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>
34 #include <linux/slab.h>
38 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
39 * @NAME_LESS: name corresponding to the first argument is less than second
40 * @NAME_MATCHES: names match
41 * @NAME_GREATER: name corresponding to the second argument is greater than
43 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
45 * These constants were introduce to improve readability.
55 * insert_old_idx - record an index node obsoleted since the last commit start.
56 * @c: UBIFS file-system description object
57 * @lnum: LEB number of obsoleted index node
58 * @offs: offset of obsoleted index node
60 * Returns %0 on success, and a negative error code on failure.
62 * For recovery, there must always be a complete intact version of the index on
63 * flash at all times. That is called the "old index". It is the index as at the
64 * time of the last successful commit. Many of the index nodes in the old index
65 * may be dirty, but they must not be erased until the next successful commit
66 * (at which point that index becomes the old index).
68 * That means that the garbage collection and the in-the-gaps method of
69 * committing must be able to determine if an index node is in the old index.
70 * Most of the old index nodes can be found by looking up the TNC using the
71 * 'lookup_znode()' function. However, some of the old index nodes may have
72 * been deleted from the current index or may have been changed so much that
73 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
74 * That is what this function does. The RB-tree is ordered by LEB number and
75 * offset because they uniquely identify the old index node.
77 static int insert_old_idx(struct ubifs_info
*c
, int lnum
, int offs
)
79 struct ubifs_old_idx
*old_idx
, *o
;
80 struct rb_node
**p
, *parent
= NULL
;
82 old_idx
= kmalloc(sizeof(struct ubifs_old_idx
), GFP_NOFS
);
83 if (unlikely(!old_idx
))
88 p
= &c
->old_idx
.rb_node
;
91 o
= rb_entry(parent
, struct ubifs_old_idx
, rb
);
94 else if (lnum
> o
->lnum
)
96 else if (offs
< o
->offs
)
98 else if (offs
> o
->offs
)
101 ubifs_err("old idx added twice!");
106 rb_link_node(&old_idx
->rb
, parent
, p
);
107 rb_insert_color(&old_idx
->rb
, &c
->old_idx
);
112 * insert_old_idx_znode - record a znode obsoleted since last commit start.
113 * @c: UBIFS file-system description object
114 * @znode: znode of obsoleted index node
116 * Returns %0 on success, and a negative error code on failure.
118 int insert_old_idx_znode(struct ubifs_info
*c
, struct ubifs_znode
*znode
)
121 struct ubifs_zbranch
*zbr
;
123 zbr
= &znode
->parent
->zbranch
[znode
->iip
];
125 return insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
128 return insert_old_idx(c
, c
->zroot
.lnum
,
134 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
135 * @c: UBIFS file-system description object
136 * @znode: znode of obsoleted index node
138 * Returns %0 on success, and a negative error code on failure.
140 static int ins_clr_old_idx_znode(struct ubifs_info
*c
,
141 struct ubifs_znode
*znode
)
146 struct ubifs_zbranch
*zbr
;
148 zbr
= &znode
->parent
->zbranch
[znode
->iip
];
150 err
= insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
159 err
= insert_old_idx(c
, c
->zroot
.lnum
, c
->zroot
.offs
);
170 * destroy_old_idx - destroy the old_idx RB-tree.
171 * @c: UBIFS file-system description object
173 * During start commit, the old_idx RB-tree is used to avoid overwriting index
174 * nodes that were in the index last commit but have since been deleted. This
175 * is necessary for recovery i.e. the old index must be kept intact until the
176 * new index is successfully written. The old-idx RB-tree is used for the
177 * in-the-gaps method of writing index nodes and is destroyed every commit.
179 void destroy_old_idx(struct ubifs_info
*c
)
181 struct ubifs_old_idx
*old_idx
, *n
;
183 rbtree_postorder_for_each_entry_safe(old_idx
, n
, &c
->old_idx
, rb
)
186 c
->old_idx
= RB_ROOT
;
190 * copy_znode - copy a dirty znode.
191 * @c: UBIFS file-system description object
192 * @znode: znode to copy
194 * A dirty znode being committed may not be changed, so it is copied.
196 static struct ubifs_znode
*copy_znode(struct ubifs_info
*c
,
197 struct ubifs_znode
*znode
)
199 struct ubifs_znode
*zn
;
201 zn
= kmalloc(c
->max_znode_sz
, GFP_NOFS
);
203 return ERR_PTR(-ENOMEM
);
205 memcpy(zn
, znode
, c
->max_znode_sz
);
207 __set_bit(DIRTY_ZNODE
, &zn
->flags
);
208 __clear_bit(COW_ZNODE
, &zn
->flags
);
210 ubifs_assert(!ubifs_zn_obsolete(znode
));
211 __set_bit(OBSOLETE_ZNODE
, &znode
->flags
);
213 if (znode
->level
!= 0) {
215 const int n
= zn
->child_cnt
;
217 /* The children now have new parent */
218 for (i
= 0; i
< n
; i
++) {
219 struct ubifs_zbranch
*zbr
= &zn
->zbranch
[i
];
222 zbr
->znode
->parent
= zn
;
226 atomic_long_inc(&c
->dirty_zn_cnt
);
231 * add_idx_dirt - add dirt due to a dirty znode.
232 * @c: UBIFS file-system description object
233 * @lnum: LEB number of index node
234 * @dirt: size of index node
236 * This function updates lprops dirty space and the new size of the index.
238 static int add_idx_dirt(struct ubifs_info
*c
, int lnum
, int dirt
)
240 c
->calc_idx_sz
-= ALIGN(dirt
, 8);
241 return ubifs_add_dirt(c
, lnum
, dirt
);
245 * dirty_cow_znode - ensure a znode is not being committed.
246 * @c: UBIFS file-system description object
247 * @zbr: branch of znode to check
249 * Returns dirtied znode on success or negative error code on failure.
251 static struct ubifs_znode
*dirty_cow_znode(struct ubifs_info
*c
,
252 struct ubifs_zbranch
*zbr
)
254 struct ubifs_znode
*znode
= zbr
->znode
;
255 struct ubifs_znode
*zn
;
258 if (!ubifs_zn_cow(znode
)) {
259 /* znode is not being committed */
260 if (!test_and_set_bit(DIRTY_ZNODE
, &znode
->flags
)) {
261 atomic_long_inc(&c
->dirty_zn_cnt
);
262 atomic_long_dec(&c
->clean_zn_cnt
);
263 atomic_long_dec(&ubifs_clean_zn_cnt
);
264 err
= add_idx_dirt(c
, zbr
->lnum
, zbr
->len
);
271 zn
= copy_znode(c
, znode
);
276 err
= insert_old_idx(c
, zbr
->lnum
, zbr
->offs
);
279 err
= add_idx_dirt(c
, zbr
->lnum
, zbr
->len
);
294 * lnc_add - add a leaf node to the leaf node cache.
295 * @c: UBIFS file-system description object
296 * @zbr: zbranch of leaf node
299 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
300 * purpose of the leaf node cache is to save re-reading the same leaf node over
301 * and over again. Most things are cached by VFS, however the file system must
302 * cache directory entries for readdir and for resolving hash collisions. The
303 * present implementation of the leaf node cache is extremely simple, and
304 * allows for error returns that are not used but that may be needed if a more
305 * complex implementation is created.
307 * Note, this function does not add the @node object to LNC directly, but
308 * allocates a copy of the object and adds the copy to LNC. The reason for this
309 * is that @node has been allocated outside of the TNC subsystem and will be
310 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
311 * may be changed at any time, e.g. freed by the shrinker.
313 static int lnc_add(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
318 const struct ubifs_dent_node
*dent
= node
;
320 ubifs_assert(!zbr
->leaf
);
321 ubifs_assert(zbr
->len
!= 0);
322 ubifs_assert(is_hash_key(c
, &zbr
->key
));
324 err
= ubifs_validate_entry(c
, dent
);
327 ubifs_dump_node(c
, dent
);
331 lnc_node
= kmemdup(node
, zbr
->len
, GFP_NOFS
);
333 /* We don't have to have the cache, so no error */
336 zbr
->leaf
= lnc_node
;
341 * lnc_add_directly - add a leaf node to the leaf-node-cache.
342 * @c: UBIFS file-system description object
343 * @zbr: zbranch of leaf node
346 * This function is similar to 'lnc_add()', but it does not create a copy of
347 * @node but inserts @node to TNC directly.
349 static int lnc_add_directly(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
354 ubifs_assert(!zbr
->leaf
);
355 ubifs_assert(zbr
->len
!= 0);
357 err
= ubifs_validate_entry(c
, node
);
360 ubifs_dump_node(c
, node
);
369 * lnc_free - remove a leaf node from the leaf node cache.
370 * @zbr: zbranch of leaf node
373 static void lnc_free(struct ubifs_zbranch
*zbr
)
382 * tnc_read_node_nm - read a "hashed" leaf node.
383 * @c: UBIFS file-system description object
384 * @zbr: key and position of the node
385 * @node: node is returned here
387 * This function reads a "hashed" node defined by @zbr from the leaf node cache
388 * (in it is there) or from the hash media, in which case the node is also
389 * added to LNC. Returns zero in case of success or a negative negative error
390 * code in case of failure.
392 static int tnc_read_node_nm(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
397 ubifs_assert(is_hash_key(c
, &zbr
->key
));
400 /* Read from the leaf node cache */
401 ubifs_assert(zbr
->len
!= 0);
402 memcpy(node
, zbr
->leaf
, zbr
->len
);
406 err
= ubifs_tnc_read_node(c
, zbr
, node
);
410 /* Add the node to the leaf node cache */
411 err
= lnc_add(c
, zbr
, node
);
416 * try_read_node - read a node if it is a node.
417 * @c: UBIFS file-system description object
418 * @buf: buffer to read to
420 * @len: node length (not aligned)
421 * @lnum: LEB number of node to read
422 * @offs: offset of node to read
424 * This function tries to read a node of known type and length, checks it and
425 * stores it in @buf. This function returns %1 if a node is present and %0 if
426 * a node is not present. A negative error code is returned for I/O errors.
427 * This function performs that same function as ubifs_read_node except that
428 * it does not require that there is actually a node present and instead
429 * the return code indicates if a node was read.
431 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
432 * is true (it is controlled by corresponding mount option). However, if
433 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
434 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
435 * because during mounting or re-mounting from R/O mode to R/W mode we may read
436 * journal nodes (when replying the journal or doing the recovery) and the
437 * journal nodes may potentially be corrupted, so checking is required.
