2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * SPDX-License-Identifier: GPL-2.0+
8 * Authors: Adrian Hunter
9 * Artem Bityutskiy (Битюцкий Артём)
13 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
16 * At the moment the locking rules of the TNC tree are quite simple and
17 * straightforward. We just have a mutex and lock it when we traverse the
18 * tree. If a znode is not in memory, we read it from flash while still having
23 #include <linux/crc32.h>
24 #include <linux/slab.h>
26 #include <linux/compat.h>
27 #include <linux/err.h>
28 #include <linux/stat.h>
33 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
34 * @NAME_LESS: name corresponding to the first argument is less than second
35 * @NAME_MATCHES: names match
36 * @NAME_GREATER: name corresponding to the second argument is greater than
38 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
40 * These constants were introduce to improve readability.
50 * insert_old_idx - record an index node obsoleted since the last commit start.
51 * @c: UBIFS file-system description object
52 * @lnum: LEB number of obsoleted index node
53 * @offs: offset of obsoleted index node
55 * Returns %0 on success, and a negative error code on failure.
57 * For recovery, there must always be a complete intact version of the index on
58 * flash at all times. That is called the "old index". It is the index as at the
59 * time of the last successful commit. Many of the index nodes in the old index
60 * may be dirty, but they must not be erased until the next successful commit
61 * (at which point that index becomes the old index).
63 * That means that the garbage collection and the in-the-gaps method of
64 * committing must be able to determine if an index node is in the old index.
65 * Most of the old index nodes can be found by looking up the TNC using the
66 * 'lookup_znode()' function. However, some of the old index nodes may have
67 * been deleted from the current index or may have been changed so much that
68 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
69 * That is what this function does. The RB-tree is ordered by LEB number and
70 * offset because they uniquely identify the old index node.
72 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
74 struct ubifs_old_idx *old_idx, *o;
75 struct rb_node **p, *parent = NULL;
77 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
78 if (unlikely(!old_idx))
83 p = &c->old_idx.rb_node;
86 o = rb_entry(parent, struct ubifs_old_idx, rb);
89 else if (lnum > o->lnum)
91 else if (offs < o->offs)
93 else if (offs > o->offs)
96 ubifs_err("old idx added twice!");
101 rb_link_node(&old_idx->rb, parent, p);
102 rb_insert_color(&old_idx->rb, &c->old_idx);
107 * insert_old_idx_znode - record a znode obsoleted since last commit start.
108 * @c: UBIFS file-system description object
109 * @znode: znode of obsoleted index node
111 * Returns %0 on success, and a negative error code on failure.
113 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
116 struct ubifs_zbranch *zbr;
118 zbr = &znode->parent->zbranch[znode->iip];
120 return insert_old_idx(c, zbr->lnum, zbr->offs);
123 return insert_old_idx(c, c->zroot.lnum,
129 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
130 * @c: UBIFS file-system description object
131 * @znode: znode of obsoleted index node
133 * Returns %0 on success, and a negative error code on failure.
135 static int ins_clr_old_idx_znode(struct ubifs_info *c,
136 struct ubifs_znode *znode)
141 struct ubifs_zbranch *zbr;
143 zbr = &znode->parent->zbranch[znode->iip];
145 err = insert_old_idx(c, zbr->lnum, zbr->offs);
154 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
165 * destroy_old_idx - destroy the old_idx RB-tree.
166 * @c: UBIFS file-system description object
168 * During start commit, the old_idx RB-tree is used to avoid overwriting index
169 * nodes that were in the index last commit but have since been deleted. This
170 * is necessary for recovery i.e. the old index must be kept intact until the
171 * new index is successfully written. The old-idx RB-tree is used for the
172 * in-the-gaps method of writing index nodes and is destroyed every commit.
174 void destroy_old_idx(struct ubifs_info *c)
176 struct ubifs_old_idx *old_idx, *n;
178 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
181 c->old_idx = RB_ROOT;
185 * copy_znode - copy a dirty znode.
