2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
27 * At the moment the locking rules of the TNC tree are quite simple and
28 * straightforward. We just have a mutex and lock it when we traverse the
29 * tree. If a znode is not in memory, we read it from flash while still having
33 #include <linux/crc32.h>
37 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
38 * @NAME_LESS: name corresponding to the first argument is less than second
39 * @NAME_MATCHES: names match
40 * @NAME_GREATER: name corresponding to the second argument is greater than
42 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
44 * These constants were introduce to improve readability.
54 * insert_old_idx - record an index node obsoleted since the last commit start.
55 * @c: UBIFS file-system description object
56 * @lnum: LEB number of obsoleted index node
57 * @offs: offset of obsoleted index node
59 * Returns %0 on success, and a negative error code on failure.
61 * For recovery, there must always be a complete intact version of the index on
62 * flash at all times. That is called the "old index". It is the index as at the
63 * time of the last successful commit. Many of the index nodes in the old index
64 * may be dirty, but they must not be erased until the next successful commit
65 * (at which point that index becomes the old index).
67 * That means that the garbage collection and the in-the-gaps method of
68 * committing must be able to determine if an index node is in the old index.
69 * Most of the old index nodes can be found by looking up the TNC using the
70 * 'lookup_znode()' function. However, some of the old index nodes may have
71 * been deleted from the current index or may have been changed so much that
72 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
73 * That is what this function does. The RB-tree is ordered by LEB number and
74 * offset because they uniquely identify the old index node.
76 static int insert_old_idx(struct ubifs_info
*c
, int lnum
, int offs
)
78 struct ubifs_old_idx
*old_idx
, *o
;
79 struct rb_node
**p
, *parent
= NULL
;
81 old_idx
= kmalloc(sizeof(struct ubifs_old_idx
), GFP_NOFS
);
82 if (unlikely(!old_idx
))
87 p
= &c
->old_idx
.rb_node
;
90 o
= rb_entry(parent
, struct ubifs_old_idx
, rb
);
93 else if (lnum
> o
->lnum
)
95 else if (offs
< o
->offs
)
97 else if (offs
> o
->offs
)
100 ubifs_err("old idx added twice!");
105 rb_link_node(&old_idx
->rb
, parent
, p
);
106 rb_insert_color(&old_idx
->rb
, &c
->old_idx
);
111 * insert_old_idx_znode - record a znode obsoleted since last commit start.
112 * @c: UBIFS file-system description object
113 * @znode: znode of obsoleted index node
115 * Returns %0 on success, and a negative error code on failure.
117 int insert_old_idx_znode(struct ubifs_info
*c
, struct ubifs_znode
*znode
)
120 struct ubifs_zbranch
*zbr
;
122 zbr
= &znode
->parent
->zbranch
[znode
->iip
];
124 return insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
127 return insert_old_idx(c
, c
->zroot
.lnum
,
133 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
134 * @c: UBIFS file-system description object
135 * @znode: znode of obsoleted index node
137 * Returns %0 on success, and a negative error code on failure.
139 static int ins_clr_old_idx_znode(struct ubifs_info
*c
,
140 struct ubifs_znode
*znode
)
145 struct ubifs_zbranch
*zbr
;
147 zbr
= &znode
->parent
->zbranch
[znode
->iip
];
149 err
= insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
158 err
= insert_old_idx(c
, c
->zroot
.lnum
, c
->zroot
.offs
);
169 * destroy_old_idx - destroy the old_idx RB-tree.
170 * @c: UBIFS file-system description object
172 * During start commit, the old_idx RB-tree is used to avoid overwriting index
173 * nodes that were in the index last commit but have since been deleted. This
174 * is necessary for recovery i.e. the old index must be kept intact until the
175 * new index is successfully written. The old-idx RB-tree is used for the
176 * in-the-gaps method of writing index nodes and is destroyed every commit.
178 void destroy_old_idx(struct ubifs_info
*c
)
180 struct rb_node
*this = c
->old_idx
.rb_node
;
181 struct ubifs_old_idx
*old_idx
;
185 this = this->rb_left
;
187 } else if (this->rb_right
) {
188 this = this->rb_right
;
191 old_idx
= rb_entry(this, struct ubifs_old_idx
, rb
);
192 this = rb_parent(this);
194 if (this->rb_left
== &old_idx
->rb
)
195 this->rb_left
= NULL
;
197 this->rb_right
= NULL
;
201 c
->old_idx
= RB_ROOT
;
205 * copy_znode - copy a dirty znode.
206 * @c: UBIFS file-system description object
207 * @znode: znode to copy
209 * A dirty znode being committed may not be changed, so it is copied.
211 static struct ubifs_znode
*copy_znode(struct ubifs_info
*c
,
212 struct ubifs_znode
*znode
)
214 struct ubifs_znode
*zn
;
216 zn
= kmalloc(c
->max_znode_sz
, GFP_NOFS
);
218 return ERR_PTR(-ENOMEM
);
220 memcpy(zn
, znode
, c
->max_znode_sz
);
222 __set_bit(DIRTY_ZNODE
, &zn
->flags
);
223 __clear_bit(COW_ZNODE
, &zn
->flags
);
225 ubifs_assert(!test_bit(OBSOLETE_ZNODE
, &znode
->flags
));
226 __set_bit(OBSOLETE_ZNODE
, &znode
->flags
);
228 if (znode
->level
!= 0) {
230 const int n
= zn
->child_cnt
;
232 /* The children now have new parent */
233 for (i
= 0; i
< n
; i
++) {
234 struct ubifs_zbranch
*zbr
= &zn
->zbranch
[i
];
237 zbr
->znode
->parent
= zn
;
241 atomic_long_inc(&c
->dirty_zn_cnt
);
246 * add_idx_dirt - add dirt due to a dirty znode.
247 * @c: UBIFS file-system description object
248 * @lnum: LEB number of index node
249 * @dirt: size of index node
251 * This function updates lprops dirty space and the new size of the index.
253 static int add_idx_dirt(struct ubifs_info
*c
, int lnum
, int dirt
)
255 c
->calc_idx_sz
-= ALIGN(dirt
, 8);
256 return ubifs_add_dirt(c
, lnum
, dirt
);
260 * dirty_cow_znode - ensure a znode is not being committed.
261 * @c: UBIFS file-system description object
262 * @zbr: branch of znode to check
264 * Returns dirtied znode on success or negative error code on failure.
266 static struct ubifs_znode
*dirty_cow_znode(struct ubifs_info
*c
,
267 struct ubifs_zbranch
*zbr
)
269 struct ubifs_znode
*znode
= zbr
->znode
;
270 struct ubifs_znode
*zn
;
273 if (!test_bit(COW_ZNODE
, &znode
->flags
)) {
274 /* znode is not being committed */
275 if (!test_and_set_bit(DIRTY_ZNODE
, &znode
->flags
)) {
276 atomic_long_inc(&c
->dirty_zn_cnt
);
277 atomic_long_dec(&c
->clean_zn_cnt
);
278 atomic_long_dec(&ubifs_clean_zn_cnt
);
279 err
= add_idx_dirt(c
, zbr
->lnum
, zbr
->len
);
286 zn
= copy_znode(c
, znode
);
291 err
= insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
294 err
= add_idx_dirt(c
, zbr
->lnum
, zbr
->len
);
309 * lnc_add - add a leaf node to the leaf node cache.
310 * @c: UBIFS file-system description object
311 * @zbr: zbranch of leaf node
314 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
315 * purpose of the leaf node cache is to save re-reading the same leaf node over
316 * and over again. Most things are cached by VFS, however the file system must
317 * cache directory entries for readdir and for resolving hash collisions. The
318 * present implementation of the leaf node cache is extremely simple, and
319 * allows for error returns that are not used but that may be needed if a more
320 * complex implementation is created.
322 * Note, this function does not add the @node object to LNC directly, but
323 * allocates a copy of the object and adds the copy to LNC. The reason for this
324 * is that @node has been allocated outside of the TNC subsystem and will be
325 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
326 * may be changed at any time, e.g. freed by the shrinker.
328 static int lnc_add(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
333 const struct ubifs_dent_node
*dent
= node
;
335 ubifs_assert(!zbr
->leaf
);
336 ubifs_assert(zbr
->len
!= 0);
337 ubifs_assert(is_hash_key(c
, &zbr
->key
));
339 err
= ubifs_validate_entry(c
, dent
);
342 dbg_dump_node(c
, dent
);
346 lnc_node
= kmalloc(zbr
->len
, GFP_NOFS
);
348 /* We don't have to have the cache, so no error */
351 memcpy(lnc_node
, node
, zbr
->len
);
352 zbr
->leaf
= lnc_node
;
357 * lnc_add_directly - add a leaf node to the leaf-node-cache.
358 * @c: UBIFS file-system description object
359 * @zbr: zbranch of leaf node
362 * This function is similar to 'lnc_add()', but it does not create a copy of
363 * @node but inserts @node to TNC directly.
365 static int lnc_add_directly(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
370 ubifs_assert(!zbr
->leaf
);
371 ubifs_assert(zbr
->len
!= 0);
373 err
= ubifs_validate_entry(c
, node
);
376 dbg_dump_node(c
, node
);
385 * lnc_free - remove a leaf node from the leaf node cache.
386 * @zbr: zbranch of leaf node
389 static void lnc_free(struct ubifs_zbranch
*zbr
)
398 * tnc_read_node_nm - read a "hashed" leaf node.
399 * @c: UBIFS file-system description object
400 * @zbr: key and position of the node
401 * @node: node is returned here
403 * This function reads a "hashed" node defined by @zbr from the leaf node cache
404 * (in it is there) or from the hash media, in which case the node is also
405 * added to LNC. Returns zero in case of success or a negative negative error
406 * code in case of failure.
408 static int tnc_read_node_nm(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
413 ubifs_assert(is_hash_key(c
, &zbr
->key
));
416 /* Read from the leaf node cache */
417 ubifs_assert(zbr
->len
!= 0);
418 memcpy(node
, zbr
->leaf
, zbr
->len
);
422 err
= ubifs_tnc_read_node(c
, zbr
, node
);
426 /* Add the node to the leaf node cache */
427 err
= lnc_add(c
, zbr
, node
);
432 * try_read_node - read a node if it is a node.