439 static int try_read_node(const struct ubifs_info
*c
, void *buf
, int type
,
440 int len
, int lnum
, int offs
)
443 struct ubifs_ch
*ch
= buf
;
444 uint32_t crc
, node_crc
;
446 dbg_io("LEB %d:%d, %s, length %d", lnum
, offs
, dbg_ntype(type
), len
);
448 err
= ubifs_leb_read(c
, lnum
, buf
, offs
, len
, 1);
450 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
451 type
, lnum
, offs
, err
);
455 if (le32_to_cpu(ch
->magic
) != UBIFS_NODE_MAGIC
)
458 if (ch
->node_type
!= type
)
461 node_len
= le32_to_cpu(ch
->len
);
465 if (type
== UBIFS_DATA_NODE
&& c
->no_chk_data_crc
&& !c
->mounting
&&
469 crc
= crc32(UBIFS_CRC32_INIT
, buf
+ 8, node_len
- 8);
470 node_crc
= le32_to_cpu(ch
->crc
);
478 * fallible_read_node - try to read a leaf node.
479 * @c: UBIFS file-system description object
480 * @key: key of node to read
481 * @zbr: position of node
482 * @node: node returned
484 * This function tries to read a node and returns %1 if the node is read, %0
485 * if the node is not present, and a negative error code in the case of error.
487 static int fallible_read_node(struct ubifs_info
*c
, const union ubifs_key
*key
,
488 struct ubifs_zbranch
*zbr
, void *node
)
492 dbg_tnck(key
, "LEB %d:%d, key ", zbr
->lnum
, zbr
->offs
);
494 ret
= try_read_node(c
, node
, key_type(c
, key
), zbr
->len
, zbr
->lnum
,
497 union ubifs_key node_key
;
498 struct ubifs_dent_node
*dent
= node
;
500 /* All nodes have key in the same place */
501 key_read(c
, &dent
->key
, &node_key
);
502 if (keys_cmp(c
, key
, &node_key
) != 0)
505 if (ret
== 0 && c
->replaying
)
506 dbg_mntk(key
, "dangling branch LEB %d:%d len %d, key ",
507 zbr
->lnum
, zbr
->offs
, zbr
->len
);
512 * matches_name - determine if a direntry or xattr entry matches a given name.
513 * @c: UBIFS file-system description object
514 * @zbr: zbranch of dent
517 * This function checks if xentry/direntry referred by zbranch @zbr matches name
518 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
519 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
520 * of failure, a negative error code is returned.
522 static int matches_name(struct ubifs_info
*c
, struct ubifs_zbranch
*zbr
,
523 const struct qstr
*nm
)
525 struct ubifs_dent_node
*dent
;
528 /* If possible, match against the dent in the leaf node cache */
530 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
534 err
= ubifs_tnc_read_node(c
, zbr
, dent
);
538 /* Add the node to the leaf node cache */
539 err
= lnc_add_directly(c
, zbr
, dent
);
545 nlen
= le16_to_cpu(dent
->nlen
);
546 err
= memcmp(dent
->name
, nm
->name
, min_t(int, nlen
, nm
->len
));
550 else if (nlen
< nm
->len
)
565 * get_znode - get a TNC znode that may not be loaded yet.
566 * @c: UBIFS file-system description object
567 * @znode: parent znode
568 * @n: znode branch slot number
570 * This function returns the znode or a negative error code.
572 static struct ubifs_znode
*get_znode(struct ubifs_info
*c
,
573 struct ubifs_znode
*znode
, int n
)
575 struct ubifs_zbranch
*zbr
;
577 zbr
= &znode
->zbranch
[n
];
581 znode
= ubifs_load_znode(c
, zbr
, znode
, n
);
586 * tnc_next - find next TNC entry.
587 * @c: UBIFS file-system description object
588 * @zn: znode is passed and returned here
589 * @n: znode branch slot number is passed and returned here
591 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
592 * no next entry, or a negative error code otherwise.
594 static int tnc_next(struct ubifs_info
*c
, struct ubifs_znode
**zn
, int *n
)
596 struct ubifs_znode
*znode
= *zn
;
600 if (nn
< znode
->child_cnt
) {
605 struct ubifs_znode
*zp
;
612 if (nn
< znode
->child_cnt
) {
613 znode
= get_znode(c
, znode
, nn
);
615 return PTR_ERR(znode
);
616 while (znode
->level
!= 0) {
617 znode
= get_znode(c
, znode
, 0);
619 return PTR_ERR(znode
);
631 * tnc_prev - find previous TNC entry.
632 * @c: UBIFS file-system description object
633 * @zn: znode is returned here
634 * @n: znode branch slot number is passed and returned here
636 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
637 * there is no next entry, or a negative error code otherwise.
639 static int tnc_prev(struct ubifs_info
*c
, struct ubifs_znode
**zn
, int *n
)
641 struct ubifs_znode
*znode
= *zn
;
649 struct ubifs_znode
*zp
;
657 znode
= get_znode(c
, znode
, nn
);
659 return PTR_ERR(znode
);
660 while (znode
->level
!= 0) {
661 nn
= znode
->child_cnt
- 1;
662 znode
= get_znode(c
, znode
, nn
);
664 return PTR_ERR(znode
);
666 nn
= znode
->child_cnt
- 1;
676 * resolve_collision - resolve a collision.
677 * @c: UBIFS file-system description object
678 * @key: key of a directory or extended attribute entry
679 * @zn: znode is returned here
680 * @n: zbranch number is passed and returned here
681 * @nm: name of the entry
683 * This function is called for "hashed" keys to make sure that the found key
684 * really corresponds to the looked up node (directory or extended attribute
685 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
686 * %0 is returned if @nm is not found and @zn and @n are set to the previous
687 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
688 * This means that @n may be set to %-1 if the leftmost key in @zn is the
689 * previous one. A negative error code is returned on failures.
691 static int resolve_collision(struct ubifs_info
*c
, const union ubifs_key
*key
,
692 struct ubifs_znode
**zn
, int *n
,
693 const struct qstr
*nm
)
697 err
= matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
698 if (unlikely(err
< 0))
700 if (err
== NAME_MATCHES
)
703 if (err
== NAME_GREATER
) {
706 err
= tnc_prev(c
, zn
, n
);
707 if (err
== -ENOENT
) {
708 ubifs_assert(*n
== 0);
714 if (keys_cmp(c
, &(*zn
)->zbranch
[*n
].key
, key
)) {
716 * We have found the branch after which we would
717 * like to insert, but inserting in this znode
718 * may still be wrong. Consider the following 3
719 * znodes, in the case where we are resolving a
720 * collision with Key2.
723 * ----------------------
724 * level 1 | Key0 | Key1 |
725 * -----------------------
727 * znode za | | znode zb
728 * ------------ ------------
729 * level 0 | Key0 | | Key2 |
730 * ------------ ------------
732 * The lookup finds Key2 in znode zb. Lets say
733 * there is no match and the name is greater so
734 * we look left. When we find Key0, we end up
735 * here. If we return now, we will insert into
736 * znode za at slot n = 1. But that is invalid
737 * according to the parent's keys. Key2 must
738 * be inserted into znode zb.
740 * Note, this problem is not relevant for the
741 * case when we go right, because
742 * 'tnc_insert()' would correct the parent key.
744 if (*n
== (*zn
)->child_cnt
- 1) {
745 err
= tnc_next(c
, zn
, n
);
747 /* Should be impossible */
753 ubifs_assert(*n
== 0);
758 err
= matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
761 if (err
== NAME_LESS
)
763 if (err
== NAME_MATCHES
)
765 ubifs_assert(err
== NAME_GREATER
);
769 struct ubifs_znode
*znode
= *zn
;
773 err
= tnc_next(c
, &znode
, &nn
);
778 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
780 err
= matches_name(c
, &znode
->zbranch
[nn
], nm
);
783 if (err
== NAME_GREATER
)
787 if (err
== NAME_MATCHES
)
789 ubifs_assert(err
== NAME_LESS
);
795 * fallible_matches_name - determine if a dent matches a given name.
796 * @c: UBIFS file-system description object
797 * @zbr: zbranch of dent
800 * This is a "fallible" version of 'matches_name()' function which does not
801 * panic if the direntry/xentry referred by @zbr does not exist on the media.
803 * This function checks if xentry/direntry referred by zbranch @zbr matches name
804 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
805 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
806 * if xentry/direntry referred by @zbr does not exist on the media. A negative
807 * error code is returned in case of failure.
809 static int fallible_matches_name(struct ubifs_info
*c
,
810 struct ubifs_zbranch
*zbr
,
811 const struct qstr
*nm
)
813 struct ubifs_dent_node
*dent
;
816 /* If possible, match against the dent in the leaf node cache */
818 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
822 err
= fallible_read_node(c
, &zbr
->key
, zbr
, dent
);
826 /* The node was not present */
830 ubifs_assert(err
== 1);
832 err
= lnc_add_directly(c
, zbr
, dent
);
838 nlen
= le16_to_cpu(dent
->nlen
);
839 err
= memcmp(dent
->name
, nm
->name
, min_t(int, nlen
, nm
->len
));
843 else if (nlen
< nm
->len
)
858 * fallible_resolve_collision - resolve a collision even if nodes are missing.
859 * @c: UBIFS file-system description object
861 * @zn: znode is returned here
862 * @n: branch number is passed and returned here
863 * @nm: name of directory entry
864 * @adding: indicates caller is adding a key to the TNC
866 * This is a "fallible" version of the 'resolve_collision()' function which
867 * does not panic if one of the nodes referred to by TNC does not exist on the
868 * media. This may happen when replaying the journal if a deleted node was
869 * Garbage-collected and the commit was not done. A branch that refers to a node
870 * that is not present is called a dangling branch. The following are the return
871 * codes for this function:
872 * o if @nm was found, %1 is returned and @zn and @n are set to the found
874 * o if we are @adding and @nm was not found, %0 is returned;
875 * o if we are not @adding and @nm was not found, but a dangling branch was
876 * found, then %1 is returned and @zn and @n are set to the dangling branch;
877 * o a negative error code is returned in case of failure.
879 static int fallible_resolve_collision(struct ubifs_info
*c
,
880 const union ubifs_key
*key
,
881 struct ubifs_znode
**zn
, int *n
,
882 const struct qstr
*nm
, int adding
)
884 struct ubifs_znode
*o_znode
= NULL
, *znode
= *zn
;
885 int uninitialized_var(o_n
), err
, cmp
, unsure
= 0, nn
= *n
;
887 cmp
= fallible_matches_name(c
, &znode
->zbranch
[nn
], nm
);
888 if (unlikely(cmp
< 0))
890 if (cmp
== NAME_MATCHES
)
892 if (cmp
== NOT_ON_MEDIA
) {
896 * We are unlucky and hit a dangling branch straight away.
897 * Now we do not really know where to go to find the needed
898 * branch - to the left or to the right. Well, let's try left.