186 * @c: UBIFS file-system description object
187 * @znode: znode to copy
189 * A dirty znode being committed may not be changed, so it is copied.
191 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
192 struct ubifs_znode *znode)
194 struct ubifs_znode *zn;
196 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
198 return ERR_PTR(-ENOMEM);
200 memcpy(zn, znode, c->max_znode_sz);
202 __set_bit(DIRTY_ZNODE, &zn->flags);
203 __clear_bit(COW_ZNODE, &zn->flags);
205 ubifs_assert(!ubifs_zn_obsolete(znode));
206 __set_bit(OBSOLETE_ZNODE, &znode->flags);
208 if (znode->level != 0) {
210 const int n = zn->child_cnt;
212 /* The children now have new parent */
213 for (i = 0; i < n; i++) {
214 struct ubifs_zbranch *zbr = &zn->zbranch[i];
217 zbr->znode->parent = zn;
221 atomic_long_inc(&c->dirty_zn_cnt);
226 * add_idx_dirt - add dirt due to a dirty znode.
227 * @c: UBIFS file-system description object
228 * @lnum: LEB number of index node
229 * @dirt: size of index node
231 * This function updates lprops dirty space and the new size of the index.
233 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
235 c->calc_idx_sz -= ALIGN(dirt, 8);
236 return ubifs_add_dirt(c, lnum, dirt);
240 * dirty_cow_znode - ensure a znode is not being committed.
241 * @c: UBIFS file-system description object
242 * @zbr: branch of znode to check
244 * Returns dirtied znode on success or negative error code on failure.
246 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
247 struct ubifs_zbranch *zbr)
249 struct ubifs_znode *znode = zbr->znode;
250 struct ubifs_znode *zn;
253 if (!ubifs_zn_cow(znode)) {
254 /* znode is not being committed */
255 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
256 atomic_long_inc(&c->dirty_zn_cnt);
257 atomic_long_dec(&c->clean_zn_cnt);
258 atomic_long_dec(&ubifs_clean_zn_cnt);
259 err = add_idx_dirt(c, zbr->lnum, zbr->len);
266 zn = copy_znode(c, znode);
271 err = insert_old_idx(c, zbr->lnum, zbr->offs);
274 err = add_idx_dirt(c, zbr->lnum, zbr->len);
289 * lnc_add - add a leaf node to the leaf node cache.
290 * @c: UBIFS file-system description object
291 * @zbr: zbranch of leaf node
294 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
295 * purpose of the leaf node cache is to save re-reading the same leaf node over
296 * and over again. Most things are cached by VFS, however the file system must
297 * cache directory entries for readdir and for resolving hash collisions. The
298 * present implementation of the leaf node cache is extremely simple, and
299 * allows for error returns that are not used but that may be needed if a more
300 * complex implementation is created.
302 * Note, this function does not add the @node object to LNC directly, but
303 * allocates a copy of the object and adds the copy to LNC. The reason for this
304 * is that @node has been allocated outside of the TNC subsystem and will be
305 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
306 * may be changed at any time, e.g. freed by the shrinker.
308 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
313 const struct ubifs_dent_node *dent = node;
315 ubifs_assert(!zbr->leaf);
316 ubifs_assert(zbr->len != 0);
317 ubifs_assert(is_hash_key(c, &zbr->key));
319 err = ubifs_validate_entry(c, dent);
322 ubifs_dump_node(c, dent);
326 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
328 /* We don't have to have the cache, so no error */
331 zbr->leaf = lnc_node;
336 * lnc_add_directly - add a leaf node to the leaf-node-cache.
337 * @c: UBIFS file-system description object
338 * @zbr: zbranch of leaf node
341 * This function is similar to 'lnc_add()', but it does not create a copy of
342 * @node but inserts @node to TNC directly.