433 * @c: UBIFS file-system description object
434 * @buf: buffer to read to
436 * @len: node length (not aligned)
437 * @lnum: LEB number of node to read
438 * @offs: offset of node to read
440 * This function tries to read a node of known type and length, checks it and
441 * stores it in @buf. This function returns %1 if a node is present and %0 if
442 * a node is not present. A negative error code is returned for I/O errors.
443 * This function performs that same function as ubifs_read_node except that
444 * it does not require that there is actually a node present and instead
445 * the return code indicates if a node was read.
447 static int try_read_node(const struct ubifs_info
*c
, void *buf
, int type
,
448 int len
, int lnum
, int offs
)
451 struct ubifs_ch
*ch
= buf
;
452 uint32_t crc
, node_crc
;
454 dbg_io("LEB %d:%d, %s, length %d", lnum
, offs
, dbg_ntype(type
), len
);
456 err
= ubi_read(c
->ubi
, lnum
, buf
, offs
, len
);
458 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
459 type
, lnum
, offs
, err
);
463 if (le32_to_cpu(ch
->magic
) != UBIFS_NODE_MAGIC
)
466 if (ch
->node_type
!= type
)
469 node_len
= le32_to_cpu(ch
->len
);
473 if (type
== UBIFS_DATA_NODE
&& !c
->always_chk_crc
)
474 if (c
->no_chk_data_crc
)
477 crc
= crc32(UBIFS_CRC32_INIT
, buf
+ 8, node_len
- 8);
478 node_crc
= le32_to_cpu(ch
->crc
);
486 * fallible_read_node - try to read a leaf node.
487 * @c: UBIFS file-system description object
488 * @key: key of node to read
489 * @zbr: position of node
490 * @node: node returned
492 * This function tries to read a node and returns %1 if the node is read, %0
493 * if the node is not present, and a negative error code in the case of error.
495 static int fallible_read_node(struct ubifs_info
*c
, const union ubifs_key
*key
,
496 struct ubifs_zbranch
*zbr
, void *node
)
500 dbg_tnc("LEB %d:%d, key %s", zbr
->lnum
, zbr
->offs
, DBGKEY(key
));
502 ret
= try_read_node(c
, node
, key_type(c
, key
), zbr
->len
, zbr
->lnum
,
505 union ubifs_key node_key
;
506 struct ubifs_dent_node
*dent
= node
;
508 /* All nodes have key in the same place */
509 key_read(c
, &dent
->key
, &node_key
);
510 if (keys_cmp(c
, key
, &node_key
) != 0)
513 if (ret
== 0 && c
->replaying
)
514 dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
515 zbr
->lnum
, zbr
->offs
, zbr
->len
, DBGKEY(key
));
520 * matches_name - determine if a direntry or xattr entry matches a given name.
521 * @c: UBIFS file-system description object
522 * @zbr: zbranch of dent
525 * This function checks if xentry/direntry referred by zbranch @zbr matches name
526 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
527 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
528 * of failure, a negative error code is returned.
530 static int matches_name(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
531 const struct qstr
*nm
)
533 struct ubifs_dent_node
*dent
;
536 /* If possible, match against the dent in the leaf node cache */
538 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
542 err
= ubifs_tnc_read_node(c
, zbr
, dent
);
546 /* Add the node to the leaf node cache */
547 err
= lnc_add_directly(c
, zbr
, dent
);
553 nlen
= le16_to_cpu(dent
->nlen
);
554 err
= memcmp(dent
->name
, nm
->name
, min_t(int, nlen
, nm
->len
));
558 else if (nlen
< nm
->len
)
573 * get_znode - get a TNC znode that may not be loaded yet.
574 * @c: UBIFS file-system description object
575 * @znode: parent znode
576 * @n: znode branch slot number
578 * This function returns the znode or a negative error code.
580 static struct ubifs_znode
*get_znode(struct ubifs_info
*c
,
581 struct ubifs_znode
*znode
, int n
)
583 struct ubifs_zbranch
*zbr
;
585 zbr
= &znode
->zbranch
[n
];
589 znode
= ubifs_load_znode(c
, zbr
, znode
, n
);
594 * tnc_next - find next TNC entry.
595 * @c: UBIFS file-system description object
596 * @zn: znode is passed and returned here
597 * @n: znode branch slot number is passed and returned here
599 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
600 * no next entry, or a negative error code otherwise.
602 static int tnc_next(struct ubifs_info
*c
, struct ubifs_znode
**zn
, int *n
)
604 struct ubifs_znode
*znode
= *zn
;
608 if (nn
< znode
->child_cnt
) {
613 struct ubifs_znode
*zp
;
620 if (nn
< znode
->child_cnt
) {
621 znode
= get_znode(c
, znode
, nn
);
623 return PTR_ERR(znode
);
624 while (znode
->level
!= 0) {
625 znode
= get_znode(c
, znode
, 0);
627 return PTR_ERR(znode
);
639 * tnc_prev - find previous TNC entry.
640 * @c: UBIFS file-system description object
641 * @zn: znode is returned here
642 * @n: znode branch slot number is passed and returned here
644 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
645 * there is no next entry, or a negative error code otherwise.
647 static int tnc_prev(struct ubifs_info
*c
, struct ubifs_znode
**zn
, int *n
)
649 struct ubifs_znode
*znode
= *zn
;
657 struct ubifs_znode
*zp
;
665 znode
= get_znode(c
, znode
, nn
);
667 return PTR_ERR(znode
);
668 while (znode
->level
!= 0) {
669 nn
= znode
->child_cnt
- 1;
670 znode
= get_znode(c
, znode
, nn
);
672 return PTR_ERR(znode
);
674 nn
= znode
->child_cnt
- 1;
684 * resolve_collision - resolve a collision.
685 * @c: UBIFS file-system description object
686 * @key: key of a directory or extended attribute entry
687 * @zn: znode is returned here
688 * @n: zbranch number is passed and returned here
689 * @nm: name of the entry
691 * This function is called for "hashed" keys to make sure that the found key
692 * really corresponds to the looked up node (directory or extended attribute
693 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
694 * %0 is returned if @nm is not found and @zn and @n are set to the previous
695 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
696 * This means that @n may be set to %-1 if the leftmost key in @zn is the
697 * previous one. A negative error code is returned on failures.
699 static int resolve_collision(struct ubifs_info
*c
, const union ubifs_key
*key
,
700 struct ubifs_znode
**zn
, int *n
,
701 const struct qstr
*nm
)
705 err
= matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
706 if (unlikely(err
< 0))
708 if (err
== NAME_MATCHES
)
711 if (err
== NAME_GREATER
) {
714 err
= tnc_prev(c
, zn
, n
);
715 if (err
== -ENOENT
) {
716 ubifs_assert(*n
== 0);
722 if (keys_cmp(c
, &(*zn
)->zbranch
[*n
].key
, key
)) {
724 * We have found the branch after which we would
725 * like to insert, but inserting in this znode
726 * may still be wrong. Consider the following 3
727 * znodes, in the case where we are resolving a
728 * collision with Key2.
731 * ----------------------
732 * level 1 | Key0 | Key1 |
733 * -----------------------
735 * znode za | | znode zb
736 * ------------ ------------
737 * level 0 | Key0 | | Key2 |
738 * ------------ ------------
740 * The lookup finds Key2 in znode zb. Lets say
741 * there is no match and the name is greater so
742 * we look left. When we find Key0, we end up
743 * here. If we return now, we will insert into
744 * znode za at slot n = 1. But that is invalid
745 * according to the parent's keys. Key2 must
746 * be inserted into znode zb.
748 * Note, this problem is not relevant for the
749 * case when we go right, because
750 * 'tnc_insert()' would correct the parent key.
752 if (*n
== (*zn
)->child_cnt
- 1) {
753 err
= tnc_next(c
, zn
, n
);
755 /* Should be impossible */
761 ubifs_assert(*n
== 0);
766 err
= matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
769 if (err
== NAME_LESS
)
771 if (err
== NAME_MATCHES
)
773 ubifs_assert(err
== NAME_GREATER
);
777 struct ubifs_znode
*znode
= *zn
;
781 err
= tnc_next(c
, &znode
, &nn
);
786 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
788 err
= matches_name(c
, &znode
->zbranch
[nn
], nm
);
791 if (err
== NAME_GREATER
)
795 if (err
== NAME_MATCHES
)
797 ubifs_assert(err
== NAME_LESS
);
803 * fallible_matches_name - determine if a dent matches a given name.
804 * @c: UBIFS file-system description object
805 * @zbr: zbranch of dent
808 * This is a "fallible" version of 'matches_name()' function which does not
809 * panic if the direntry/xentry referred by @zbr does not exist on the media.
811 * This function checks if xentry/direntry referred by zbranch @zbr matches name
812 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
813 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
814 * if xentry/direntry referred by @zbr does not exist on the media. A negative
815 * error code is returned in case of failure.
817 static int fallible_matches_name(struct ubifs_info
*c
,
818 struct ubifs_zbranch
*zbr
,
819 const struct qstr
*nm
)
821 struct ubifs_dent_node
*dent
;
824 /* If possible, match against the dent in the leaf node cache */
826 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
830 err
= fallible_read_node(c
, &zbr
->key
, zbr
, dent
);
834 /* The node was not present */
838 ubifs_assert(err
== 1);
840 err
= lnc_add_directly(c
, zbr
, dent
);
846 nlen
= le16_to_cpu(dent
->nlen
);
847 err
= memcmp(dent
->name
, nm
->name
, min_t(int, nlen
, nm
->len
));
851 else if (nlen
< nm
->len
)
866 * fallible_resolve_collision - resolve a collision even if nodes are missing.
867 * @c: UBIFS file-system description object
869 * @zn: znode is returned here
870 * @n: branch number is passed and returned here
871 * @nm: name of directory entry
872 * @adding: indicates caller is adding a key to the TNC
874 * This is a "fallible" version of the 'resolve_collision()' function which
875 * does not panic if one of the nodes referred to by TNC does not exist on the
876 * media. This may happen when replaying the journal if a deleted node was
877 * Garbage-collected and the commit was not done. A branch that refers to a node
878 * that is not present is called a dangling branch. The following are the return
879 * codes for this function:
880 * o if @nm was found, %1 is returned and @zn and @n are set to the found
882 * o if we are @adding and @nm was not found, %0 is returned;
883 * o if we are not @adding and @nm was not found, but a dangling branch was
884 * found, then %1 is returned and @zn and @n are set to the dangling branch;
885 * o a negative error code is returned in case of failure.