902 unsure
= 1; /* Remove a dangling branch wherever it is */
904 if (cmp
== NAME_GREATER
|| unsure
) {
907 err
= tnc_prev(c
, zn
, n
);
908 if (err
== -ENOENT
) {
909 ubifs_assert(*n
== 0);
915 if (keys_cmp(c
, &(*zn
)->zbranch
[*n
].key
, key
)) {
916 /* See comments in 'resolve_collision()' */
917 if (*n
== (*zn
)->child_cnt
- 1) {
918 err
= tnc_next(c
, zn
, n
);
920 /* Should be impossible */
926 ubifs_assert(*n
== 0);
931 err
= fallible_matches_name(c
, &(*zn
)->zbranch
[*n
], nm
);
934 if (err
== NAME_MATCHES
)
936 if (err
== NOT_ON_MEDIA
) {
943 if (err
== NAME_LESS
)
950 if (cmp
== NAME_LESS
|| unsure
) {
955 err
= tnc_next(c
, &znode
, &nn
);
960 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
962 err
= fallible_matches_name(c
, &znode
->zbranch
[nn
], nm
);
965 if (err
== NAME_GREATER
)
969 if (err
== NAME_MATCHES
)
971 if (err
== NOT_ON_MEDIA
) {
978 /* Never match a dangling branch when adding */
979 if (adding
|| !o_znode
)
982 dbg_mntk(key
, "dangling match LEB %d:%d len %d key ",
983 o_znode
->zbranch
[o_n
].lnum
, o_znode
->zbranch
[o_n
].offs
,
984 o_znode
->zbranch
[o_n
].len
);
991 * matches_position - determine if a zbranch matches a given position.
992 * @zbr: zbranch of dent
993 * @lnum: LEB number of dent to match
994 * @offs: offset of dent to match
996 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
998 static int matches_position(struct ubifs_zbranch
*zbr
, int lnum
, int offs
)
1000 if (zbr
->lnum
== lnum
&& zbr
->offs
== offs
)
1007 * resolve_collision_directly - resolve a collision directly.
1008 * @c: UBIFS file-system description object
1009 * @key: key of directory entry
1010 * @zn: znode is passed and returned here
1011 * @n: zbranch number is passed and returned here
1012 * @lnum: LEB number of dent node to match
1013 * @offs: offset of dent node to match
1015 * This function is used for "hashed" keys to make sure the found directory or
1016 * extended attribute entry node is what was looked for. It is used when the
1017 * flash address of the right node is known (@lnum:@offs) which makes it much
1018 * easier to resolve collisions (no need to read entries and match full
1019 * names). This function returns %1 and sets @zn and @n if the collision is
1020 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1021 * previous directory entry. Otherwise a negative error code is returned.
1023 static int resolve_collision_directly(struct ubifs_info
*c
,
1024 const union ubifs_key
*key
,
1025 struct ubifs_znode
**zn
, int *n
,
1028 struct ubifs_znode
*znode
;
1033 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
))
1038 err
= tnc_prev(c
, &znode
, &nn
);
1043 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
1045 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
)) {
1056 err
= tnc_next(c
, &znode
, &nn
);
1061 if (keys_cmp(c
, &znode
->zbranch
[nn
].key
, key
))
1065 if (matches_position(&znode
->zbranch
[nn
], lnum
, offs
))
1071 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1072 * @c: UBIFS file-system description object
1073 * @znode: znode to dirty
1075 * If we do not have a unique key that resides in a znode, then we cannot
1076 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1077 * This function records the path back to the last dirty ancestor, and then
1078 * dirties the znodes on that path.
1080 static struct ubifs_znode
*dirty_cow_bottom_up(struct ubifs_info
*c
,
1081 struct ubifs_znode
*znode
)
1083 struct ubifs_znode
*zp
;
1084 int *path
= c
->bottom_up_buf
, p
= 0;
1086 ubifs_assert(c
->zroot
.znode
);
1087 ubifs_assert(znode
);
1088 if (c
->zroot
.znode
->level
> BOTTOM_UP_HEIGHT
) {
1089 kfree(c
->bottom_up_buf
);
1090 c
->bottom_up_buf
= kmalloc(c
->zroot
.znode
->level
* sizeof(int),
1092 if (!c
->bottom_up_buf
)
1093 return ERR_PTR(-ENOMEM
);
1094 path
= c
->bottom_up_buf
;
1096 if (c
->zroot
.znode
->level
) {
1097 /* Go up until parent is dirty */
1105 ubifs_assert(p
< c
->zroot
.znode
->level
);
1107 if (!zp
->cnext
&& ubifs_zn_dirty(znode
))
1113 /* Come back down, dirtying as we go */
1115 struct ubifs_zbranch
*zbr
;
1119 ubifs_assert(path
[p
- 1] >= 0);
1120 ubifs_assert(path
[p
- 1] < zp
->child_cnt
);
1121 zbr
= &zp
->zbranch
[path
[--p
]];
1122 znode
= dirty_cow_znode(c
, zbr
);
1124 ubifs_assert(znode
== c
->zroot
.znode
);
1125 znode
= dirty_cow_znode(c
, &c
->zroot
);
1127 if (IS_ERR(znode
) || !p
)
1129 ubifs_assert(path
[p
- 1] >= 0);
1130 ubifs_assert(path
[p
- 1] < znode
->child_cnt
);
1131 znode
= znode
->zbranch
[path
[p
- 1]].znode
;
1138 * ubifs_lookup_level0 - search for zero-level znode.
1139 * @c: UBIFS file-system description object
1140 * @key: key to lookup
1141 * @zn: znode is returned here
1142 * @n: znode branch slot number is returned here
1144 * This function looks up the TNC tree and search for zero-level znode which
1145 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1147 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1148 * is returned and slot number of the matched branch is stored in @n;
1149 * o not exact match, which means that zero-level znode does not contain
1150 * @key, then %0 is returned and slot number of the closest branch is stored
1152 * o @key is so small that it is even less than the lowest key of the
1153 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1155 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1156 * function reads corresponding indexing nodes and inserts them to TNC. In
1157 * case of failure, a negative error code is returned.
1159 int ubifs_lookup_level0(struct ubifs_info
*c
, const union ubifs_key
*key
,
1160 struct ubifs_znode
**zn
, int *n
)
1163 struct ubifs_znode
*znode
;
1164 unsigned long time
= get_seconds();
1166 dbg_tnck(key
, "search key ");
1167 ubifs_assert(key_type(c
, key
) < UBIFS_INVALID_KEY
);
1169 znode
= c
->zroot
.znode
;
1170 if (unlikely(!znode
)) {
1171 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
1173 return PTR_ERR(znode
);
1179 struct ubifs_zbranch
*zbr
;
1181 exact
= ubifs_search_zbranch(c
, znode
, key
, n
);
1183 if (znode
->level
== 0)
1188 zbr
= &znode
->zbranch
[*n
];
1196 /* znode is not in TNC cache, load it from the media */
1197 znode
= ubifs_load_znode(c
, zbr
, znode
, *n
);
1199 return PTR_ERR(znode
);
1203 if (exact
|| !is_hash_key(c
, key
) || *n
!= -1) {
1204 dbg_tnc("found %d, lvl %d, n %d", exact
, znode
->level
, *n
);
1209 * Here is a tricky place. We have not found the key and this is a
1210 * "hashed" key, which may collide. The rest of the code deals with
1211 * situations like this:
1215 * | 3 | 5 | | 6 | 7 | (x)
1217 * Or more a complex example:
1221 * | 1 | 3 | | 5 | 8 |
1223 * | 5 | 5 | | 6 | 7 | (x)
1225 * In the examples, if we are looking for key "5", we may reach nodes
1226 * marked with "(x)". In this case what we have do is to look at the
1227 * left and see if there is "5" key there. If there is, we have to
1230 * Note, this whole situation is possible because we allow to have
1231 * elements which are equivalent to the next key in the parent in the
1232 * children of current znode. For example, this happens if we split a
1233 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1237 * | 3 | 5 | | 5 | 6 | 7 |
1239 * And this becomes what is at the first "picture" after key "5" marked
1240 * with "^" is removed. What could be done is we could prohibit
1241 * splitting in the middle of the colliding sequence. Also, when
1242 * removing the leftmost key, we would have to correct the key of the
1243 * parent node, which would introduce additional complications. Namely,
1244 * if we changed the leftmost key of the parent znode, the garbage
1245 * collector would be unable to find it (GC is doing this when GC'ing
1246 * indexing LEBs). Although we already have an additional RB-tree where
1247 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1248 * after the commit. But anyway, this does not look easy to implement
1249 * so we did not try this.
1251 err
= tnc_prev(c
, &znode
, n
);
1252 if (err
== -ENOENT
) {
1253 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1257 if (unlikely(err
< 0))
1259 if (keys_cmp(c
, key
, &znode
->zbranch
[*n
].key
)) {
1260 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1265 dbg_tnc("found 1, lvl %d, n %d", znode
->level
, *n
);
1271 * lookup_level0_dirty - search for zero-level znode dirtying.
1272 * @c: UBIFS file-system description object
1273 * @key: key to lookup
1274 * @zn: znode is returned here
1275 * @n: znode branch slot number is returned here
1277 * This function looks up the TNC tree and search for zero-level znode which
1278 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1280 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1281 * is returned and slot number of the matched branch is stored in @n;
1282 * o not exact match, which means that zero-level znode does not contain @key
1283 * then %0 is returned and slot number of the closed branch is stored in
1285 * o @key is so small that it is even less than the lowest key of the
1286 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1288 * Additionally all znodes in the path from the root to the located zero-level
1289 * znode are marked as dirty.
1291 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1292 * function reads corresponding indexing nodes and inserts them to TNC. In
1293 * case of failure, a negative error code is returned.