344 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
349 ubifs_assert(!zbr->leaf);
350 ubifs_assert(zbr->len != 0);
352 err = ubifs_validate_entry(c, node);
355 ubifs_dump_node(c, node);
364 * lnc_free - remove a leaf node from the leaf node cache.
365 * @zbr: zbranch of leaf node
368 static void lnc_free(struct ubifs_zbranch *zbr)
377 * tnc_read_node_nm - read a "hashed" leaf node.
378 * @c: UBIFS file-system description object
379 * @zbr: key and position of the node
380 * @node: node is returned here
382 * This function reads a "hashed" node defined by @zbr from the leaf node cache
383 * (in it is there) or from the hash media, in which case the node is also
384 * added to LNC. Returns zero in case of success or a negative negative error
385 * code in case of failure.
387 static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
392 ubifs_assert(is_hash_key(c, &zbr->key));
395 /* Read from the leaf node cache */
396 ubifs_assert(zbr->len != 0);
397 memcpy(node, zbr->leaf, zbr->len);
401 err = ubifs_tnc_read_node(c, zbr, node);
405 /* Add the node to the leaf node cache */
406 err = lnc_add(c, zbr, node);
411 * try_read_node - read a node if it is a node.
412 * @c: UBIFS file-system description object
413 * @buf: buffer to read to
415 * @len: node length (not aligned)
416 * @lnum: LEB number of node to read
417 * @offs: offset of node to read
419 * This function tries to read a node of known type and length, checks it and
420 * stores it in @buf. This function returns %1 if a node is present and %0 if
421 * a node is not present. A negative error code is returned for I/O errors.
422 * This function performs that same function as ubifs_read_node except that
423 * it does not require that there is actually a node present and instead
424 * the return code indicates if a node was read.
426 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
427 * is true (it is controlled by corresponding mount option). However, if
428 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
429 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
430 * because during mounting or re-mounting from R/O mode to R/W mode we may read
431 * journal nodes (when replying the journal or doing the recovery) and the
432 * journal nodes may potentially be corrupted, so checking is required.
434 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
435 int len, int lnum, int offs)
438 struct ubifs_ch *ch = buf;
439 uint32_t crc, node_crc;
441 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
443 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
445 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
446 type, lnum, offs, err);
450 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
453 if (ch->node_type != type)
456 node_len = le32_to_cpu(ch->len);
460 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
464 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
465 node_crc = le32_to_cpu(ch->crc);
473 * fallible_read_node - try to read a leaf node.
474 * @c: UBIFS file-system description object
475 * @key: key of node to read
476 * @zbr: position of node
477 * @node: node returned
479 * This function tries to read a node and returns %1 if the node is read, %0
480 * if the node is not present, and a negative error code in the case of error.
482 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
483 struct ubifs_zbranch *zbr, void *node)
487 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
489 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
492 union ubifs_key node_key;
493 struct ubifs_dent_node *dent = node;
495 /* All nodes have key in the same place */
496 key_read(c, &dent->key, &node_key);
497 if (keys_cmp(c, key, &node_key) != 0)
500 if (ret == 0 && c->replaying)
501 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
502 zbr->lnum, zbr->offs, zbr->len);
507 * matches_name - determine if a direntry or xattr entry matches a given name.
508 * @c: UBIFS file-system description object
509 * @zbr: zbranch of dent
512 * This function checks if xentry/direntry referred by zbranch @zbr matches name
513 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
514 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
515 * of failure, a negative error code is returned.
517 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
518 const struct qstr *nm)
520 struct ubifs_dent_node *dent;
523 /* If possible, match against the dent in the leaf node cache */
525 dent = kmalloc(zbr->len, GFP_NOFS);
529 err = ubifs_tnc_read_node(c, zbr, dent);
533 /* Add the node to the leaf node cache */
534 err = lnc_add_directly(c, zbr, dent);
540 nlen = le16_to_cpu(dent->nlen);
541 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
545 else if (nlen < nm->len)
560 * get_znode - get a TNC znode that may not be loaded yet.