887 static int fallible_resolve_collision(struct ubifs_info
*c
,
888 const union ubifs_key
*key
,
889 struct ubifs_znode
**zn
, int *n
,
890 const struct qstr
*nm
, int adding
)
892 struct ubifs_znode
*o_znode
= NULL
, *znode
= *zn
;
893 int uninitialized_var(o_n
), err
, cmp
, unsure
= 0, nn
= *n
;
895 cmp
= fallible_matches_name(c
, &znode
->zbranch
[nn
], nm
);
896 if (unlikely(cmp
< 0))
898 if (cmp
== NAME_MATCHES
)
900 if (cmp
== NOT_ON_MEDIA
) {
904 * We are unlucky and hit a dangling branch straight away.
905 * Now we do not really know where to go to find the needed
906 * branch - to the left or to the right. Well, let's try left.
910 unsure
= 1; /* Remove a dangling branch wherever it is */
912 if (cmp
== NAME_GREATER
|| unsure
) {
915 err
= tnc_prev(c
, zn
, n
);
916 if (err
== -ENOENT
) {
917 ubifs_assert(*n
== 0);
923 if (keys_cmp(c
, &(*zn
)->zbranch
[*n
].key
, key
)) {
924 /* See comments in 'resolve_collision()' */
925 if (*n
== (*zn
)->child_cnt
- 1) {
926 err
= tnc_next(c
, zn
, n
);
928 /* Should be impossible */
934 ubifs_assert(*n
== 0);
939 err
= fallible_matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
942 if (err
== NAME_MATCHES
)
944 if (err
== NOT_ON_MEDIA
) {
951 if (err
== NAME_LESS
)
958 if (cmp
== NAME_LESS
|| unsure
) {
963 err
= tnc_next(c
, &znode
, &nn
);
968 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
970 err
= fallible_matches_name(c
, &znode
->zbranch
[nn
], nm
);
973 if (err
== NAME_GREATER
)
977 if (err
== NAME_MATCHES
)
979 if (err
== NOT_ON_MEDIA
) {
986 /* Never match a dangling branch when adding */
987 if (adding
|| !o_znode
)
990 dbg_mnt("dangling match LEB %d:%d len %d %s",
991 o_znode
->zbranch
[o_n
].lnum
, o_znode
->zbranch
[o_n
].offs
,
992 o_znode
->zbranch
[o_n
].len
, DBGKEY(key
));
999 * matches_position - determine if a zbranch matches a given position.
1000 * @zbr: zbranch of dent
1001 * @lnum: LEB number of dent to match
1002 * @offs: offset of dent to match
1004 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1006 static int matches_position(struct ubifs_zbranch
*zbr
, int lnum
, int offs
)
1008 if (zbr
->lnum
== lnum
&& zbr
->offs
== offs
)
1015 * resolve_collision_directly - resolve a collision directly.
1016 * @c: UBIFS file-system description object
1017 * @key: key of directory entry
1018 * @zn: znode is passed and returned here
1019 * @n: zbranch number is passed and returned here
1020 * @lnum: LEB number of dent node to match
1021 * @offs: offset of dent node to match
1023 * This function is used for "hashed" keys to make sure the found directory or
1024 * extended attribute entry node is what was looked for. It is used when the
1025 * flash address of the right node is known (@lnum:@offs) which makes it much
1026 * easier to resolve collisions (no need to read entries and match full
1027 * names). This function returns %1 and sets @zn and @n if the collision is
1028 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1029 * previous directory entry. Otherwise a negative error code is returned.
1031 static int resolve_collision_directly(struct ubifs_info
*c
,
1032 const union ubifs_key
*key
,
1033 struct ubifs_znode
**zn
, int *n
,
1036 struct ubifs_znode
*znode
;
1041 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
))
1046 err
= tnc_prev(c
, &znode
, &nn
);
1051 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
1053 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
)) {
1064 err
= tnc_next(c
, &znode
, &nn
);
1069 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
1073 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
))
1079 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1080 * @c: UBIFS file-system description object
1081 * @znode: znode to dirty
1083 * If we do not have a unique key that resides in a znode, then we cannot
1084 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1085 * This function records the path back to the last dirty ancestor, and then
1086 * dirties the znodes on that path.
1088 static struct ubifs_znode
*dirty_cow_bottom_up(struct ubifs_info
*c
,
1089 struct ubifs_znode
*znode
)
1091 struct ubifs_znode
*zp
;
1092 int *path
= c
->bottom_up_buf
, p
= 0;
1094 ubifs_assert(c
->zroot
.znode
);
1095 ubifs_assert(znode
);
1096 if (c
->zroot
.znode
->level
> BOTTOM_UP_HEIGHT
) {
1097 kfree(c
->bottom_up_buf
);
1098 c
->bottom_up_buf
= kmalloc(c
->zroot
.znode
->level
* sizeof(int),
1100 if (!c
->bottom_up_buf
)
1101 return ERR_PTR(-ENOMEM
);
1102 path
= c
->bottom_up_buf
;
1104 if (c
->zroot
.znode
->level
) {
1105 /* Go up until parent is dirty */
1113 ubifs_assert(p
< c
->zroot
.znode
->level
);
1115 if (!zp
->cnext
&& ubifs_zn_dirty(znode
))
1121 /* Come back down, dirtying as we go */
1123 struct ubifs_zbranch
*zbr
;
1127 ubifs_assert(path
[p
- 1] >= 0);
1128 ubifs_assert(path
[p
- 1] < zp
->child_cnt
);
1129 zbr
= &zp
->zbranch
[path
[--p
]];
1130 znode
= dirty_cow_znode(c
, zbr
);
1132 ubifs_assert(znode
== c
->zroot
.znode
);
1133 znode
= dirty_cow_znode(c
, &c
->zroot
);
1135 if (IS_ERR(znode
) || !p
)
1137 ubifs_assert(path
[p
- 1] >= 0);
1138 ubifs_assert(path
[p
- 1] < znode
->child_cnt
);
1139 znode
= znode
->zbranch
[path
[p
- 1]].znode
;
1146 * ubifs_lookup_level0 - search for zero-level znode.
1147 * @c: UBIFS file-system description object
1148 * @key: key to lookup
1149 * @zn: znode is returned here
1150 * @n: znode branch slot number is returned here
1152 * This function looks up the TNC tree and search for zero-level znode which
1153 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1155 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1156 * is returned and slot number of the matched branch is stored in @n;
1157 * o not exact match, which means that zero-level znode does not contain
1158 * @key, then %0 is returned and slot number of the closed branch is stored
1160 * o @key is so small that it is even less than the lowest key of the
1161 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1163 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1164 * function reads corresponding indexing nodes and inserts them to TNC. In
1165 * case of failure, a negative error code is returned.
1167 int ubifs_lookup_level0(struct ubifs_info
*c
, const union ubifs_key
*key
,
1168 struct ubifs_znode
**zn
, int *n
)
1171 struct ubifs_znode
*znode
;
1172 unsigned long time
= get_seconds();
1174 dbg_tnc("search key %s", DBGKEY(key
));
1176 znode
= c
->zroot
.znode
;
1177 if (unlikely(!znode
)) {
1178 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
1180 return PTR_ERR(znode
);
1186 struct ubifs_zbranch
*zbr
;
1188 exact
= ubifs_search_zbranch(c
, znode
, key
, n
);
1190 if (znode
->level
== 0)
1195 zbr
= &znode
->zbranch
[*n
];
1203 /* znode is not in TNC cache, load it from the media */
1204 znode
= ubifs_load_znode(c
, zbr
, znode
, *n
);
1206 return PTR_ERR(znode
);
1210 if (exact
|| !is_hash_key(c
, key
) || *n
!= -1) {
1211 dbg_tnc("found %d, lvl %d, n %d", exact
, znode
->level
, *n
);
1216 * Here is a tricky place. We have not found the key and this is a
1217 * "hashed" key, which may collide. The rest of the code deals with
1218 * situations like this:
1222 * | 3 | 5 | | 6 | 7 | (x)
1224 * Or more a complex example:
1228 * | 1 | 3 | | 5 | 8 |
1230 * | 5 | 5 | | 6 | 7 | (x)
1232 * In the examples, if we are looking for key "5", we may reach nodes
1233 * marked with "(x)". In this case what we have do is to look at the
1234 * left and see if there is "5" key there. If there is, we have to
1237 * Note, this whole situation is possible because we allow to have
1238 * elements which are equivalent to the next key in the parent in the
1239 * children of current znode. For example, this happens if we split a
1240 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1244 * | 3 | 5 | | 5 | 6 | 7 |
1246 * And this becomes what is at the first "picture" after key "5" marked
1247 * with "^" is removed. What could be done is we could prohibit
1248 * splitting in the middle of the colliding sequence. Also, when
1249 * removing the leftmost key, we would have to correct the key of the
1250 * parent node, which would introduce additional complications. Namely,
1251 * if we changed the the leftmost key of the parent znode, the garbage
1252 * collector would be unable to find it (GC is doing this when GC'ing
1253 * indexing LEBs). Although we already have an additional RB-tree where
1254 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1255 * after the commit. But anyway, this does not look easy to implement
1256 * so we did not try this.
1258 err
= tnc_prev(c
, &znode
, n
);
1259 if (err
== -ENOENT
) {
1260 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1264 if (unlikely(err
< 0))
1266 if (keys_cmp(c
, key
, &znode
->zbranch
[*n
].key
)) {
1267 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1272 dbg_tnc("found 1, lvl %d, n %d", znode
->level
, *n
);
1278 * lookup_level0_dirty - search for zero-level znode dirtying.
1279 * @c: UBIFS file-system description object
1280 * @key: key to lookup
1281 * @zn: znode is returned here
1282 * @n: znode branch slot number is returned here
1284 * This function looks up the TNC tree and search for zero-level znode which
1285 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1287 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1288 * is returned and slot number of the matched branch is stored in @n;
1289 * o not exact match, which means that zero-level znode does not contain @key
1290 * then %0 is returned and slot number of the closed branch is stored in
1292 * o @key is so small that it is even less than the lowest key of the
1293 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1295 * Additionally all znodes in the path from the root to the located zero-level
1296 * znode are marked as dirty.
1298 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1299 * function reads corresponding indexing nodes and inserts them to TNC. In
1300 * case of failure, a negative error code is returned.