1295 static int lookup_level0_dirty(struct ubifs_info
*c
, const union ubifs_key
*key
,
1296 struct ubifs_znode
**zn
, int *n
)
1299 struct ubifs_znode
*znode
;
1300 unsigned long time
= get_seconds();
1302 dbg_tnck(key
, "search and dirty key ");
1304 znode
= c
->zroot
.znode
;
1305 if (unlikely(!znode
)) {
1306 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
1308 return PTR_ERR(znode
);
1311 znode
= dirty_cow_znode(c
, &c
->zroot
);
1313 return PTR_ERR(znode
);
1318 struct ubifs_zbranch
*zbr
;
1320 exact
= ubifs_search_zbranch(c
, znode
, key
, n
);
1322 if (znode
->level
== 0)
1327 zbr
= &znode
->zbranch
[*n
];
1331 znode
= dirty_cow_znode(c
, zbr
);
1333 return PTR_ERR(znode
);
1337 /* znode is not in TNC cache, load it from the media */
1338 znode
= ubifs_load_znode(c
, zbr
, znode
, *n
);
1340 return PTR_ERR(znode
);
1341 znode
= dirty_cow_znode(c
, zbr
);
1343 return PTR_ERR(znode
);
1347 if (exact
|| !is_hash_key(c
, key
) || *n
!= -1) {
1348 dbg_tnc("found %d, lvl %d, n %d", exact
, znode
->level
, *n
);
1353 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1356 err
= tnc_prev(c
, &znode
, n
);
1357 if (err
== -ENOENT
) {
1359 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1362 if (unlikely(err
< 0))
1364 if (keys_cmp(c
, key
, &znode
->zbranch
[*n
].key
)) {
1366 dbg_tnc("found 0, lvl %d, n -1", znode
->level
);
1370 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
1371 znode
= dirty_cow_bottom_up(c
, znode
);
1373 return PTR_ERR(znode
);
1376 dbg_tnc("found 1, lvl %d, n %d", znode
->level
, *n
);
1382 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1383 * @c: UBIFS file-system description object
1385 * @gc_seq1: garbage collection sequence number
1387 * This function determines if @lnum may have been garbage collected since
1388 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1391 static int maybe_leb_gced(struct ubifs_info
*c
, int lnum
, int gc_seq1
)
1393 int gc_seq2
, gced_lnum
;
1395 gced_lnum
= c
->gced_lnum
;
1397 gc_seq2
= c
->gc_seq
;
1398 /* Same seq means no GC */
1399 if (gc_seq1
== gc_seq2
)
1401 /* Different by more than 1 means we don't know */
1402 if (gc_seq1
+ 1 != gc_seq2
)
1405 * We have seen the sequence number has increased by 1. Now we need to
1406 * be sure we read the right LEB number, so read it again.
1409 if (gced_lnum
!= c
->gced_lnum
)
1411 /* Finally we can check lnum */
1412 if (gced_lnum
== lnum
)
1418 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1419 * @c: UBIFS file-system description object
1420 * @key: node key to lookup
1421 * @node: the node is returned here
1422 * @lnum: LEB number is returned here
1423 * @offs: offset is returned here
1425 * This function looks up and reads node with key @key. The caller has to make
1426 * sure the @node buffer is large enough to fit the node. Returns zero in case
1427 * of success, %-ENOENT if the node was not found, and a negative error code in
1428 * case of failure. The node location can be returned in @lnum and @offs.
1430 int ubifs_tnc_locate(struct ubifs_info
*c
, const union ubifs_key
*key
,
1431 void *node
, int *lnum
, int *offs
)
1433 int found
, n
, err
, safely
= 0, gc_seq1
;
1434 struct ubifs_znode
*znode
;
1435 struct ubifs_zbranch zbr
, *zt
;
1438 mutex_lock(&c
->tnc_mutex
);
1439 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
1443 } else if (found
< 0) {
1447 zt
= &znode
->zbranch
[n
];
1452 if (is_hash_key(c
, key
)) {
1454 * In this case the leaf node cache gets used, so we pass the
1455 * address of the zbranch and keep the mutex locked
1457 err
= tnc_read_node_nm(c
, zt
, node
);
1461 err
= ubifs_tnc_read_node(c
, zt
, node
);
1464 /* Drop the TNC mutex prematurely and race with garbage collection */
1465 zbr
= znode
->zbranch
[n
];
1466 gc_seq1
= c
->gc_seq
;
1467 mutex_unlock(&c
->tnc_mutex
);
1469 if (ubifs_get_wbuf(c
, zbr
.lnum
)) {
1470 /* We do not GC journal heads */
1471 err
= ubifs_tnc_read_node(c
, &zbr
, node
);
1475 err
= fallible_read_node(c
, key
, &zbr
, node
);
1476 if (err
<= 0 || maybe_leb_gced(c
, zbr
.lnum
, gc_seq1
)) {
1478 * The node may have been GC'ed out from under us so try again
1479 * while keeping the TNC mutex locked.
1487 mutex_unlock(&c
->tnc_mutex
);
1492 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1493 * @c: UBIFS file-system description object
1494 * @bu: bulk-read parameters and results
1496 * Lookup consecutive data node keys for the same inode that reside
1497 * consecutively in the same LEB. This function returns zero in case of success
1498 * and a negative error code in case of failure.
1500 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1501 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1502 * maximum possible amount of nodes for bulk-read.
1504 int ubifs_tnc_get_bu_keys(struct ubifs_info
*c
, struct bu_info
*bu
)
1506 int n
, err
= 0, lnum
= -1, uninitialized_var(offs
);
1507 int uninitialized_var(len
);
1508 unsigned int block
= key_block(c
, &bu
->key
);
1509 struct ubifs_znode
*znode
;
1515 mutex_lock(&c
->tnc_mutex
);
1516 /* Find first key */
1517 err
= ubifs_lookup_level0(c
, &bu
->key
, &znode
, &n
);
1522 len
= znode
->zbranch
[n
].len
;
1523 /* The buffer must be big enough for at least 1 node */
1524 if (len
> bu
->buf_len
) {
1529 bu
->zbranch
[bu
->cnt
++] = znode
->zbranch
[n
];
1531 lnum
= znode
->zbranch
[n
].lnum
;
1532 offs
= ALIGN(znode
->zbranch
[n
].offs
+ len
, 8);
1535 struct ubifs_zbranch
*zbr
;
1536 union ubifs_key
*key
;
1537 unsigned int next_block
;
1540 err
= tnc_next(c
, &znode
, &n
);
1543 zbr
= &znode
->zbranch
[n
];
1545 /* See if there is another data key for this file */
1546 if (key_inum(c
, key
) != key_inum(c
, &bu
->key
) ||
1547 key_type(c
, key
) != UBIFS_DATA_KEY
) {
1552 /* First key found */
1554 offs
= ALIGN(zbr
->offs
+ zbr
->len
, 8);
1556 if (len
> bu
->buf_len
) {
1562 * The data nodes must be in consecutive positions in
1565 if (zbr
->lnum
!= lnum
|| zbr
->offs
!= offs
)
1567 offs
+= ALIGN(zbr
->len
, 8);
1568 len
= ALIGN(len
, 8) + zbr
->len
;
1569 /* Must not exceed buffer length */
1570 if (len
> bu
->buf_len
)
1573 /* Allow for holes */
1574 next_block
= key_block(c
, key
);
1575 bu
->blk_cnt
+= (next_block
- block
- 1);
1576 if (bu
->blk_cnt
>= UBIFS_MAX_BULK_READ
)
1580 bu
->zbranch
[bu
->cnt
++] = *zbr
;
1582 /* See if we have room for more */
1583 if (bu
->cnt
>= UBIFS_MAX_BULK_READ
)
1585 if (bu
->blk_cnt
>= UBIFS_MAX_BULK_READ
)
1589 if (err
== -ENOENT
) {
1593 bu
->gc_seq
= c
->gc_seq
;
1594 mutex_unlock(&c
->tnc_mutex
);
1598 * An enormous hole could cause bulk-read to encompass too many
1599 * page cache pages, so limit the number here.
1601 if (bu
->blk_cnt
> UBIFS_MAX_BULK_READ
)
1602 bu
->blk_cnt
= UBIFS_MAX_BULK_READ
;
1604 * Ensure that bulk-read covers a whole number of page cache
1607 if (UBIFS_BLOCKS_PER_PAGE
== 1 ||
1608 !(bu
->blk_cnt
& (UBIFS_BLOCKS_PER_PAGE
- 1)))
1611 /* At the end of file we can round up */
1612 bu
->blk_cnt
+= UBIFS_BLOCKS_PER_PAGE
- 1;
1615 /* Exclude data nodes that do not make up a whole page cache page */
1616 block
= key_block(c
, &bu
->key
) + bu
->blk_cnt
;
1617 block
&= ~(UBIFS_BLOCKS_PER_PAGE
- 1);
1619 if (key_block(c
, &bu
->zbranch
[bu
->cnt
- 1].key
) < block
)
1627 * read_wbuf - bulk-read from a LEB with a wbuf.
1628 * @wbuf: wbuf that may overlap the read
1629 * @buf: buffer into which to read
1631 * @lnum: LEB number from which to read
1632 * @offs: offset from which to read
1634 * This functions returns %0 on success or a negative error code on failure.
1636 static int read_wbuf(struct ubifs_wbuf
*wbuf
, void *buf
, int len
, int lnum
,
1639 const struct ubifs_info
*c
= wbuf
->c
;
1642 dbg_io("LEB %d:%d, length %d", lnum
, offs
, len
);
1643 ubifs_assert(wbuf
&& lnum
>= 0 && lnum
< c
->leb_cnt
&& offs
>= 0);
1644 ubifs_assert(!(offs
& 7) && offs
< c
->leb_size
);
1645 ubifs_assert(offs
+ len
<= c
->leb_size
);
1647 spin_lock(&wbuf
->lock
);
1648 overlap
= (lnum
== wbuf
->lnum
&& offs
+ len
> wbuf
->offs
);
1650 /* We may safely unlock the write-buffer and read the data */
1651 spin_unlock(&wbuf
->lock
);
1652 return ubifs_leb_read(c
, lnum
, buf
, offs
, len
, 0);
1655 /* Don't read under wbuf */
1656 rlen
= wbuf
->offs
- offs
;
1660 /* Copy the rest from the write-buffer */
1661 memcpy(buf
+ rlen
, wbuf
->buf
+ offs
+ rlen
- wbuf
->offs
, len
- rlen
);
1662 spin_unlock(&wbuf
->lock
);
1665 /* Read everything that goes before write-buffer */
1666 return ubifs_leb_read(c
, lnum
, buf
, offs
, rlen
, 0);
1672 * validate_data_node - validate data nodes for bulk-read.
1673 * @c: UBIFS file-system description object
1674 * @buf: buffer containing data node to validate
1675 * @zbr: zbranch of data node to validate
1677 * This functions returns %0 on success or a negative error code on failure.