561 * @c: UBIFS file-system description object
562 * @znode: parent znode
563 * @n: znode branch slot number
565 * This function returns the znode or a negative error code.
567 static struct ubifs_znode *get_znode(struct ubifs_info *c,
568 struct ubifs_znode *znode, int n)
570 struct ubifs_zbranch *zbr;
572 zbr = &znode->zbranch[n];
576 znode = ubifs_load_znode(c, zbr, znode, n);
581 * tnc_next - find next TNC entry.
582 * @c: UBIFS file-system description object
583 * @zn: znode is passed and returned here
584 * @n: znode branch slot number is passed and returned here
586 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
587 * no next entry, or a negative error code otherwise.
589 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
591 struct ubifs_znode *znode = *zn;
595 if (nn < znode->child_cnt) {
600 struct ubifs_znode *zp;
607 if (nn < znode->child_cnt) {
608 znode = get_znode(c, znode, nn);
610 return PTR_ERR(znode);
611 while (znode->level != 0) {
612 znode = get_znode(c, znode, 0);
614 return PTR_ERR(znode);
626 * tnc_prev - find previous TNC entry.
627 * @c: UBIFS file-system description object
628 * @zn: znode is returned here
629 * @n: znode branch slot number is passed and returned here
631 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
632 * there is no next entry, or a negative error code otherwise.
634 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
636 struct ubifs_znode *znode = *zn;
644 struct ubifs_znode *zp;
652 znode = get_znode(c, znode, nn);
654 return PTR_ERR(znode);
655 while (znode->level != 0) {
656 nn = znode->child_cnt - 1;
657 znode = get_znode(c, znode, nn);
659 return PTR_ERR(znode);
661 nn = znode->child_cnt - 1;
671 * resolve_collision - resolve a collision.
672 * @c: UBIFS file-system description object
673 * @key: key of a directory or extended attribute entry
674 * @zn: znode is returned here
675 * @n: zbranch number is passed and returned here
676 * @nm: name of the entry
678 * This function is called for "hashed" keys to make sure that the found key
679 * really corresponds to the looked up node (directory or extended attribute
680 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
681 * %0 is returned if @nm is not found and @zn and @n are set to the previous
682 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
683 * This means that @n may be set to %-1 if the leftmost key in @zn is the
684 * previous one. A negative error code is returned on failures.
686 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
687 struct ubifs_znode **zn, int *n,
688 const struct qstr *nm)
692 err = matches_name(c, &(*zn)->zbranch[*n], nm);
693 if (unlikely(err < 0))
695 if (err == NAME_MATCHES)
698 if (err == NAME_GREATER) {
701 err = tnc_prev(c, zn, n);
702 if (err == -ENOENT) {
703 ubifs_assert(*n == 0);
709 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
711 * We have found the branch after which we would
712 * like to insert, but inserting in this znode
713 * may still be wrong. Consider the following 3
714 * znodes, in the case where we are resolving a
715 * collision with Key2.
718 * ----------------------
719 * level 1 | Key0 | Key1 |
720 * -----------------------
722 * znode za | | znode zb
723 * ------------ ------------
724 * level 0 | Key0 | | Key2 |
725 * ------------ ------------
727 * The lookup finds Key2 in znode zb. Lets say
728 * there is no match and the name is greater so
729 * we look left. When we find Key0, we end up
730 * here. If we return now, we will insert into
731 * znode za at slot n = 1. But that is invalid
732 * according to the parent's keys. Key2 must
733 * be inserted into znode zb.
735 * Note, this problem is not relevant for the
736 * case when we go right, because
737 * 'tnc_insert()' would correct the parent key.