1302 static int lookup_level0_dirty(struct ubifs_info
*c
, const union ubifs_key
*key
,
1303 struct ubifs_znode
**zn
, int *n
)
1306 struct ubifs_znode
*znode
;
1307 unsigned long time
= get_seconds();
1309 dbg_tnc("search and dirty key %s", DBGKEY(key
));
1311 znode
= c
->zroot
.znode
;
1312 if (unlikely(!znode
)) {
1313 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
1315 return PTR_ERR(znode
);
1318 znode
= dirty_cow_znode(c
, &c
->zroot
);
1320 return PTR_ERR(znode
);
1325 struct ubifs_zbranch
*zbr
;
1327 exact
= ubifs_search_zbranch(c
, znode
, key
, n
);
1329 if (znode
->level
== 0)
1334 zbr
= &znode
->zbranch
[*n
];
1338 znode
= dirty_cow_znode(c
, zbr
);
1340 return PTR_ERR(znode
);
1344 /* znode is not in TNC cache, load it from the media */
1345 znode
= ubifs_load_znode(c
, zbr
, znode
, *n
);
1347 return PTR_ERR(znode
);
1348 znode
= dirty_cow_znode(c
, zbr
);
1350 return PTR_ERR(znode
);
1354 if (exact
|| !is_hash_key(c
, key
) || *n
!= -1) {
1355 dbg_tnc("found %d, lvl %d, n %d", exact
, znode
->level
, *n
);
1360 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1363 err
= tnc_prev(c
, &znode
, n
);
1364 if (err
== -ENOENT
) {
1366 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1369 if (unlikely(err
< 0))
1371 if (keys_cmp(c
, key
, &znode
->zbranch
[*n
].key
)) {
1373 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1377 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
1378 znode
= dirty_cow_bottom_up(c
, znode
);
1380 return PTR_ERR(znode
);
1383 dbg_tnc("found 1, lvl %d, n %d", znode
->level
, *n
);
1389 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1390 * @c: UBIFS file-system description object
1392 * @gc_seq1: garbage collection sequence number
1394 * This function determines if @lnum may have been garbage collected since
1395 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1398 static int maybe_leb_gced(struct ubifs_info
*c
, int lnum
, int gc_seq1
)
1400 int gc_seq2
, gced_lnum
;
1402 gced_lnum
= c
->gced_lnum
;
1404 gc_seq2
= c
->gc_seq
;
1405 /* Same seq means no GC */
1406 if (gc_seq1
== gc_seq2
)
1408 /* Different by more than 1 means we don't know */
1409 if (gc_seq1
+ 1 != gc_seq2
)
1412 * We have seen the sequence number has increased by 1. Now we need to
1413 * be sure we read the right LEB number, so read it again.
1416 if (gced_lnum
!= c
->gced_lnum
)
1418 /* Finally we can check lnum */
1419 if (gced_lnum
== lnum
)
1425 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1426 * @c: UBIFS file-system description object
1427 * @key: node key to lookup
1428 * @node: the node is returned here
1429 * @lnum: LEB number is returned here
1430 * @offs: offset is returned here
1432 * This function look up and reads node with key @key. The caller has to make
1433 * sure the @node buffer is large enough to fit the node. Returns zero in case
1434 * of success, %-ENOENT if the node was not found, and a negative error code in
1435 * case of failure. The node location can be returned in @lnum and @offs.
1437 int ubifs_tnc_locate(struct ubifs_info
*c
, const union ubifs_key
*key
,
1438 void *node
, int *lnum
, int *offs
)
1440 int found
, n
, err
, safely
= 0, gc_seq1
;
1441 struct ubifs_znode
*znode
;
1442 struct ubifs_zbranch zbr
, *zt
;
1445 mutex_lock(&c
->tnc_mutex
);
1446 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
1450 } else if (found
< 0) {
1454 zt
= &znode
->zbranch
[n
];
1459 if (is_hash_key(c
, key
)) {
1461 * In this case the leaf node cache gets used, so we pass the
1462 * address of the zbranch and keep the mutex locked
1464 err
= tnc_read_node_nm(c
, zt
, node
);
1468 err
= ubifs_tnc_read_node(c
, zt
, node
);
1471 /* Drop the TNC mutex prematurely and race with garbage collection */
1472 zbr
= znode
->zbranch
[n
];
1473 gc_seq1
= c
->gc_seq
;
1474 mutex_unlock(&c
->tnc_mutex
);
1476 if (ubifs_get_wbuf(c
, zbr
.lnum
)) {
1477 /* We do not GC journal heads */
1478 err
= ubifs_tnc_read_node(c
, &zbr
, node
);
1482 err
= fallible_read_node(c
, key
, &zbr
, node
);
1483 if (err
<= 0 || maybe_leb_gced(c
, zbr
.lnum
, gc_seq1
)) {
1485 * The node may have been GC'ed out from under us so try again
1486 * while keeping the TNC mutex locked.
1494 mutex_unlock(&c
->tnc_mutex
);
1499 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1500 * @c: UBIFS file-system description object
1501 * @bu: bulk-read parameters and results
1503 * Lookup consecutive data node keys for the same inode that reside
1504 * consecutively in the same LEB.
1506 int ubifs_tnc_get_bu_keys(struct ubifs_info
*c
, struct bu_info
*bu
)
1508 int n
, err
= 0, lnum
= -1, uninitialized_var(offs
);
1509 int uninitialized_var(len
);
1510 unsigned int block
= key_block(c
, &bu
->key
);
1511 struct ubifs_znode
*znode
;
1517 mutex_lock(&c
->tnc_mutex
);
1518 /* Find first key */
1519 err
= ubifs_lookup_level0(c
, &bu
->key
, &znode
, &n
);
1524 len
= znode
->zbranch
[n
].len
;
1525 /* The buffer must be big enough for at least 1 node */
1526 if (len
> bu
->buf_len
) {
1531 bu
->zbranch
[bu
->cnt
++] = znode
->zbranch
[n
];
1533 lnum
= znode
->zbranch
[n
].lnum
;
1534 offs
= ALIGN(znode
->zbranch
[n
].offs
+ len
, 8);
1537 struct ubifs_zbranch
*zbr
;
1538 union ubifs_key
*key
;
1539 unsigned int next_block
;
1542 err
= tnc_next(c
, &znode
, &n
);
1545 zbr
= &znode
->zbranch
[n
];
1547 /* See if there is another data key for this file */
1548 if (key_inum(c
, key
) != key_inum(c
, &bu
->key
) ||
1549 key_type(c
, key
) != UBIFS_DATA_KEY
) {
1554 /* First key found */
1556 offs
= ALIGN(zbr
->offs
+ zbr
->len
, 8);
1558 if (len
> bu
->buf_len
) {
1564 * The data nodes must be in consecutive positions in
1567 if (zbr
->lnum
!= lnum
|| zbr
->offs
!= offs
)
1569 offs
+= ALIGN(zbr
->len
, 8);
1570 len
= ALIGN(len
, 8) + zbr
->len
;
1571 /* Must not exceed buffer length */
1572 if (len
> bu
->buf_len
)
1575 /* Allow for holes */
1576 next_block
= key_block(c
, key
);
1577 bu
->blk_cnt
+= (next_block
- block
- 1);
1578 if (bu
->blk_cnt
>= UBIFS_MAX_BULK_READ
)
1582 bu
->zbranch
[bu
->cnt
++] = *zbr
;
1584 /* See if we have room for more */
1585 if (bu
->cnt
>= UBIFS_MAX_BULK_READ
)
1587 if (bu
->blk_cnt
>= UBIFS_MAX_BULK_READ
)
1591 if (err
== -ENOENT
) {
1595 bu
->gc_seq
= c
->gc_seq
;
1596 mutex_unlock(&c
->tnc_mutex
);
1600 * An enormous hole could cause bulk-read to encompass too many
1601 * page cache pages, so limit the number here.
1603 if (bu
->blk_cnt
> UBIFS_MAX_BULK_READ
)
1604 bu
->blk_cnt
= UBIFS_MAX_BULK_READ
;
1606 * Ensure that bulk-read covers a whole number of page cache
1609 if (UBIFS_BLOCKS_PER_PAGE
== 1 ||
1610 !(bu
->blk_cnt
& (UBIFS_BLOCKS_PER_PAGE
- 1)))
1613 /* At the end of file we can round up */
1614 bu
->blk_cnt
+= UBIFS_BLOCKS_PER_PAGE
- 1;
1617 /* Exclude data nodes that do not make up a whole page cache page */
1618 block
= key_block(c
, &bu
->key
) + bu
->blk_cnt
;
1619 block
&= ~(UBIFS_BLOCKS_PER_PAGE
- 1);
1621 if (key_block(c
, &bu
->zbranch
[bu
->cnt
- 1].key
) < block
)
1629 * read_wbuf - bulk-read from a LEB with a wbuf.
1630 * @wbuf: wbuf that may overlap the read
1631 * @buf: buffer into which to read
1633 * @lnum: LEB number from which to read
1634 * @offs: offset from which to read
1636 * This functions returns %0 on success or a negative error code on failure.
1638 static int read_wbuf(struct ubifs_wbuf
*wbuf
, void *buf
, int len
, int lnum
,
1641 const struct ubifs_info
*c
= wbuf
->c
;
1644 dbg_io("LEB %d:%d, length %d", lnum
, offs
, len
);
1645 ubifs_assert(wbuf
&& lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
1646 ubifs_assert(!(offs
& 7) && offs
< c
->leb_size
);
1647 ubifs_assert(offs
+ len
<= c
->leb_size
);
1649 spin_lock(&wbuf
->lock
);
1650 overlap
= (lnum
== wbuf
->lnum
&& offs
+ len
> wbuf
->offs
);
1652 /* We may safely unlock the write-buffer and read the data */
1653 spin_unlock(&wbuf
->lock
);
1654 return ubi_read(c
->ubi
, lnum
, buf
, offs
, len
);
1657 /* Don't read under wbuf */
1658 rlen
= wbuf
->offs
- offs
;
1662 /* Copy the rest from the write-buffer */
1663 memcpy(buf
+ rlen
, wbuf
->buf
+ offs
+ rlen
- wbuf
->offs
, len
- rlen
);
1664 spin_unlock(&wbuf
->lock
);
1667 /* Read everything that goes before write-buffer */
1668 return ubi_read(c
->ubi
, lnum
, buf
, offs
, rlen
);
1674 * validate_data_node - validate data nodes for bulk-read.
