs_iam_release(struct file *filp, struct inode *inode);
1115 int iam_uapi_ioctl(struct inode * inode, struct file * filp, unsigned int cmd,
1119 #if LDISKFS_INVARIANT_ON
1120 extern int ldiskfs_check_dir_entry(const char *, struct inode *,
1121 struct ldiskfs_dir_entry_2 *,
1122 struct buffer_head *, unsigned long);
1124 static inline int ldiskfs_check_dir_entry(const char * function,
1126 struct ldiskfs_dir_entry_2 * de,
1127 struct buffer_head * bh,
1128 unsigned long offset)
1139 * User level API. Copy exists in lustre/lustre/tests/iam_ut.c
1142 struct iam_uapi_info {
1147 char iui_fmt_name[DX_FMT_NAME_LEN];
1150 struct iam_uapi_op {
1155 struct iam_uapi_it {
1156 struct iam_uapi_op iui_op;
1160 enum iam_ioctl_cmd {
1161 IAM_IOC_INIT = _IOW('i', 1, struct iam_uapi_info),
1162 IAM_IOC_GETINFO = _IOR('i', 2, struct iam_uapi_info),
1163 IAM_IOC_INSERT = _IOR('i', 3, struct iam_uapi_op),
1164 IAM_IOC_LOOKUP = _IOWR('i', 4, struct iam_uapi_op),
1165 IAM_IOC_DELETE = _IOR('i', 5, struct iam_uapi_op),
1166 IAM_IOC_IT_START = _IOR('i', 6, struct iam_uapi_it),
1167 IAM_IOC_IT_NEXT = _IOW('i', 7, struct iam_uapi_it),
1168 IAM_IOC_IT_STOP = _IOR('i', 8, struct iam_uapi_it),
1170 IAM_IOC_POLYMORPH = _IOR('i', 9, unsigned long)
1173 /* __LINUX_LUSTRE_IAM_H__ */