739 if (*n == (*zn)->child_cnt - 1) {
740 err = tnc_next(c, zn, n);
742 /* Should be impossible */
748 ubifs_assert(*n == 0);
753 err = matches_name(c, &(*zn)->zbranch[*n], nm);
756 if (err == NAME_LESS)
758 if (err == NAME_MATCHES)
760 ubifs_assert(err == NAME_GREATER);
764 struct ubifs_znode *znode = *zn;
768 err = tnc_next(c, &znode, &nn);
773 if (keys_cmp(c, &znode->zbranch[nn].key, key))
775 err = matches_name(c, &znode->zbranch[nn], nm);
778 if (err == NAME_GREATER)
782 if (err == NAME_MATCHES)
784 ubifs_assert(err == NAME_LESS);
790 * fallible_matches_name - determine if a dent matches a given name.
791 * @c: UBIFS file-system description object
792 * @zbr: zbranch of dent
795 * This is a "fallible" version of 'matches_name()' function which does not
796 * panic if the direntry/xentry referred by @zbr does not exist on the media.
798 * This function checks if xentry/direntry referred by zbranch @zbr matches name
799 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
800 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
801 * if xentry/direntry referred by @zbr does not exist on the media. A negative
802 * error code is returned in case of failure.
804 static int fallible_matches_name(struct ubifs_info *c,
805 struct ubifs_zbranch *zbr,
806 const struct qstr *nm)
808 struct ubifs_dent_node *dent;
811 /* If possible, match against the dent in the leaf node cache */
813 dent = kmalloc(zbr->len, GFP_NOFS);
817 err = fallible_read_node(c, &zbr->key, zbr, dent);
821 /* The node was not present */
825 ubifs_assert(err == 1);
827 err = lnc_add_directly(c, zbr, dent);
833 nlen = le16_to_cpu(dent->nlen);
834 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
838 else if (nlen < nm->len)
853 * fallible_resolve_collision - resolve a collision even if nodes are missing.
854 * @c: UBIFS file-system description object
856 * @zn: znode is returned here
857 * @n: branch number is passed and returned here
858 * @nm: name of directory entry
859 * @adding: indicates caller is adding a key to the TNC
861 * This is a "fallible" version of the 'resolve_collision()' function which
862 * does not panic if one of the nodes referred to by TNC does not exist on the
863 * media. This may happen when replaying the journal if a deleted node was
864 * Garbage-collected and the commit was not done. A branch that refers to a node
865 * that is not present is called a dangling branch. The following are the return
866 * codes for this function:
867 * o if @nm was found, %1 is returned and @zn and @n are set to the found
869 * o if we are @adding and @nm was not found, %0 is returned;
870 * o if we are not @adding and @nm was not found, but a dangling branch was
871 * found, then %1 is returned and @zn and @n are set to the dangling branch;
872 * o a negative error code is returned in case of failure.
874 static int fallible_resolve_collision(struct ubifs_info *c,
875 const union ubifs_key *key,
876 struct ubifs_znode **zn, int *n,
877 const struct qstr *nm, int adding)
879 struct ubifs_znode *o_znode = NULL, *znode = *zn;
880 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
882 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
883 if (unlikely(cmp < 0))
885 if (cmp == NAME_MATCHES)
887 if (cmp == NOT_ON_MEDIA) {
891 * We are unlucky and hit a dangling branch straight away.
892 * Now we do not really know where to go to find the needed
893 * branch - to the left or to the right. Well, let's try left.