1675 * @c: UBIFS file-system description object
1676 * @buf: buffer containing data node to validate
1677 * @zbr: zbranch of data node to validate
1679 * This functions returns %0 on success or a negative error code on failure.
1681 static int validate_data_node(struct ubifs_info
*c
, void *buf
,
1682 struct ubifs_zbranch
*zbr
)
1684 union ubifs_key key1
;
1685 struct ubifs_ch
*ch
= buf
;
1688 if (ch
->node_type
!= UBIFS_DATA_NODE
) {
1689 ubifs_err("bad node type (%d but expected %d)",
1690 ch
->node_type
, UBIFS_DATA_NODE
);
1694 err
= ubifs_check_node(c
, buf
, zbr
->lnum
, zbr
->offs
, 0, 0);
1696 ubifs_err("expected node type %d", UBIFS_DATA_NODE
);
1700 len
= le32_to_cpu(ch
->len
);
1701 if (len
!= zbr
->len
) {
1702 ubifs_err("bad node length %d, expected %d", len
, zbr
->len
);
1706 /* Make sure the key of the read node is correct */
1707 key_read(c
, buf
+ UBIFS_KEY_OFFSET
, &key1
);
1708 if (!keys_eq(c
, &zbr
->key
, &key1
)) {
1709 ubifs_err("bad key in node at LEB %d:%d",
1710 zbr
->lnum
, zbr
->offs
);
1711 dbg_tnc("looked for key %s found node's key %s",
1712 DBGKEY(&zbr
->key
), DBGKEY1(&key1
));
1721 ubifs_err("bad node at LEB %d:%d", zbr
->lnum
, zbr
->offs
);
1722 dbg_dump_node(c
, buf
);
1728 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1729 * @c: UBIFS file-system description object
1730 * @bu: bulk-read parameters and results
1732 * This functions reads and validates the data nodes that were identified by the
1733 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1734 * -EAGAIN to indicate a race with GC, or another negative error code on
1737 int ubifs_tnc_bulk_read(struct ubifs_info
*c
, struct bu_info
*bu
)
1739 int lnum
= bu
->zbranch
[0].lnum
, offs
= bu
->zbranch
[0].offs
, len
, err
, i
;
1740 struct ubifs_wbuf
*wbuf
;
1743 len
= bu
->zbranch
[bu
->cnt
- 1].offs
;
1744 len
+= bu
->zbranch
[bu
->cnt
- 1].len
- offs
;
1745 if (len
> bu
->buf_len
) {
1746 ubifs_err("buffer too small %d vs %d", bu
->buf_len
, len
);
1751 wbuf
= ubifs_get_wbuf(c
, lnum
);
1753 err
= read_wbuf(wbuf
, bu
->buf
, len
, lnum
, offs
);
1755 err
= ubi_read(c
->ubi
, lnum
, bu
->buf
, offs
, len
);
1757 /* Check for a race with GC */
1758 if (maybe_leb_gced(c
, lnum
, bu
->gc_seq
))
1761 if (err
&& err
!= -EBADMSG
) {
1762 ubifs_err("failed to read from LEB %d:%d, error %d",
1765 dbg_tnc("key %s", DBGKEY(&bu
->key
));
1769 /* Validate the nodes read */
1771 for (i
= 0; i
< bu
->cnt
; i
++) {
1772 err
= validate_data_node(c
, buf
, &bu
->zbranch
[i
]);
1775 buf
= buf
+ ALIGN(bu
->zbranch
[i
].len
, 8);
1782 * do_lookup_nm- look up a "hashed" node.
1783 * @c: UBIFS file-system description object
1784 * @key: node key to lookup
1785 * @node: the node is returned here
1788 * This function look up and reads a node which contains name hash in the key.
1789 * Since the hash may have collisions, there may be many nodes with the same
1790 * key, so we have to sequentially look to all of them until the needed one is
1791 * found. This function returns zero in case of success, %-ENOENT if the node
1792 * was not found, and a negative error code in case of failure.
1794 static int do_lookup_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
1795 void *node
, const struct qstr
*nm
)
1798 struct ubifs_znode
*znode
;
1800 dbg_tnc("name '%.*s' key %s", nm
->len
, nm
->name
, DBGKEY(key
));
1801 mutex_lock(&c
->tnc_mutex
);
1802 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
1806 } else if (found
< 0) {
1811 ubifs_assert(n
>= 0);
1813 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
1814 dbg_tnc("rc returned %d, znode %p, n %d", err
, znode
, n
);
1815 if (unlikely(err
< 0))
1822 err
= tnc_read_node_nm(c
, &znode
->zbranch
[n
], node
);
1825 mutex_unlock(&c
->tnc_mutex
);
1830 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1831 * @c: UBIFS file-system description object
1832 * @key: node key to lookup
1833 * @node: the node is returned here
1836 * This function look up and reads a node which contains name hash in the key.
1837 * Since the hash may have collisions, there may be many nodes with the same
1838 * key, so we have to sequentially look to all of them until the needed one is
1839 * found. This function returns zero in case of success, %-ENOENT if the node
1840 * was not found, and a negative error code in case of failure.
1842 int ubifs_tnc_lookup_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
1843 void *node
, const struct qstr
*nm
)
1846 const struct ubifs_dent_node
*dent
= node
;
1849 * We assume that in most of the cases there are no name collisions and
1850 * 'ubifs_tnc_lookup()' returns us the right direntry.
1852 err
= ubifs_tnc_lookup(c
, key
, node
);
1856 len
= le16_to_cpu(dent
->nlen
);
1857 if (nm
->len
== len
&& !memcmp(dent
->name
, nm
->name
, len
))
1861 * Unluckily, there are hash collisions and we have to iterate over
1862 * them look at each direntry with colliding name hash sequentially.
1864 return do_lookup_nm(c
, key
, node
, nm
);
1868 * correct_parent_keys - correct parent znodes' keys.
1869 * @c: UBIFS file-system description object
1870 * @znode: znode to correct parent znodes for
1872 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1873 * zbranch changes, keys of parent znodes have to be corrected. This helper
1874 * function is called in such situations and corrects the keys if needed.
1876 static void correct_parent_keys(const struct ubifs_info
*c
,
1877 struct ubifs_znode
*znode
)
1879 union ubifs_key
*key
, *key1
;
1881 ubifs_assert(znode
->parent
);
1882 ubifs_assert(znode
->iip
== 0);
1884 key
= &znode
->zbranch
[0].key
;
1885 key1
= &znode
->parent
->zbranch
[0].key
;
1887 while (keys_cmp(c
, key
, key1
) < 0) {
1888 key_copy(c
, key
, key1
);
1889 znode
= znode
->parent
;
1891 if (!znode
->parent
|| znode
->iip
)
1893 key1
= &znode
->parent
->zbranch
[0].key
;
1898 * insert_zbranch - insert a zbranch into a znode.
1899 * @znode: znode into which to insert
1900 * @zbr: zbranch to insert
1901 * @n: slot number to insert to
1903 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1904 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1905 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1906 * slot, zbranches starting from @n have to be moved right.
1908 static void insert_zbranch(struct ubifs_znode
*znode
,
1909 const struct ubifs_zbranch
*zbr
, int n
)
1913 ubifs_assert(ubifs_zn_dirty(znode
));
1916 for (i
= znode
->child_cnt
; i
> n
; i
--) {
1917 znode
->zbranch
[i
] = znode
->zbranch
[i
- 1];
1918 if (znode
->zbranch
[i
].znode
)
1919 znode
->zbranch
[i
].znode
->iip
= i
;
1922 zbr
->znode
->iip
= n
;
1924 for (i
= znode
->child_cnt
; i
> n
; i
--)
1925 znode
->zbranch
[i
] = znode
->zbranch
[i
- 1];
1927 znode
->zbranch
[n
] = *zbr
;
1928 znode
->child_cnt
+= 1;
1931 * After inserting at slot zero, the lower bound of the key range of
1932 * this znode may have changed. If this znode is subsequently split
1933 * then the upper bound of the key range may change, and furthermore
1934 * it could change to be lower than the original lower bound. If that
1935 * happens, then it will no longer be possible to find this znode in the
1936 * TNC using the key from the index node on flash. That is bad because
1937 * if it is not found, we will assume it is obsolete and may overwrite
1938 * it. Then if there is an unclean unmount, we will start using the
1939 * old index which will be broken.
1941 * So we first mark znodes that have insertions at slot zero, and then
1942 * if they are split we add their lnum/offs to the old_idx tree.
1949 * tnc_insert - insert a node into TNC.
1950 * @c: UBIFS file-system description object
1951 * @znode: znode to insert into
1952 * @zbr: branch to insert
1953 * @n: slot number to insert new zbranch to
1955 * This function inserts a new node described by @zbr into znode @znode. If
1956 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1957 * are splat as well if needed. Returns zero in case of success or a negative
1958 * error code in case of failure.
1960 static int tnc_insert(struct ubifs_info
*c
, struct ubifs_znode
*znode
,
1961 struct ubifs_zbranch
*zbr
, int n
)
1963 struct ubifs_znode
*zn
, *zi
, *zp
;
1964 int i
, keep
, move
, appending
= 0;
1965 union ubifs_key
*key
= &zbr
->key
, *key1
;
1967 ubifs_assert(n
>= 0 && n
<= c
->fanout
);
1969 /* Implement naive insert for now */
1972 if (znode
->child_cnt
< c
->fanout
) {
1973 ubifs_assert(n
!= c
->fanout
);
1974 dbg_tnc("inserted at %d level %d, key %s", n
, znode
->level
,
1977 insert_zbranch(znode
, zbr
, n
);
1979 /* Ensure parent's key is correct */
1980 if (n
== 0 && zp
&& znode
->iip
== 0)
1981 correct_parent_keys(c
, znode
);
1987 * Unfortunately, @znode does not have more empty slots and we have to
1990 dbg_tnc("splitting level %d, key %s", znode
->level
, DBGKEY(key
));
1994 * We can no longer be sure of finding this znode by key, so we
1995 * record it in the old_idx tree.