1679 static int validate_data_node(struct ubifs_info
*c
, void *buf
,
1680 struct ubifs_zbranch
*zbr
)
1682 union ubifs_key key1
;
1683 struct ubifs_ch
*ch
= buf
;
1686 if (ch
->node_type
!= UBIFS_DATA_NODE
) {
1687 ubifs_err("bad node type (%d but expected %d)",
1688 ch
->node_type
, UBIFS_DATA_NODE
);
1692 err
= ubifs_check_node(c
, buf
, zbr
->lnum
, zbr
->offs
, 0, 0);
1694 ubifs_err("expected node type %d", UBIFS_DATA_NODE
);
1698 len
= le32_to_cpu(ch
->len
);
1699 if (len
!= zbr
->len
) {
1700 ubifs_err("bad node length %d, expected %d", len
, zbr
->len
);
1704 /* Make sure the key of the read node is correct */
1705 key_read(c
, buf
+ UBIFS_KEY_OFFSET
, &key1
);
1706 if (!keys_eq(c
, &zbr
->key
, &key1
)) {
1707 ubifs_err("bad key in node at LEB %d:%d",
1708 zbr
->lnum
, zbr
->offs
);
1709 dbg_tnck(&zbr
->key
, "looked for key ");
1710 dbg_tnck(&key1
, "found node's key ");
1719 ubifs_err("bad node at LEB %d:%d", zbr
->lnum
, zbr
->offs
);
1720 ubifs_dump_node(c
, buf
);
1726 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1727 * @c: UBIFS file-system description object
1728 * @bu: bulk-read parameters and results
1730 * This functions reads and validates the data nodes that were identified by the
1731 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1732 * -EAGAIN to indicate a race with GC, or another negative error code on
1735 int ubifs_tnc_bulk_read(struct ubifs_info
*c
, struct bu_info
*bu
)
1737 int lnum
= bu
->zbranch
[0].lnum
, offs
= bu
->zbranch
[0].offs
, len
, err
, i
;
1738 struct ubifs_wbuf
*wbuf
;
1741 len
= bu
->zbranch
[bu
->cnt
- 1].offs
;
1742 len
+= bu
->zbranch
[bu
->cnt
- 1].len
- offs
;
1743 if (len
> bu
->buf_len
) {
1744 ubifs_err("buffer too small %d vs %d", bu
->buf_len
, len
);
1749 wbuf
= ubifs_get_wbuf(c
, lnum
);
1751 err
= read_wbuf(wbuf
, bu
->buf
, len
, lnum
, offs
);
1753 err
= ubifs_leb_read(c
, lnum
, bu
->buf
, offs
, len
, 0);
1755 /* Check for a race with GC */
1756 if (maybe_leb_gced(c
, lnum
, bu
->gc_seq
))
1759 if (err
&& err
!= -EBADMSG
) {
1760 ubifs_err("failed to read from LEB %d:%d, error %d",
1763 dbg_tnck(&bu
->key
, "key ");
1767 /* Validate the nodes read */
1769 for (i
= 0; i
< bu
->cnt
; i
++) {
1770 err
= validate_data_node(c
, buf
, &bu
->zbranch
[i
]);
1773 buf
= buf
+ ALIGN(bu
->zbranch
[i
].len
, 8);
1780 * do_lookup_nm- look up a "hashed" node.
1781 * @c: UBIFS file-system description object
1782 * @key: node key to lookup
1783 * @node: the node is returned here
1786 * This function look up and reads a node which contains name hash in the key.
1787 * Since the hash may have collisions, there may be many nodes with the same
1788 * key, so we have to sequentially look to all of them until the needed one is
1789 * found. This function returns zero in case of success, %-ENOENT if the node
1790 * was not found, and a negative error code in case of failure.
1792 static int do_lookup_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
1793 void *node
, const struct qstr
*nm
)
1796 struct ubifs_znode
*znode
;
1798 dbg_tnck(key
, "name '%.*s' key ", nm
->len
, nm
->name
);
1799 mutex_lock(&c
->tnc_mutex
);
1800 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
1804 } else if (found
< 0) {
1809 ubifs_assert(n
>= 0);
1811 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
1812 dbg_tnc("rc returned %d, znode %p, n %d", err
, znode
, n
);
1813 if (unlikely(err
< 0))
1820 err
= tnc_read_node_nm(c
, &znode
->zbranch
[n
], node
);
1823 mutex_unlock(&c
->tnc_mutex
);
1828 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1829 * @c: UBIFS file-system description object
1830 * @key: node key to lookup
1831 * @node: the node is returned here
1834 * This function look up and reads a node which contains name hash in the key.
1835 * Since the hash may have collisions, there may be many nodes with the same
1836 * key, so we have to sequentially look to all of them until the needed one is
1837 * found. This function returns zero in case of success, %-ENOENT if the node
1838 * was not found, and a negative error code in case of failure.
1840 int ubifs_tnc_lookup_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
1841 void *node
, const struct qstr
*nm
)
1844 const struct ubifs_dent_node
*dent
= node
;
1847 * We assume that in most of the cases there are no name collisions and
1848 * 'ubifs_tnc_lookup()' returns us the right direntry.
1850 err
= ubifs_tnc_lookup(c
, key
, node
);
1854 len
= le16_to_cpu(dent
->nlen
);
1855 if (nm
->len
== len
&& !memcmp(dent
->name
, nm
->name
, len
))
1859 * Unluckily, there are hash collisions and we have to iterate over
1860 * them look at each direntry with colliding name hash sequentially.
1862 return do_lookup_nm(c
, key
, node
, nm
);
1866 * correct_parent_keys - correct parent znodes' keys.
1867 * @c: UBIFS file-system description object
1868 * @znode: znode to correct parent znodes for
1870 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1871 * zbranch changes, keys of parent znodes have to be corrected. This helper
1872 * function is called in such situations and corrects the keys if needed.
1874 static void correct_parent_keys(const struct ubifs_info
*c
,
1875 struct ubifs_znode
*znode
)
1877 union ubifs_key
*key
, *key1
;
1879 ubifs_assert(znode
->parent
);
1880 ubifs_assert(znode
->iip
== 0);
1882 key
= &znode
->zbranch
[0].key
;
1883 key1
= &znode
->parent
->zbranch
[0].key
;
1885 while (keys_cmp(c
, key
, key1
) < 0) {
1886 key_copy(c
, key
, key1
);
1887 znode
= znode
->parent
;
1889 if (!znode
->parent
|| znode
->iip
)
1891 key1
= &znode
->parent
->zbranch
[0].key
;
1896 * insert_zbranch - insert a zbranch into a znode.
1897 * @znode: znode into which to insert
1898 * @zbr: zbranch to insert
1899 * @n: slot number to insert to
1901 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1902 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1903 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1904 * slot, zbranches starting from @n have to be moved right.
1906 static void insert_zbranch(struct ubifs_znode
*znode
,
1907 const struct ubifs_zbranch
*zbr
, int n
)
1911 ubifs_assert(ubifs_zn_dirty(znode
));
1914 for (i
= znode
->child_cnt
; i
> n
; i
--) {
1915 znode
->zbranch
[i
] = znode
->zbranch
[i
- 1];
1916 if (znode
->zbranch
[i
].znode
)
1917 znode
->zbranch
[i
].znode
->iip
= i
;
1920 zbr
->znode
->iip
= n
;
1922 for (i
= znode
->child_cnt
; i
> n
; i
--)
1923 znode
->zbranch
[i
] = znode
->zbranch
[i
- 1];
1925 znode
->zbranch
[n
] = *zbr
;
1926 znode
->child_cnt
+= 1;
1929 * After inserting at slot zero, the lower bound of the key range of
1930 * this znode may have changed. If this znode is subsequently split
1931 * then the upper bound of the key range may change, and furthermore
1932 * it could change to be lower than the original lower bound. If that
1933 * happens, then it will no longer be possible to find this znode in the
1934 * TNC using the key from the index node on flash. That is bad because
1935 * if it is not found, we will assume it is obsolete and may overwrite
1936 * it. Then if there is an unclean unmount, we will start using the
1937 * old index which will be broken.
1939 * So we first mark znodes that have insertions at slot zero, and then
1940 * if they are split we add their lnum/offs to the old_idx tree.
1947 * tnc_insert - insert a node into TNC.
1948 * @c: UBIFS file-system description object
1949 * @znode: znode to insert into
1950 * @zbr: branch to insert
1951 * @n: slot number to insert new zbranch to
1953 * This function inserts a new node described by @zbr into znode @znode. If
1954 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1955 * are splat as well if needed. Returns zero in case of success or a negative
1956 * error code in case of failure.
1958 static int tnc_insert(struct ubifs_info
*c
, struct ubifs_znode
*znode
,
1959 struct ubifs_zbranch
*zbr
, int n
)
1961 struct ubifs_znode
*zn
, *zi
, *zp
;
1962 int i
, keep
, move
, appending
= 0;
1963 union ubifs_key
*key
= &zbr
->key
, *key1
;
1965 ubifs_assert(n
>= 0 && n
<= c
->fanout
);
1967 /* Implement naive insert for now */
1970 if (znode
->child_cnt
< c
->fanout
) {
1971 ubifs_assert(n
!= c
->fanout
);
1972 dbg_tnck(key
, "inserted at %d level %d, key ", n
, znode
->level
);
1974 insert_zbranch(znode
, zbr
, n
);
1976 /* Ensure parent's key is correct */
1977 if (n
== 0 && zp
&& znode
->iip
== 0)
1978 correct_parent_keys(c
, znode
);
1984 * Unfortunately, @znode does not have more empty slots and we have to
1987 dbg_tnck(key
, "splitting level %d, key ", znode
->level
);
1991 * We can no longer be sure of finding this znode by key, so we
1992 * record it in the old_idx tree.
1994 ins_clr_old_idx_znode(c
, znode
);
1996 zn
= kzalloc(c
->max_znode_sz
, GFP_NOFS
);
2000 zn
->level
= znode
->level
;
2002 /* Decide where to split */
2003 if (znode
->level
== 0 && key_type(c
, key
) == UBIFS_DATA_KEY
) {
2004 /* Try not to split consecutive data keys */
2005 if (n
== c
->fanout
) {
2006 key1
= &znode
->zbranch
[n
- 1].key
;
2007 if (key_inum(c
, key1
) == key_inum(c
, key
) &&
2008 key_type(c
, key1
) == UBIFS_DATA_KEY
)
2012 } else if (appending
&& n
!= c
->fanout
) {
2013 /* Try not to split consecutive data keys */
2016 if (n
>= (c
->fanout
+ 1) / 2) {
2017 key1
= &znode
->zbranch
[0].key
;
2018 if (key_inum(c
, key1
) == key_inum(c
, key
) &&
2019 key_type(c
, key1
) == UBIFS_DATA_KEY
) {
2020 key1
= &znode
->zbranch
[n
].key
;
2021 if (key_inum(c
, key1
) != key_inum(c
, key
) ||
2022 key_type(c
, key1
) != UBIFS_DATA_KEY
) {
2024 move
= c
->fanout
- keep
;
2036 keep
= (c
->fanout
+ 1) / 2;
2037 move
= c
->fanout
- keep
;
2041 * Although we don't at present, we could look at the neighbors and see
2042 * if we can move some zbranches there.