897 unsure = 1; /* Remove a dangling branch wherever it is */
899 if (cmp == NAME_GREATER || unsure) {
902 err = tnc_prev(c, zn, n);
903 if (err == -ENOENT) {
904 ubifs_assert(*n == 0);
910 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
911 /* See comments in 'resolve_collision()' */
912 if (*n == (*zn)->child_cnt - 1) {
913 err = tnc_next(c, zn, n);
915 /* Should be impossible */
921 ubifs_assert(*n == 0);
926 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
929 if (err == NAME_MATCHES)
931 if (err == NOT_ON_MEDIA) {
938 if (err == NAME_LESS)
945 if (cmp == NAME_LESS || unsure) {
950 err = tnc_next(c, &znode, &nn);
955 if (keys_cmp(c, &znode->zbranch[nn].key, key))
957 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
960 if (err == NAME_GREATER)
964 if (err == NAME_MATCHES)
966 if (err == NOT_ON_MEDIA) {
973 /* Never match a dangling branch when adding */
974 if (adding || !o_znode)
977 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
978 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
979 o_znode->zbranch[o_n].len);
986 * matches_position - determine if a zbranch matches a given position.
987 * @zbr: zbranch of dent
988 * @lnum: LEB number of dent to match
989 * @offs: offset of dent to match
991 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
993 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
995 if (zbr->lnum == lnum && zbr->offs == offs)
1002 * resolve_collision_directly - resolve a collision directly.
1003 * @c: UBIFS file-system description object
1004 * @key: key of directory entry
1005 * @zn: znode is passed and returned here
1006 * @n: zbranch number is passed and returned here
1007 * @lnum: LEB number of dent node to match
1008 * @offs: offset of dent node to match
1010 * This function is used for "hashed" keys to make sure the found directory or
1011 * extended attribute entry node is what was looked for. It is used when the
1012 * flash address of the right node is known (@lnum:@offs) which makes it much
1013 * easier to resolve collisions (no need to read entries and match full
1014 * names). This function returns %1 and sets @zn and @n if the collision is
1015 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1016 * previous directory entry. Otherwise a negative error code is returned.
1018 static int resolve_collision_directly(struct ubifs_info *c,
1019 const union ubifs_key *key,
1020 struct ubifs_znode **zn, int *n,
1023 struct ubifs_znode *znode;
1028 if (matches_position(&znode->zbranch[nn], lnum, offs))
1033 err = tnc_prev(c, &znode, &nn);
1038 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1040 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1051 err = tnc_next(c, &znode, &nn);
1056 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1060 if (matches_position(&znode->zbranch[nn], lnum, offs))
1066 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1067 * @c: UBIFS file-system description object
1068 * @znode: znode to dirty
1070 * If we do not have a unique key that resides in a znode, then we cannot
1071 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1072 * This function records the path back to the last dirty ancestor, and then
1073 * dirties the znodes on that path.
1075 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1076 struct ubifs_znode *znode)
1078 struct ubifs_znode *zp;
1079 int *path = c->bottom_up_buf, p = 0;
1081 ubifs_assert(c->zroot.znode);
1082 ubifs_assert(znode);
1083 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1084 kfree(c->bottom_up_buf);
1085 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1087 if (!c->bottom_up_buf)
1088 return ERR_PTR(-ENOMEM);
1089 path = c->bottom_up_buf;
1091 if (c->zroot.znode->level) {
1092 /* Go up until parent is dirty */
1100 ubifs_assert(p < c->zroot.znode->level);
1102 if (!zp->cnext && ubifs_zn_dirty(znode))
1108 /* Come back down, dirtying as we go */
1110 struct ubifs_zbranch *zbr;
1114 ubifs_assert(path[p - 1] >= 0);
1115 ubifs_assert(path[p - 1] < zp->child_cnt);
1116 zbr = &zp->zbranch[path[--p]];
1117 znode = dirty_cow_znode(c, zbr);
1119 ubifs_assert(znode == c->zroot.znode);
1120 znode = dirty_cow_znode(c, &c->zroot);
1122 if (IS_ERR(znode) || !p)
1124 ubifs_assert(path[p - 1] >= 0);
1125 ubifs_assert(path[p - 1] < znode->child_cnt);
1126 znode = znode->zbranch[path[p - 1]].znode;
1133 * ubifs_lookup_level0 - search for zero-level znode.