1997 ins_clr_old_idx_znode(c
, znode
);
1999 zn
= kzalloc(c
->max_znode_sz
, GFP_NOFS
);
2003 zn
->level
= znode
->level
;
2005 /* Decide where to split */
2006 if (znode
->level
== 0 && key_type(c
, key
) == UBIFS_DATA_KEY
) {
2007 /* Try not to split consecutive data keys */
2008 if (n
== c
->fanout
) {
2009 key1
= &znode
->zbranch
[n
- 1].key
;
2010 if (key_inum(c
, key1
) == key_inum(c
, key
) &&
2011 key_type(c
, key1
) == UBIFS_DATA_KEY
)
2015 } else if (appending
&& n
!= c
->fanout
) {
2016 /* Try not to split consecutive data keys */
2019 if (n
>= (c
->fanout
+ 1) / 2) {
2020 key1
= &znode
->zbranch
[0].key
;
2021 if (key_inum(c
, key1
) == key_inum(c
, key
) &&
2022 key_type(c
, key1
) == UBIFS_DATA_KEY
) {
2023 key1
= &znode
->zbranch
[n
].key
;
2024 if (key_inum(c
, key1
) != key_inum(c
, key
) ||
2025 key_type(c
, key1
) != UBIFS_DATA_KEY
) {
2027 move
= c
->fanout
- keep
;
2039 keep
= (c
->fanout
+ 1) / 2;
2040 move
= c
->fanout
- keep
;
2044 * Although we don't at present, we could look at the neighbors and see
2045 * if we can move some zbranches there.
2049 /* Insert into existing znode */
2054 /* Insert into new znode */
2059 zbr
->znode
->parent
= zn
;
2064 __set_bit(DIRTY_ZNODE
, &zn
->flags
);
2065 atomic_long_inc(&c
->dirty_zn_cnt
);
2067 zn
->child_cnt
= move
;
2068 znode
->child_cnt
= keep
;
2070 dbg_tnc("moving %d, keeping %d", move
, keep
);
2073 for (i
= 0; i
< move
; i
++) {
2074 zn
->zbranch
[i
] = znode
->zbranch
[keep
+ i
];
2077 if (zn
->zbranch
[i
].znode
) {
2078 zn
->zbranch
[i
].znode
->parent
= zn
;
2079 zn
->zbranch
[i
].znode
->iip
= i
;
2083 /* Insert new key and branch */
2084 dbg_tnc("inserting at %d level %d, key %s", n
, zn
->level
, DBGKEY(key
));
2086 insert_zbranch(zi
, zbr
, n
);
2088 /* Insert new znode (produced by spitting) into the parent */
2090 if (n
== 0 && zi
== znode
&& znode
->iip
== 0)
2091 correct_parent_keys(c
, znode
);
2093 /* Locate insertion point */
2096 /* Tail recursion */
2097 zbr
->key
= zn
->zbranch
[0].key
;
2107 /* We have to split root znode */
2108 dbg_tnc("creating new zroot at level %d", znode
->level
+ 1);
2110 zi
= kzalloc(c
->max_znode_sz
, GFP_NOFS
);
2115 zi
->level
= znode
->level
+ 1;
2117 __set_bit(DIRTY_ZNODE
, &zi
->flags
);
2118 atomic_long_inc(&c
->dirty_zn_cnt
);
2120 zi
->zbranch
[0].key
= znode
->zbranch
[0].key
;
2121 zi
->zbranch
[0].znode
= znode
;
2122 zi
->zbranch
[0].lnum
= c
->zroot
.lnum
;
2123 zi
->zbranch
[0].offs
= c
->zroot
.offs
;
2124 zi
->zbranch
[0].len
= c
->zroot
.len
;
2125 zi
->zbranch
[1].key
= zn
->zbranch
[0].key
;
2126 zi
->zbranch
[1].znode
= zn
;
2131 c
->zroot
.znode
= zi
;
2142 * ubifs_tnc_add - add a node to TNC.
2143 * @c: UBIFS file-system description object
2145 * @lnum: LEB number of node
2146 * @offs: node offset
2149 * This function adds a node with key @key to TNC. The node may be new or it may
2150 * obsolete some existing one. Returns %0 on success or negative error code on
2153 int ubifs_tnc_add(struct ubifs_info
*c
, const union ubifs_key
*key
, int lnum
,
2156 int found
, n
, err
= 0;
2157 struct ubifs_znode
*znode
;
2159 mutex_lock(&c
->tnc_mutex
);
2160 dbg_tnc("%d:%d, len %d, key %s", lnum
, offs
, len
, DBGKEY(key
));
2161 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2163 struct ubifs_zbranch zbr
;
2169 key_copy(c
, key
, &zbr
.key
);
2170 err
= tnc_insert(c
, znode
, &zbr
, n
+ 1);
2171 } else if (found
== 1) {
2172 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2175 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2182 err
= dbg_check_tnc(c
, 0);
2183 mutex_unlock(&c
->tnc_mutex
);
2189 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2190 * @c: UBIFS file-system description object
2192 * @old_lnum: LEB number of old node
2193 * @old_offs: old node offset
2194 * @lnum: LEB number of node
2195 * @offs: node offset
2198 * This function replaces a node with key @key in the TNC only if the old node
2199 * is found. This function is called by garbage collection when node are moved.
2200 * Returns %0 on success or negative error code on failure.
2202 int ubifs_tnc_replace(struct ubifs_info
*c
, const union ubifs_key
*key
,
2203 int old_lnum
, int old_offs
, int lnum
, int offs
, int len
)
2205 int found
, n
, err
= 0;
2206 struct ubifs_znode
*znode
;
2208 mutex_lock(&c
->tnc_mutex
);
2209 dbg_tnc("old LEB %d:%d, new LEB %d:%d, len %d, key %s", old_lnum
,
2210 old_offs
, lnum
, offs
, len
, DBGKEY(key
));
2211 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2218 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2221 if (zbr
->lnum
== old_lnum
&& zbr
->offs
== old_offs
) {
2223 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2230 } else if (is_hash_key(c
, key
)) {
2231 found
= resolve_collision_directly(c
, key
, &znode
, &n
,
2232 old_lnum
, old_offs
);
2233 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2234 found
, znode
, n
, old_lnum
, old_offs
);
2241 /* Ensure the znode is dirtied */
2242 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2243 znode
= dirty_cow_bottom_up(c
,
2245 if (IS_ERR(znode
)) {
2246 err
= PTR_ERR(znode
);
2250 zbr
= &znode
->zbranch
[n
];
2252 err
= ubifs_add_dirt(c
, zbr
->lnum
,
2264 err
= ubifs_add_dirt(c
, lnum
, len
);
2267 err
= dbg_check_tnc(c
, 0);
2270 mutex_unlock(&c
->tnc_mutex
);
2275 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2276 * @c: UBIFS file-system description object
2278 * @lnum: LEB number of node
2279 * @offs: node offset
2283 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2284 * may have collisions, like directory entry keys.
2286 int ubifs_tnc_add_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
2287 int lnum
, int offs
, int len
, const struct qstr
*nm
)
2289 int found
, n
, err
= 0;
2290 struct ubifs_znode
*znode
;
2292 mutex_lock(&c
->tnc_mutex
);
2293 dbg_tnc("LEB %d:%d, name '%.*s', key %s", lnum
, offs
, nm
->len
, nm
->name
,
2295 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2303 found
= fallible_resolve_collision(c
, key
, &znode
, &n
,
2306 found
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2307 dbg_tnc("rc returned %d, znode %p, n %d", found
, znode
, n
);
2313 /* Ensure the znode is dirtied */
2314 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2315 znode
= dirty_cow_bottom_up(c
, znode
);
2316 if (IS_ERR(znode
)) {
2317 err
= PTR_ERR(znode
);
2323 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2326 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2335 struct ubifs_zbranch zbr
;
2341 key_copy(c
, key
, &zbr
.key
);
2342 err
= tnc_insert(c
, znode
, &zbr
, n
+ 1);
2347 * We did not find it in the index so there may be a
2348 * dangling branch still in the index. So we remove it
2349 * by passing 'ubifs_tnc_remove_nm()' the same key but
2350 * an unmatchable name.
2352 struct qstr noname
= { .len
= 0, .name
= "" };
2354 err
= dbg_check_tnc(c
, 0);
2355 mutex_unlock(&c
->tnc_mutex
);
2358 return ubifs_tnc_remove_nm(c
, key
, &noname
);
2364 err
= dbg_check_tnc(c
, 0);
2365 mutex_unlock(&c
->tnc_mutex
);
2370 * tnc_delete - delete a znode form TNC.
2371 * @c: UBIFS file-system description object
2372 * @znode: znode to delete from
2373 * @n: zbranch slot number to delete
2375 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2376 * case of success and a negative error code in case of failure.
2378 static int tnc_delete(struct ubifs_info
*c
, struct ubifs_znode
*znode
, int n
)
2380 struct ubifs_zbranch
*zbr
;
2381 struct ubifs_znode
*zp
;
2384 /* Delete without merge for now */
2385 ubifs_assert(znode
->level
== 0);
2386 ubifs_assert(n
>= 0 && n
< c
->fanout
);
2387 dbg_tnc("deleting %s", DBGKEY(&znode
->zbranch
[n
].key
));
2389 zbr
= &znode
->zbranch
[n
];
2392 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2394 dbg_dump_znode(c
, znode
);
2398 /* We do not "gap" zbranch slots */
2399 for (i
= n
; i
< znode
->child_cnt
- 1; i
++)
2400 znode
->zbranch
[i
] = znode
->zbranch
[i
+ 1];
2401 znode
->child_cnt
-= 1;
2403 if (znode
->child_cnt
> 0)
2407 * This was the last zbranch, we have to delete this znode from the
2412 ubifs_assert(!test_bit(OBSOLETE_ZNODE
, &znode
->flags
));
2413 ubifs_assert(ubifs_zn_dirty(znode
));
2418 atomic_long_dec(&c
->dirty_zn_cnt
);
2420 err
= insert_old_idx_znode(c
, znode
);
2425 __set_bit(OBSOLETE_ZNODE
, &znode
->flags
);
2426 atomic_long_inc(&c
->clean_zn_cnt
);
2427 atomic_long_inc(&ubifs_clean_zn_cnt
);
2431 } while (znode
->child_cnt
== 1); /* while removing last child */
2433 /* Remove from znode, entry n - 1 */
2434 znode
->child_cnt
-= 1;
2435 ubifs_assert(znode
->level
!= 0);
2436 for (i
= n
; i
< znode
->child_cnt
; i
++) {
2437 znode
->zbranch
[i
] = znode
->zbranch
[i
+ 1];
2438 if (znode
->zbranch
[i
].znode
)
2439 znode
->zbranch
[i
].znode
->iip
= i
;
2443 * If this is the root and it has only 1 child then
2444 * collapse the tree.