2046 /* Insert into existing znode */
2051 /* Insert into new znode */
2056 zbr
->znode
->parent
= zn
;
2061 __set_bit(DIRTY_ZNODE
, &zn
->flags
);
2062 atomic_long_inc(&c
->dirty_zn_cnt
);
2064 zn
->child_cnt
= move
;
2065 znode
->child_cnt
= keep
;
2067 dbg_tnc("moving %d, keeping %d", move
, keep
);
2070 for (i
= 0; i
< move
; i
++) {
2071 zn
->zbranch
[i
] = znode
->zbranch
[keep
+ i
];
2074 if (zn
->zbranch
[i
].znode
) {
2075 zn
->zbranch
[i
].znode
->parent
= zn
;
2076 zn
->zbranch
[i
].znode
->iip
= i
;
2080 /* Insert new key and branch */
2081 dbg_tnck(key
, "inserting at %d level %d, key ", n
, zn
->level
);
2083 insert_zbranch(zi
, zbr
, n
);
2085 /* Insert new znode (produced by spitting) into the parent */
2087 if (n
== 0 && zi
== znode
&& znode
->iip
== 0)
2088 correct_parent_keys(c
, znode
);
2090 /* Locate insertion point */
2093 /* Tail recursion */
2094 zbr
->key
= zn
->zbranch
[0].key
;
2104 /* We have to split root znode */
2105 dbg_tnc("creating new zroot at level %d", znode
->level
+ 1);
2107 zi
= kzalloc(c
->max_znode_sz
, GFP_NOFS
);
2112 zi
->level
= znode
->level
+ 1;
2114 __set_bit(DIRTY_ZNODE
, &zi
->flags
);
2115 atomic_long_inc(&c
->dirty_zn_cnt
);
2117 zi
->zbranch
[0].key
= znode
->zbranch
[0].key
;
2118 zi
->zbranch
[0].znode
= znode
;
2119 zi
->zbranch
[0].lnum
= c
->zroot
.lnum
;
2120 zi
->zbranch
[0].offs
= c
->zroot
.offs
;
2121 zi
->zbranch
[0].len
= c
->zroot
.len
;
2122 zi
->zbranch
[1].key
= zn
->zbranch
[0].key
;
2123 zi
->zbranch
[1].znode
= zn
;
2128 c
->zroot
.znode
= zi
;
2139 * ubifs_tnc_add - add a node to TNC.
2140 * @c: UBIFS file-system description object
2142 * @lnum: LEB number of node
2143 * @offs: node offset
2146 * This function adds a node with key @key to TNC. The node may be new or it may
2147 * obsolete some existing one. Returns %0 on success or negative error code on
2150 int ubifs_tnc_add(struct ubifs_info
*c
, const union ubifs_key
*key
, int lnum
,
2153 int found
, n
, err
= 0;
2154 struct ubifs_znode
*znode
;
2156 mutex_lock(&c
->tnc_mutex
);
2157 dbg_tnck(key
, "%d:%d, len %d, key ", lnum
, offs
, len
);
2158 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2160 struct ubifs_zbranch zbr
;
2166 key_copy(c
, key
, &zbr
.key
);
2167 err
= tnc_insert(c
, znode
, &zbr
, n
+ 1);
2168 } else if (found
== 1) {
2169 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2172 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2179 err
= dbg_check_tnc(c
, 0);
2180 mutex_unlock(&c
->tnc_mutex
);
2186 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2187 * @c: UBIFS file-system description object
2189 * @old_lnum: LEB number of old node
2190 * @old_offs: old node offset
2191 * @lnum: LEB number of node
2192 * @offs: node offset
2195 * This function replaces a node with key @key in the TNC only if the old node
2196 * is found. This function is called by garbage collection when node are moved.
2197 * Returns %0 on success or negative error code on failure.
2199 int ubifs_tnc_replace(struct ubifs_info
*c
, const union ubifs_key
*key
,
2200 int old_lnum
, int old_offs
, int lnum
, int offs
, int len
)
2202 int found
, n
, err
= 0;
2203 struct ubifs_znode
*znode
;
2205 mutex_lock(&c
->tnc_mutex
);
2206 dbg_tnck(key
, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum
,
2207 old_offs
, lnum
, offs
, len
);
2208 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2215 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2218 if (zbr
->lnum
== old_lnum
&& zbr
->offs
== old_offs
) {
2220 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2227 } else if (is_hash_key(c
, key
)) {
2228 found
= resolve_collision_directly(c
, key
, &znode
, &n
,
2229 old_lnum
, old_offs
);
2230 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2231 found
, znode
, n
, old_lnum
, old_offs
);
2238 /* Ensure the znode is dirtied */
2239 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2240 znode
= dirty_cow_bottom_up(c
, znode
);
2241 if (IS_ERR(znode
)) {
2242 err
= PTR_ERR(znode
);
2246 zbr
= &znode
->zbranch
[n
];
2248 err
= ubifs_add_dirt(c
, zbr
->lnum
,
2260 err
= ubifs_add_dirt(c
, lnum
, len
);
2263 err
= dbg_check_tnc(c
, 0);
2266 mutex_unlock(&c
->tnc_mutex
);
2271 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2272 * @c: UBIFS file-system description object
2274 * @lnum: LEB number of node
2275 * @offs: node offset
2279 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2280 * may have collisions, like directory entry keys.
2282 int ubifs_tnc_add_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
2283 int lnum
, int offs
, int len
, const struct qstr
*nm
)
2285 int found
, n
, err
= 0;
2286 struct ubifs_znode
*znode
;
2288 mutex_lock(&c
->tnc_mutex
);
2289 dbg_tnck(key
, "LEB %d:%d, name '%.*s', key ",
2290 lnum
, offs
, nm
->len
, nm
->name
);
2291 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2299 found
= fallible_resolve_collision(c
, key
, &znode
, &n
,
2302 found
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2303 dbg_tnc("rc returned %d, znode %p, n %d", found
, znode
, n
);
2309 /* Ensure the znode is dirtied */
2310 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2311 znode
= dirty_cow_bottom_up(c
, znode
);
2312 if (IS_ERR(znode
)) {
2313 err
= PTR_ERR(znode
);
2319 struct ubifs_zbranch
*zbr
= &znode
->zbranch
[n
];
2322 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2331 struct ubifs_zbranch zbr
;
2337 key_copy(c
, key
, &zbr
.key
);
2338 err
= tnc_insert(c
, znode
, &zbr
, n
+ 1);
2343 * We did not find it in the index so there may be a
2344 * dangling branch still in the index. So we remove it
2345 * by passing 'ubifs_tnc_remove_nm()' the same key but
2346 * an unmatchable name.
2348 struct qstr noname
= { .name
= "" };
2350 err
= dbg_check_tnc(c
, 0);
2351 mutex_unlock(&c
->tnc_mutex
);
2354 return ubifs_tnc_remove_nm(c
, key
, &noname
);
2360 err
= dbg_check_tnc(c
, 0);
2361 mutex_unlock(&c
->tnc_mutex
);
2366 * tnc_delete - delete a znode form TNC.
2367 * @c: UBIFS file-system description object
2368 * @znode: znode to delete from
2369 * @n: zbranch slot number to delete
2371 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2372 * case of success and a negative error code in case of failure.
2374 static int tnc_delete(struct ubifs_info
*c
, struct ubifs_znode
*znode
, int n
)
2376 struct ubifs_zbranch
*zbr
;
2377 struct ubifs_znode
*zp
;
2380 /* Delete without merge for now */
2381 ubifs_assert(znode
->level
== 0);
2382 ubifs_assert(n
>= 0 && n
< c
->fanout
);
2383 dbg_tnck(&znode
->zbranch
[n
].key
, "deleting key ");
2385 zbr
= &znode
->zbranch
[n
];
2388 err
= ubifs_add_dirt(c
, zbr
->lnum
, zbr
->len
);
2390 ubifs_dump_znode(c
, znode
);
2394 /* We do not "gap" zbranch slots */
2395 for (i
= n
; i
< znode
->child_cnt
- 1; i
++)
2396 znode
->zbranch
[i
] = znode
->zbranch
[i
+ 1];
2397 znode
->child_cnt
-= 1;
2399 if (znode
->child_cnt
> 0)
2403 * This was the last zbranch, we have to delete this znode from the
2408 ubifs_assert(!ubifs_zn_obsolete(znode
));
2409 ubifs_assert(ubifs_zn_dirty(znode
));
2414 atomic_long_dec(&c
->dirty_zn_cnt
);
2416 err
= insert_old_idx_znode(c
, znode
);
2421 __set_bit(OBSOLETE_ZNODE
, &znode
->flags
);
2422 atomic_long_inc(&c
->clean_zn_cnt
);
2423 atomic_long_inc(&ubifs_clean_zn_cnt
);
2427 } while (znode
->child_cnt
== 1); /* while removing last child */
2429 /* Remove from znode, entry n - 1 */
2430 znode
->child_cnt
-= 1;
2431 ubifs_assert(znode
->level
!= 0);
2432 for (i
= n
; i
< znode
->child_cnt
; i
++) {
2433 znode
->zbranch
[i
] = znode
->zbranch
[i
+ 1];
2434 if (znode
->zbranch
[i
].znode
)
2435 znode
->zbranch
[i
].znode
->iip
= i
;
2439 * If this is the root and it has only 1 child then
2440 * collapse the tree.
2442 if (!znode
->parent
) {
2443 while (znode
->child_cnt
== 1 && znode
->level
!= 0) {
2445 zbr
= &znode
->zbranch
[0];
2446 znode
= get_znode(c
, znode
, 0);
2448 return PTR_ERR(znode
);
2449 znode
= dirty_cow_znode(c
, zbr
);
2451 return PTR_ERR(znode
);
2452 znode
->parent
= NULL
;
2455 err
= insert_old_idx(c
, c
->zroot
.lnum
,
2460 c
->zroot
.lnum
= zbr
->lnum
;
2461 c
->zroot
.offs
= zbr
->offs
;
2462 c
->zroot
.len
= zbr
->len
;
2463 c
->zroot
.znode
= znode
;
2464 ubifs_assert(!ubifs_zn_obsolete(zp
));
2465 ubifs_assert(ubifs_zn_dirty(zp
));
2466 atomic_long_dec(&c
->dirty_zn_cnt
);
2469 __set_bit(OBSOLETE_ZNODE
, &zp
->flags
);
2470 atomic_long_inc(&c
->clean_zn_cnt
);
2471 atomic_long_inc(&ubifs_clean_zn_cnt
);
2481 * ubifs_tnc_remove - remove an index entry of a node.
2482 * @c: UBIFS file-system description object
2485 * Returns %0 on success or negative error code on failure.
2487 int ubifs_tnc_remove(struct ubifs_info
*c
, const union ubifs_key
*key
)
2489 int found
, n
, err
= 0;
2490 struct ubifs_znode
*znode
;
2492 mutex_lock(&c
->tnc_mutex
);
2493 dbg_tnck(key
, "key ");
2494 found
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2500 err
= tnc_delete(c
, znode
, n
);
2502 err
= dbg_check_tnc(c
, 0);
2505 mutex_unlock(&c
->tnc_mutex
);
2510 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2511 * @c: UBIFS file-system description object
2513 * @nm: directory entry name
2515 * Returns %0 on success or negative error code on failure.