2446 if (!znode
->parent
) {
2447 while (znode
->child_cnt
== 1 && znode
->level
!= 0) {
2449 zbr
= &znode
->zbranch
[0];
2450 znode
= get_znode(c
, znode
, 0);
2452 return PTR_ERR(znode
);
2453 znode
= dirty_cow_znode(c
, zbr
);
2455 return PTR_ERR(znode
);
2456 znode
->parent
= NULL
;
2459 err
= insert_old_idx(c
, c
->zroot
.lnum
,
2464 c
->zroot
.lnum
= zbr
->lnum
;
2465 c
->zroot
.offs
= zbr
->offs
;
2466 c
->zroot
.len
= zbr
->len
;
2467 c
->zroot
.znode
= znode
;
2468 ubifs_assert(!test_bit(OBSOLETE_ZNODE
,
2470 ubifs_assert(test_bit(DIRTY_ZNODE
, &zp
->flags
));
2471 atomic_long_dec(&c
->dirty_zn_cnt
);
2474 __set_bit(OBSOLETE_ZNODE
, &zp
->flags
);
2475 atomic_long_inc(&c
->clean_zn_cnt
);
2476 atomic_long_inc(&ubifs_clean_zn_cnt
);
2486 * ubifs_tnc_remove - remove an index entry of a node.
2487 * @c: UBIFS file-system description object
2490 * Returns %0 on success or negative error code on failure.
2492 int ubifs_tnc_remove(struct ubifs_info
*c
, const union ubifs_key
*key
)
2494 int found
, n
, err
= 0;
2495 struct ubifs_znode
*znode
;
2497 mutex_lock(&c
->tnc_mutex
);
2498 dbg_tnc("key %s", DBGKEY(key
));
2499 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2505 err
= tnc_delete(c
, znode
, n
);
2507 err
= dbg_check_tnc(c
, 0);
2510 mutex_unlock(&c
->tnc_mutex
);
2515 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2516 * @c: UBIFS file-system description object
2518 * @nm: directory entry name
2520 * Returns %0 on success or negative error code on failure.
2522 int ubifs_tnc_remove_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
2523 const struct qstr
*nm
)
2526 struct ubifs_znode
*znode
;
2528 mutex_lock(&c
->tnc_mutex
);
2529 dbg_tnc("%.*s, key %s", nm
->len
, nm
->name
, DBGKEY(key
));
2530 err
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2536 err
= fallible_resolve_collision(c
, key
, &znode
, &n
,
2539 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2540 dbg_tnc("rc returned %d, znode %p, n %d", err
, znode
, n
);
2544 /* Ensure the znode is dirtied */
2545 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2546 znode
= dirty_cow_bottom_up(c
, znode
);
2547 if (IS_ERR(znode
)) {
2548 err
= PTR_ERR(znode
);
2552 err
= tnc_delete(c
, znode
, n
);
2558 err
= dbg_check_tnc(c
, 0);
2559 mutex_unlock(&c
->tnc_mutex
);
2564 * key_in_range - determine if a key falls within a range of keys.
2565 * @c: UBIFS file-system description object
2566 * @key: key to check
2567 * @from_key: lowest key in range
2568 * @to_key: highest key in range
2570 * This function returns %1 if the key is in range and %0 otherwise.
2572 static int key_in_range(struct ubifs_info
*c
, union ubifs_key
*key
,
2573 union ubifs_key
*from_key
, union ubifs_key
*to_key
)
2575 if (keys_cmp(c
, key
, from_key
) < 0)
2577 if (keys_cmp(c
, key
, to_key
) > 0)
2583 * ubifs_tnc_remove_range - remove index entries in range.
2584 * @c: UBIFS file-system description object
2585 * @from_key: lowest key to remove
2586 * @to_key: highest key to remove
2588 * This function removes index entries starting at @from_key and ending at
2589 * @to_key. This function returns zero in case of success and a negative error
2590 * code in case of failure.
2592 int ubifs_tnc_remove_range(struct ubifs_info
*c
, union ubifs_key
*from_key
,
2593 union ubifs_key
*to_key
)
2595 int i
, n
, k
, err
= 0;
2596 struct ubifs_znode
*znode
;
2597 union ubifs_key
*key
;
2599 mutex_lock(&c
->tnc_mutex
);
2601 /* Find first level 0 znode that contains keys to remove */
2602 err
= ubifs_lookup_level0(c
, from_key
, &znode
, &n
);
2609 err
= tnc_next(c
, &znode
, &n
);
2610 if (err
== -ENOENT
) {
2616 key
= &znode
->zbranch
[n
].key
;
2617 if (!key_in_range(c
, key
, from_key
, to_key
)) {
2623 /* Ensure the znode is dirtied */
2624 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2625 znode
= dirty_cow_bottom_up(c
, znode
);
2626 if (IS_ERR(znode
)) {
2627 err
= PTR_ERR(znode
);
2632 /* Remove all keys in range except the first */
2633 for (i
= n
+ 1, k
= 0; i
< znode
->child_cnt
; i
++, k
++) {
2634 key
= &znode
->zbranch
[i
].key
;
2635 if (!key_in_range(c
, key
, from_key
, to_key
))
2637 lnc_free(&znode
->zbranch
[i
]);
2638 err
= ubifs_add_dirt(c
, znode
->zbranch
[i
].lnum
,
2639 znode
->zbranch
[i
].len
);
2641 dbg_dump_znode(c
, znode
);
2644 dbg_tnc("removing %s", DBGKEY(key
));
2647 for (i
= n
+ 1 + k
; i
< znode
->child_cnt
; i
++)
2648 znode
->zbranch
[i
- k
] = znode
->zbranch
[i
];
2649 znode
->child_cnt
-= k
;
2652 /* Now delete the first */
2653 err
= tnc_delete(c
, znode
, n
);
2660 err
= dbg_check_tnc(c
, 0);
2661 mutex_unlock(&c
->tnc_mutex
);
2666 * ubifs_tnc_remove_ino - remove an inode from TNC.
2667 * @c: UBIFS file-system description object
2668 * @inum: inode number to remove
2670 * This function remove inode @inum and all the extended attributes associated
2671 * with the anode from TNC and returns zero in case of success or a negative
2672 * error code in case of failure.
2674 int ubifs_tnc_remove_ino(struct ubifs_info
*c
, ino_t inum
)
2676 union ubifs_key key1
, key2
;
2677 struct ubifs_dent_node
*xent
, *pxent
= NULL
;
2678 struct qstr nm
= { .name
= NULL
};
2680 dbg_tnc("ino %lu", inum
);
2683 * Walk all extended attribute entries and remove them together with
2684 * corresponding extended attribute inodes.
2686 lowest_xent_key(c
, &key1
, inum
);
2691 xent
= ubifs_tnc_next_ent(c
, &key1
, &nm
);
2693 err
= PTR_ERR(xent
);
2699 xattr_inum
= le64_to_cpu(xent
->inum
);
2700 dbg_tnc("xent '%s', ino %lu", xent
->name
, xattr_inum
);
2702 nm
.name
= xent
->name
;
2703 nm
.len
= le16_to_cpu(xent
->nlen
);
2704 err
= ubifs_tnc_remove_nm(c
, &key1
, &nm
);
2710 lowest_ino_key(c
, &key1
, xattr_inum
);
2711 highest_ino_key(c
, &key2
, xattr_inum
);
2712 err
= ubifs_tnc_remove_range(c
, &key1
, &key2
);
2720 key_read(c
, &xent
->key
, &key1
);
2724 lowest_ino_key(c
, &key1
, inum
);
2725 highest_ino_key(c
, &key2
, inum
);
2727 return ubifs_tnc_remove_range(c
, &key1
, &key2
);
2731 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2732 * @c: UBIFS file-system description object
2733 * @key: key of last entry
2734 * @nm: name of last entry found or %NULL
2736 * This function finds and reads the next directory or extended attribute entry
2737 * after the given key (@key) if there is one. @nm is used to resolve
2740 * If the name of the current entry is not known and only the key is known,
2741 * @nm->name has to be %NULL. In this case the semantics of this function is a
2742 * little bit different and it returns the entry corresponding to this key, not
2743 * the next one. If the key was not found, the closest "right" entry is
2746 * If the fist entry has to be found, @key has to contain the lowest possible
2747 * key value for this inode and @name has to be %NULL.
2749 * This function returns the found directory or extended attribute entry node
2750 * in case of success, %-ENOENT is returned if no entry was found, and a
2751 * negative error code is returned in case of failure.
2753 struct ubifs_dent_node
*ubifs_tnc_next_ent(struct ubifs_info
*c
,
2754 union ubifs_key
*key
,
2755 const struct qstr
*nm
)
2757 int n
, err
, type
= key_type(c
, key
);
2758 struct ubifs_znode
*znode
;
2759 struct ubifs_dent_node
*dent
;
2760 struct ubifs_zbranch
*zbr
;
2761 union ubifs_key
*dkey
;
2763 dbg_tnc("%s %s", nm
->name
? (char *)nm
->name
: "(lowest)", DBGKEY(key
));
2764 ubifs_assert(is_hash_key(c
, key
));
2766 mutex_lock(&c
->tnc_mutex
);
2767 err
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
2768 if (unlikely(err
< 0))
2773 /* Handle collisions */
2774 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2775 dbg_tnc("rc returned %d, znode %p, n %d",
2777 if (unlikely(err
< 0))
2781 /* Now find next entry */
2782 err
= tnc_next(c
, &znode
, &n
);
2787 * The full name of the entry was not given, in which case the
2788 * behavior of this function is a little different and it
2789 * returns current entry, not the next one.
2793 * However, the given key does not exist in the TNC
2794 * tree and @znode/@n variables contain the closest
2795 * "preceding" element. Switch to the next one.
2797 err
= tnc_next(c
, &znode
, &n
);
2803 zbr
= &znode
->zbranch
[n
];
2804 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
2805 if (unlikely(!dent
)) {
2811 * The above 'tnc_next()' call could lead us to the next inode, check
2815 if (key_inum(c
, dkey
) != key_inum(c
, key
) ||
2816 key_type(c
, dkey
) != type
) {
2821 err
= tnc_read_node_nm(c
, zbr
, dent
);
2825 mutex_unlock(&c
->tnc_mutex
);
2831 mutex_unlock(&c
->tnc_mutex
);
2832 return ERR_PTR(err
);
2836 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2837 * @c: UBIFS file-system description object
2839 * Destroy left-over obsolete znodes from a failed commit.