2517 int ubifs_tnc_remove_nm(struct ubifs_info
*c
, const union ubifs_key
*key
,
2518 const struct qstr
*nm
)
2521 struct ubifs_znode
*znode
;
2523 mutex_lock(&c
->tnc_mutex
);
2524 dbg_tnck(key
, "%.*s, key ", nm
->len
, nm
->name
);
2525 err
= lookup_level0_dirty(c
, key
, &znode
, &n
);
2531 err
= fallible_resolve_collision(c
, key
, &znode
, &n
,
2534 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2535 dbg_tnc("rc returned %d, znode %p, n %d", err
, znode
, n
);
2539 /* Ensure the znode is dirtied */
2540 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2541 znode
= dirty_cow_bottom_up(c
, znode
);
2542 if (IS_ERR(znode
)) {
2543 err
= PTR_ERR(znode
);
2547 err
= tnc_delete(c
, znode
, n
);
2553 err
= dbg_check_tnc(c
, 0);
2554 mutex_unlock(&c
->tnc_mutex
);
2559 * key_in_range - determine if a key falls within a range of keys.
2560 * @c: UBIFS file-system description object
2561 * @key: key to check
2562 * @from_key: lowest key in range
2563 * @to_key: highest key in range
2565 * This function returns %1 if the key is in range and %0 otherwise.
2567 static int key_in_range(struct ubifs_info
*c
, union ubifs_key
*key
,
2568 union ubifs_key
*from_key
, union ubifs_key
*to_key
)
2570 if (keys_cmp(c
, key
, from_key
) < 0)
2572 if (keys_cmp(c
, key
, to_key
) > 0)
2578 * ubifs_tnc_remove_range - remove index entries in range.
2579 * @c: UBIFS file-system description object
2580 * @from_key: lowest key to remove
2581 * @to_key: highest key to remove
2583 * This function removes index entries starting at @from_key and ending at
2584 * @to_key. This function returns zero in case of success and a negative error
2585 * code in case of failure.
2587 int ubifs_tnc_remove_range(struct ubifs_info
*c
, union ubifs_key
*from_key
,
2588 union ubifs_key
*to_key
)
2590 int i
, n
, k
, err
= 0;
2591 struct ubifs_znode
*znode
;
2592 union ubifs_key
*key
;
2594 mutex_lock(&c
->tnc_mutex
);
2596 /* Find first level 0 znode that contains keys to remove */
2597 err
= ubifs_lookup_level0(c
, from_key
, &znode
, &n
);
2604 err
= tnc_next(c
, &znode
, &n
);
2605 if (err
== -ENOENT
) {
2611 key
= &znode
->zbranch
[n
].key
;
2612 if (!key_in_range(c
, key
, from_key
, to_key
)) {
2618 /* Ensure the znode is dirtied */
2619 if (znode
->cnext
|| !ubifs_zn_dirty(znode
)) {
2620 znode
= dirty_cow_bottom_up(c
, znode
);
2621 if (IS_ERR(znode
)) {
2622 err
= PTR_ERR(znode
);
2627 /* Remove all keys in range except the first */
2628 for (i
= n
+ 1, k
= 0; i
< znode
->child_cnt
; i
++, k
++) {
2629 key
= &znode
->zbranch
[i
].key
;
2630 if (!key_in_range(c
, key
, from_key
, to_key
))
2632 lnc_free(&znode
->zbranch
[i
]);
2633 err
= ubifs_add_dirt(c
, znode
->zbranch
[i
].lnum
,
2634 znode
->zbranch
[i
].len
);
2636 ubifs_dump_znode(c
, znode
);
2639 dbg_tnck(key
, "removing key ");
2642 for (i
= n
+ 1 + k
; i
< znode
->child_cnt
; i
++)
2643 znode
->zbranch
[i
- k
] = znode
->zbranch
[i
];
2644 znode
->child_cnt
-= k
;
2647 /* Now delete the first */
2648 err
= tnc_delete(c
, znode
, n
);
2655 err
= dbg_check_tnc(c
, 0);
2656 mutex_unlock(&c
->tnc_mutex
);
2661 * ubifs_tnc_remove_ino - remove an inode from TNC.
2662 * @c: UBIFS file-system description object
2663 * @inum: inode number to remove
2665 * This function remove inode @inum and all the extended attributes associated
2666 * with the anode from TNC and returns zero in case of success or a negative
2667 * error code in case of failure.
2669 int ubifs_tnc_remove_ino(struct ubifs_info
*c
, ino_t inum
)
2671 union ubifs_key key1
, key2
;
2672 struct ubifs_dent_node
*xent
, *pxent
= NULL
;
2673 struct qstr nm
= { .name
= NULL
};
2675 dbg_tnc("ino %lu", (unsigned long)inum
);
2678 * Walk all extended attribute entries and remove them together with
2679 * corresponding extended attribute inodes.
2681 lowest_xent_key(c
, &key1
, inum
);
2686 xent
= ubifs_tnc_next_ent(c
, &key1
, &nm
);
2688 err
= PTR_ERR(xent
);
2694 xattr_inum
= le64_to_cpu(xent
->inum
);
2695 dbg_tnc("xent '%s', ino %lu", xent
->name
,
2696 (unsigned long)xattr_inum
);
2698 nm
.name
= xent
->name
;
2699 nm
.len
= le16_to_cpu(xent
->nlen
);
2700 err
= ubifs_tnc_remove_nm(c
, &key1
, &nm
);
2706 lowest_ino_key(c
, &key1
, xattr_inum
);
2707 highest_ino_key(c
, &key2
, xattr_inum
);
2708 err
= ubifs_tnc_remove_range(c
, &key1
, &key2
);
2716 key_read(c
, &xent
->key
, &key1
);
2720 lowest_ino_key(c
, &key1
, inum
);
2721 highest_ino_key(c
, &key2
, inum
);
2723 return ubifs_tnc_remove_range(c
, &key1
, &key2
);
2727 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2728 * @c: UBIFS file-system description object
2729 * @key: key of last entry
2730 * @nm: name of last entry found or %NULL
2732 * This function finds and reads the next directory or extended attribute entry
2733 * after the given key (@key) if there is one. @nm is used to resolve
2736 * If the name of the current entry is not known and only the key is known,
2737 * @nm->name has to be %NULL. In this case the semantics of this function is a
2738 * little bit different and it returns the entry corresponding to this key, not
2739 * the next one. If the key was not found, the closest "right" entry is
2742 * If the fist entry has to be found, @key has to contain the lowest possible
2743 * key value for this inode and @name has to be %NULL.
2745 * This function returns the found directory or extended attribute entry node
2746 * in case of success, %-ENOENT is returned if no entry was found, and a
2747 * negative error code is returned in case of failure.
2749 struct ubifs_dent_node
*ubifs_tnc_next_ent(struct ubifs_info
*c
,
2750 union ubifs_key
*key
,
2751 const struct qstr
*nm
)
2753 int n
, err
, type
= key_type(c
, key
);
2754 struct ubifs_znode
*znode
;
2755 struct ubifs_dent_node
*dent
;
2756 struct ubifs_zbranch
*zbr
;
2757 union ubifs_key
*dkey
;
2759 dbg_tnck(key
, "%s ", nm
->name
? (char *)nm
->name
: "(lowest)");
2760 ubifs_assert(is_hash_key(c
, key
));
2762 mutex_lock(&c
->tnc_mutex
);
2763 err
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
2764 if (unlikely(err
< 0))
2769 /* Handle collisions */
2770 err
= resolve_collision(c
, key
, &znode
, &n
, nm
);
2771 dbg_tnc("rc returned %d, znode %p, n %d",
2773 if (unlikely(err
< 0))
2777 /* Now find next entry */
2778 err
= tnc_next(c
, &znode
, &n
);
2783 * The full name of the entry was not given, in which case the
2784 * behavior of this function is a little different and it
2785 * returns current entry, not the next one.
2789 * However, the given key does not exist in the TNC
2790 * tree and @znode/@n variables contain the closest
2791 * "preceding" element. Switch to the next one.
2793 err
= tnc_next(c
, &znode
, &n
);
2799 zbr
= &znode
->zbranch
[n
];
2800 dent
= kmalloc(zbr
->len
, GFP_NOFS
);
2801 if (unlikely(!dent
)) {
2807 * The above 'tnc_next()' call could lead us to the next inode, check
2811 if (key_inum(c
, dkey
) != key_inum(c
, key
) ||
2812 key_type(c
, dkey
) != type
) {
2817 err
= tnc_read_node_nm(c
, zbr
, dent
);
2821 mutex_unlock(&c
->tnc_mutex
);
2827 mutex_unlock(&c
->tnc_mutex
);
2828 return ERR_PTR(err
);
2832 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2833 * @c: UBIFS file-system description object
2835 * Destroy left-over obsolete znodes from a failed commit.
2837 static void tnc_destroy_cnext(struct ubifs_info
*c
)
2839 struct ubifs_znode
*cnext
;
2843 ubifs_assert(c
->cmt_state
== COMMIT_BROKEN
);
2846 struct ubifs_znode
*znode
= cnext
;
2848 cnext
= cnext
->cnext
;
2849 if (ubifs_zn_obsolete(znode
))
2851 } while (cnext
&& cnext
!= c
->cnext
);
2855 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2856 * @c: UBIFS file-system description object
2858 void ubifs_tnc_close(struct ubifs_info
*c
)
2860 tnc_destroy_cnext(c
);
2861 if (c
->zroot
.znode
) {
2864 ubifs_destroy_tnc_subtree(c
->zroot
.znode
);
2865 n
= atomic_long_read(&c
->clean_zn_cnt
);
2866 atomic_long_sub(n
, &ubifs_clean_zn_cnt
);
2874 * left_znode - get the znode to the left.
2875 * @c: UBIFS file-system description object
2878 * This function returns a pointer to the znode to the left of @znode or NULL if
2879 * there is not one. A negative error code is returned on failure.
2881 static struct ubifs_znode
*left_znode(struct ubifs_info
*c
,
2882 struct ubifs_znode
*znode
)
2884 int level
= znode
->level
;
2887 int n
= znode
->iip
- 1;
2889 /* Go up until we can go left */
2890 znode
= znode
->parent
;
2894 /* Now go down the rightmost branch to 'level' */
2895 znode
= get_znode(c
, znode
, n
);
2898 while (znode
->level
!= level
) {
2899 n
= znode
->child_cnt
- 1;
2900 znode
= get_znode(c
, znode
, n
);
2911 * right_znode - get the znode to the right.
2912 * @c: UBIFS file-system description object
2915 * This function returns a pointer to the znode to the right of @znode or NULL
2916 * if there is not one. A negative error code is returned on failure.
2918 static struct ubifs_znode
*right_znode(struct ubifs_info
*c
,
2919 struct ubifs_znode
*znode
)
2921 int level
= znode
->level
;
2924 int n
= znode
->iip
+ 1;
2926 /* Go up until we can go right */
2927 znode
= znode
->parent
;
2930 if (n
< znode
->child_cnt
) {
2931 /* Now go down the leftmost branch to 'level' */
2932 znode
= get_znode(c
, znode
, n
);
2935 while (znode
->level
!= level
) {
2936 znode
= get_znode(c
, znode
, 0);
2947 * lookup_znode - find a particular indexing node from TNC.
2948 * @c: UBIFS file-system description object
2949 * @key: index node key to lookup
2950 * @level: index node level
2951 * @lnum: index node LEB number
2952 * @offs: index node offset
2954 * This function searches an indexing node by its first key @key and its
2955 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2956 * nodes it traverses to TNC. This function is called for indexing nodes which
2957 * were found on the media by scanning, for example when garbage-collecting or
2958 * when doing in-the-gaps commit. This means that the indexing node which is
2959 * looked for does not have to have exactly the same leftmost key @key, because
2960 * the leftmost key may have been changed, in which case TNC will contain a
2961 * dirty znode which still refers the same @lnum:@offs. This function is clever
2962 * enough to recognize such indexing nodes.
2964 * Note, if a znode was deleted or changed too much, then this function will
2965 * not find it. For situations like this UBIFS has the old index RB-tree
2966 * (indexed by @lnum:@offs).
2968 * This function returns a pointer to the znode found or %NULL if it is not
2969 * found. A negative error code is returned on failure.
2971 static struct ubifs_znode
*lookup_znode(struct ubifs_info
*c
,
2972 union ubifs_key
*key
, int level
,
2975 struct ubifs_znode
*znode
, *zn
;
2978 ubifs_assert(key_type(c
, key
) < UBIFS_INVALID_KEY
);
2981 * The arguments have probably been read off flash, so don't assume
2985 return ERR_PTR(-EINVAL
);
2987 /* Get the root znode */
2988 znode
= c
->zroot
.znode
;
2990 znode
= ubifs_load_znode(c
, &c
->zroot
, NULL
, 0);
2994 /* Check if it is the one we are looking for */
2995 if (c
->zroot
.lnum
== lnum
&& c
->zroot
.offs
== offs
)
2997 /* Descend to the parent level i.e. (level + 1) */
2998 if (level
>= znode
->level
)
3001 ubifs_search_zbranch(c
, znode
, key
, &n
);
3004 * We reached a znode where the leftmost key is greater
3005 * than the key we are searching for. This is the same
3006 * situation as the one described in a huge comment at
3007 * the end of the 'ubifs_lookup_level0()' function. And
3008 * for exactly the same reasons we have to try to look
3009 * left before giving up.
3011 znode
= left_znode(c
, znode
);
3016 ubifs_search_zbranch(c
, znode
, key
, &n
);
3017 ubifs_assert(n
>= 0);
3019 if (znode
->level
== level
+ 1)
3021 znode
= get_znode(c
, znode
, n
);
3025 /* Check if the child is the one we are looking for */
3026 if (znode
->zbranch
[n
].lnum
== lnum
&& znode
->zbranch
[n
].offs
== offs
)
3027 return get_znode(c
, znode
, n
);
3028 /* If the key is unique, there is nowhere else to look */
3029 if (!is_hash_key(c
, key
))
3032 * The key is not unique and so may be also in the znodes to either
3039 /* Move one branch to the left */
3043 znode
= left_znode(c
, znode
);
3048 n
= znode
->child_cnt
- 1;
3051 if (znode
->zbranch
[n
].lnum
== lnum
&&
3052 znode
->zbranch
[n
].offs
== offs
)
3053 return get_znode(c
, znode
, n
);
3054 /* Stop if the key is less than the one we are looking for */
3055 if (keys_cmp(c
, &znode
->zbranch
[n
].key
, key
) < 0)
3058 /* Back to the middle */
3063 /* Move one branch to the right */
3064 if (++n
>= znode
->child_cnt
) {
3065 znode
= right_znode(c
, znode
);
3073 if (znode
->zbranch
[n
].lnum
== lnum
&&
3074 znode
->zbranch
[n
].offs
== offs
)
3075 return get_znode(c
, znode
, n
);
3076 /* Stop if the key is greater than the one we are looking for */
3077 if (keys_cmp(c
, &znode
->zbranch
[n
].key
, key
) > 0)
3084 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3085 * @c: UBIFS file-system description object
3086 * @key: key of index node
3087 * @level: index node level
3088 * @lnum: LEB number of index node
3089 * @offs: offset of index node
3091 * This function returns %0 if the index node is not referred to in the TNC, %1
3092 * if the index node is referred to in the TNC and the corresponding znode is
3093 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3094 * znode is clean, and a negative error code in case of failure.
3096 * Note, the @key argument has to be the key of the first child. Also note,
3097 * this function relies on the fact that 0:0 is never a valid LEB number and
3098 * offset for a main-area node.
3100 int is_idx_node_in_tnc(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3103 struct ubifs_znode
*znode
;
3105 znode
= lookup_znode(c
, key
, level
, lnum
, offs
);
3109 return PTR_ERR(znode
);
3111 return ubifs_zn_dirty(znode
) ? 1 : 2;
3115 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3116 * @c: UBIFS file-system description object
3118 * @lnum: node LEB number
3119 * @offs: node offset
3121 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3122 * not, and a negative error code in case of failure.
3124 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3125 * and offset for a main-area node.
3127 static int is_leaf_node_in_tnc(struct ubifs_info
*c
, union ubifs_key
*key
,
3130 struct ubifs_zbranch
*zbr
;
3131 struct ubifs_znode
*znode
, *zn
;
3132 int n
, found
, err
, nn
;
3133 const int unique
= !is_hash_key(c
, key
);
3135 found
= ubifs_lookup_level0(c
, key
, &znode
, &n
);
3137 return found
; /* Error code */
3140 zbr
= &znode
->zbranch
[n
];
3141 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3142 return 1; /* Found it */
3146 * Because the key is not unique, we have to look left
3153 err
= tnc_prev(c
, &znode
, &n
);
3158 if (keys_cmp(c
, key
, &znode
->zbranch
[n
].key
))
3160 zbr
= &znode
->zbranch
[n
];
3161 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3162 return 1; /* Found it */
3168 err
= tnc_next(c
, &znode
, &n
);
3174 if (keys_cmp(c
, key
, &znode
->zbranch
[n
].key
))
3176 zbr
= &znode
->zbranch
[n
];
3177 if (lnum
== zbr
->lnum
&& offs
== zbr
->offs
)
3178 return 1; /* Found it */
3184 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3185 * @c: UBIFS file-system description object
3187 * @level: index node level (if it is an index node)
3188 * @lnum: node LEB number
3189 * @offs: node offset
3190 * @is_idx: non-zero if the node is an index node
3192 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3193 * negative error code in case of failure. For index nodes, @key has to be the
3194 * key of the first child. An index node is considered to be in the TNC only if
3195 * the corresponding znode is clean or has not been loaded.
3197 int ubifs_tnc_has_node(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3198 int lnum
, int offs
, int is_idx
)
3202 mutex_lock(&c
->tnc_mutex
);
3204 err
= is_idx_node_in_tnc(c
, key
, level
, lnum
, offs
);
3208 /* The index node was found but it was dirty */
3211 /* The index node was found and it was clean */
3216 err
= is_leaf_node_in_tnc(c
, key
, lnum
, offs
);
3219 mutex_unlock(&c
->tnc_mutex
);
3224 * ubifs_dirty_idx_node - dirty an index node.
3225 * @c: UBIFS file-system description object
3226 * @key: index node key
3227 * @level: index node level
3228 * @lnum: index node LEB number
3229 * @offs: index node offset
3231 * This function loads and dirties an index node so that it can be garbage
3232 * collected. The @key argument has to be the key of the first child. This
3233 * function relies on the fact that 0:0 is never a valid LEB number and offset
3234 * for a main-area node. Returns %0 on success and a negative error code on
3237 int ubifs_dirty_idx_node(struct ubifs_info
*c
, union ubifs_key
*key
, int level
,
3240 struct ubifs_znode
*znode
;
3243 mutex_lock(&c
->tnc_mutex
);
3244 znode
= lookup_znode(c
, key
, level
, lnum
, offs
);
3247 if (IS_ERR(znode
)) {
3248 err
= PTR_ERR(znode
);
3251 znode
= dirty_cow_bottom_up(c
, znode
);
3252 if (IS_ERR(znode
)) {
3253 err
= PTR_ERR(znode
);
3258 mutex_unlock(&c
->tnc_mutex
);
3263 * dbg_check_inode_size - check if inode size is correct.
3264 * @c: UBIFS file-system description object
3265 * @inum: inode number
3268 * This function makes sure that the inode size (@size) is correct and it does
3269 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3270 * if it has a data page beyond @size, and other negative error code in case of
3273 int dbg_check_inode_size(struct ubifs_info
*c
, const struct inode
*inode
,
3277 union ubifs_key from_key
, to_key
, *key
;
3278 struct ubifs_znode
*znode
;
3281 if (!S_ISREG(inode
->i_mode
))
3283 if (!dbg_is_chk_gen(c
))
3286 block
= (size
+ UBIFS_BLOCK_SIZE
- 1) >> UBIFS_BLOCK_SHIFT
;
3287 data_key_init(c
, &from_key
, inode
->i_ino
, block
);
3288 highest_data_key(c
, &to_key
, inode
->i_ino
);
3290 mutex_lock(&c
->tnc_mutex
);
3291 err
= ubifs_lookup_level0(c
, &from_key
, &znode
, &n
);
3301 err
= tnc_next(c
, &znode
, &n
);
3302 if (err
== -ENOENT
) {
3309 ubifs_assert(err
== 0);
3310 key
= &znode
->zbranch
[n
].key
;
3311 if (!key_in_range(c
, key
, &from_key
, &to_key
))
3315 block
= key_block(c
, key
);
3316 ubifs_err("inode %lu has size %lld, but there are data at offset %lld",
3317 (unsigned long)inode
->i_ino
, size
,
3318 ((loff_t
)block
) << UBIFS_BLOCK_SHIFT
);
3319 mutex_unlock(&c
->tnc_mutex
);
3320 ubifs_dump_inode(c
, inode
);
3325 mutex_unlock(&c
->tnc_mutex
);