2841 static void tnc_destroy_cnext(struct ubifs_info
*c
)
2843 struct ubifs_znode
*cnext
;
2847 ubifs_assert(c
->cmt_state
== COMMIT_BROKEN
);
2850 struct ubifs_znode
*znode
= cnext
;
2852 cnext
= cnext
->cnext
;
2853 if (test_bit(OBSOLETE_ZNODE
, &znode
->flags
))
2855 } while (cnext
&& cnext
!= c
->cnext
);
2859 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2860 * @c: UBIFS file-system description object
2862 void ubifs_tnc_close(struct ubifs_info
*c
)
2866 tnc_destroy_cnext(c
);
2867 if (c
->zroot
.znode
) {
2868 clean_freed
= ubifs_destroy_tnc_subtree(c
->zroot
.znode
);
2869 atomic_long_sub(clean_freed
, &ubifs_clean_zn_cnt
);
2877 * left_znode - get the znode to the left.
2878 * @c: UBIFS file-system description object
2881 * This function returns a pointer to the znode to the left of @znode or NULL if
2882 * there is not one. A negative error code is returned on failure.
2884 static struct ubifs_znode
*left_znode(struct ubifs_info
*c
,
2885 struct ubifs_znode
*znode
)
2887 int level
= znode
->level
;
2890 int n
= znode
->iip
- 1;
2892 /* Go up until we can go left */
2893 znode
= znode
->parent
;
2897 /* Now go down the rightmost branch to 'level' */
2898 znode
= get_znode(c
, znode
, n
);
2901 while (znode
->level
!= level
) {
2902 n
= znode
->child_cnt
- 1;
2903 znode
= get_znode(c
, znode
, n
);
2914 * right_znode - get the znode to the right.
2915 * @c: UBIFS file-system description object
2918 * This function returns a pointer to the znode to the right of @znode or NULL
2919 * if there is not one. A negative error code is returned on failure.
2921 static struct ubifs_znode
*right_znode(struct ubifs_info
*c
,
2922 struct ubifs_znode
*znode
)
2924 int level
= znode
->level
;
2927 int n
= znode
->iip
+ 1;
2929 /* Go up until we can go right */
2930 znode
= znode
->parent
;
2933 if (n
< znode
->child_cnt
) {
2934 /* Now go down the leftmost branch to 'level' */
2935 znode
= get_znode(c
, znode
, n
);
2938 while (znode
->level
!= level
) {
2939 znode
= get_znode(c
, znode
, 0);
2950 * lookup_znode - find a particular indexing node from TNC.
2951 * @c: UBIFS file-system description object
2952 * @key: index node key to lookup
2953 * @level: index node level
2954 * @lnum: index node LEB number
2955 * @offs: index node offset
2957 * This function searches an indexing node by its first key @key and its
2958 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2959 * nodes it traverses to TNC. This function is called fro indexing nodes which
2960 * were found on the media by scanning, for example when garbage-collecting or
2961 * when doing in-the-gaps commit. This means that the indexing node which is
2962 * looked for does not have to have exactly the same leftmost key @key, because
2963 * the leftmost key may have been changed, in which case TNC will contain a
2964 * dirty znode which still refers the same @lnum:@offs. This function is clever
2965 * enough to recognize such indexing nodes.
2967 * Note, if a znode was deleted or changed too much, then this function will
2968 * not find it. For situations like this UBIFS has the old index RB-tree
2969 * (indexed by @lnum:@offs).
2971 * This function returns a pointer to the znode found or %NULL if it is not
2972 * found. A negative error code is returned on failure.
2974 static struct ubifs_znode
*lookup_znode(struct ubifs_info
*c
,
2975 union ubifs_key
*key
, int level
,
2978 struct ubifs_znode
*znode
, *zn
;
2982 * The arguments have probably been read off flash, so don't assume
2986 return ERR_PTR(-EINVAL
);
2988 /* Get the root znode */
2989 znode
= c
->zroot
.znode
;
2991 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
2995 /* Check if it is the one we are looking for */
2996 if (c
->zroot
.lnum
== lnum
&& c
->zroot
.offs
== offs
)
2998 /* Descend to the parent level i.e. (level + 1) */
2999 if (level
>= znode
->level
)
3002 ubifs_search_zbranch(c
, znode
, key
, &n
);
3005 * We reached a znode where the leftmost key is greater
3006 * than the key we are searching for. This is the same
3007 * situation as the one described in a huge comment at
3008 * the end of the 'ubifs_lookup_level0()' function. And
3009 * for exactly the same reasons we have to try to look
3010 * left before giving up.
3012 znode
= left_znode(c
, znode
);
3017 ubifs_search_zbranch(c
, znode
, key
, &n
);
3018 ubifs_assert(n
>= 0);
3020 if (znode
->level
== level
+ 1)
3022 znode
= get_znode(c
, znode
, n
);
3026 /* Check if the child is the one we are looking for */
3027 if (znode
->zbranch
[n
].lnum
== lnum
&& znode
->zbranch
[n
].offs
== offs
)
3028 return get_znode(c
, znode
, n
);
3029 /* If the key is unique, there is nowhere else to look */
3030 if (!is_hash_key(c
, key
))
3033 * The key is not unique and so may be also in the znodes to either
3040 /* Move one branch to the left */
3044 znode
= left_znode(c
, znode
);
3049 n
= znode
->child_cnt
- 1;
3052 if (znode
->zbranch
[n
].lnum
== lnum
&&
3053 znode
->zbranch
[n
].offs
== offs
)
3054 return get_znode(c
, znode
, n
);
3055 /* Stop if the key is less than the one we are looking for */
3056 if (keys_cmp(c
, &znode
->zbranch
[n
].key
, key
) < 0)
3059 /* Back to the middle */
3064 /* Move one branch to the right */
3065 if (++n
>= znode
->child_cnt
) {
3066 znode
= right_znode(c
, znode
);
3074 if (znode
->zbranch
[n
].lnum
== lnum
&&
3075 znode
->zbranch
[n
].offs
== offs
)
3076 return get_znode(c
, znode
, n
);
3077 /* Stop if the key is greater than the one we are looking for */
3078 if (keys_cmp(c
, &znode
->zbranch
[n
].key
, key
) > 0)
3085 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3086 * @c: UBIFS file-system description object
3087 * @key: key of index node
3088 * @level: index node level
3089 * @lnum: LEB number of index node
3090 * @offs: offset of index node
3092 * This function returns %0 if the index node is not referred to in the TNC, %1
3093 * if the index node is referred to in the TNC and the corresponding znode is
3094 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3095 * znode is clean, and a negative error code in case of failure.
3097 * Note, the @key argument has to be the key of the first child. Also note,
3098 * this function relies on the fact that 0:0 is never a valid LEB number and
3099 * offset for a main-area node.
3101 int is_idx_node_in_tnc(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3104 struct ubifs_znode
*znode
;
3106 znode
= lookup_znode(c
, key
, level
, lnum
, offs
);
3110 return PTR_ERR(znode
);
3112 return ubifs_zn_dirty(znode
) ? 1 : 2;
3116 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3117 * @c: UBIFS file-system description object
3119 * @lnum: node LEB number
3120 * @offs: node offset
3122 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3123 * not, and a negative error code in case of failure.
3125 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3126 * and offset for a main-area node.
3128 static int is_leaf_node_in_tnc(struct ubifs_info
*c
, union ubifs_key
*key
,
3131 struct ubifs_zbranch
*zbr
;
3132 struct ubifs_znode
*znode
, *zn
;
3133 int n
, found
, err
, nn
;
3134 const int unique
= !is_hash_key(c
, key
);
3136 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
3138 return found
; /* Error code */
3141 zbr
= &znode
->zbranch
[n
];
3142 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3143 return 1; /* Found it */
3147 * Because the key is not unique, we have to look left
3154 err
= tnc_prev(c
, &znode
, &n
);
3159 if (keys_cmp(c
, key
, &znode
->zbranch
[n
].key
))
3161 zbr
= &znode
->zbranch
[n
];
3162 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3163 return 1; /* Found it */
3169 err
= tnc_next(c
, &znode
, &n
);
3175 if (keys_cmp(c
, key
, &znode
->zbranch
[n
].key
))
3177 zbr
= &znode
->zbranch
[n
];
3178 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3179 return 1; /* Found it */
3185 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3186 * @c: UBIFS file-system description object
3188 * @level: index node level (if it is an index node)
3189 * @lnum: node LEB number
3190 * @offs: node offset
3191 * @is_idx: non-zero if the node is an index node
3193 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3194 * negative error code in case of failure. For index nodes, @key has to be the
3195 * key of the first child. An index node is considered to be in the TNC only if
3196 * the corresponding znode is clean or has not been loaded.
3198 int ubifs_tnc_has_node(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3199 int lnum
, int offs
, int is_idx
)
3203 mutex_lock(&c
->tnc_mutex
);
3205 err
= is_idx_node_in_tnc(c
, key
, level
, lnum
, offs
);
3209 /* The index node was found but it was dirty */
3212 /* The index node was found and it was clean */
3217 err
= is_leaf_node_in_tnc(c
, key
, lnum
, offs
);
3220 mutex_unlock(&c
->tnc_mutex
);
3225 * ubifs_dirty_idx_node - dirty an index node.
3226 * @c: UBIFS file-system description object
3227 * @key: index node key
3228 * @level: index node level
3229 * @lnum: index node LEB number
3230 * @offs: index node offset
3232 * This function loads and dirties an index node so that it can be garbage
3233 * collected. The @key argument has to be the key of the first child. This
3234 * function relies on the fact that 0:0 is never a valid LEB number and offset
3235 * for a main-area node. Returns %0 on success and a negative error code on
3238 int ubifs_dirty_idx_node(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3241 struct ubifs_znode
*znode
;
3244 mutex_lock(&c
->tnc_mutex
);
3245 znode
= lookup_znode(c
, key
, level
, lnum
, offs
);
3248 if (IS_ERR(znode
)) {
3249 err
= PTR_ERR(znode
);
3252 znode
= dirty_cow_bottom_up(c
, znode
);
3253 if (IS_ERR(znode
)) {
3254 err
= PTR_ERR(znode
);
3259 mutex_unlock(&c
->tnc_mutex
);