1134 * @c: UBIFS file-system description object
1135 * @key: key to lookup
1136 * @zn: znode is returned here
1137 * @n: znode branch slot number is returned here
1139 * This function looks up the TNC tree and search for zero-level znode which
1140 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1142 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1143 * is returned and slot number of the matched branch is stored in @n;
1144 * o not exact match, which means that zero-level znode does not contain
1145 * @key, then %0 is returned and slot number of the closest branch is stored
1147 * o @key is so small that it is even less than the lowest key of the
1148 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1150 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1151 * function reads corresponding indexing nodes and inserts them to TNC. In
1152 * case of failure, a negative error code is returned.
1154 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1155 struct ubifs_znode **zn, int *n)
1158 struct ubifs_znode *znode;
1159 unsigned long time = get_seconds();
1161 dbg_tnck(key, "search key ");
1162 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1164 znode = c->zroot.znode;
1165 if (unlikely(!znode)) {
1166 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1168 return PTR_ERR(znode);
1174 struct ubifs_zbranch *zbr;
1176 exact = ubifs_search_zbranch(c, znode, key, n);
1178 if (znode->level == 0)
1183 zbr = &znode->zbranch[*n];
1191 /* znode is not in TNC cache, load it from the media */
1192 znode = ubifs_load_znode(c, zbr, znode, *n);
1194 return PTR_ERR(znode);
1198 if (exact || !is_hash_key(c, key) || *n != -1) {
1199 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1204 * Here is a tricky place. We have not found the key and this is a
1205 * "hashed" key, which may collide. The rest of the code deals with
1206 * situations like this:
1210 * | 3 | 5 | | 6 | 7 | (x)
1212 * Or more a complex example:
1216 * | 1 | 3 | | 5 | 8 |
1218 * | 5 | 5 | | 6 | 7 | (x)
1220 * In the examples, if we are looking for key "5", we may reach nodes
1221 * marked with "(x)". In this case what we have do is to look at the
1222 * left and see if there is "5" key there. If there is, we have to
1225 * Note, this whole situation is possible because we allow to have
1226 * elements which are equivalent to the next key in the parent in the
1227 * children of current znode. For example, this happens if we split a
1228 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1232 * | 3 | 5 | | 5 | 6 | 7 |
1234 * And this becomes what is at the first "picture" after key "5" marked
1235 * with "^" is removed. What could be done is we could prohibit
1236 * splitting in the middle of the colliding sequence. Also, when
1237 * removing the leftmost key, we would have to correct the key of the
1238 * parent node, which would introduce additional complications. Namely,
1239 * if we changed the leftmost key of the parent znode, the garbage
1240 * collector would be unable to find it (GC is doing this when GC'ing
1241 * indexing LEBs). Although we already have an additional RB-tree where
1242 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1243 * after the commit. But anyway, this does not look easy to implement
1244 * so we did not try this.
1246 err = tnc_prev(c, &znode, n);
1247 if (err == -ENOENT) {
1248 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1252 if (unlikely(err < 0))
1254 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1255 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1260 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1266 * lookup_level0_dirty - search for zero-level znode dirtying.
1267 * @c: UBIFS file-system description object
1268 * @key: key to lookup
1269 * @zn: znode is returned here
1270 * @n: znode branch slot number is returned here
1272 * This function looks up the TNC tree and search for zero-level znode which
1273 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1275 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1276 * is returned and slot number of the matched branch is stored in @n;
1277 * o not exact match, which means that zero-level znode does not contain @key
1278 * then %0 is returned and slot number of the closed branch is stored in
1280 * o @key is so small that it is even less than the lowest key of the
1281 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1283 * Additionally all znodes in the path from the root to the located zero-level
1284 * znode are marked as dirty.
1286 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1287 * function reads corresponding indexing nodes and inserts them to TNC. In
1288 * case of failure, a negative error code is returned.
1290 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1291 struct ubifs_znode **zn, int *n)
1294 struct ubifs_znode *znode;
1295 unsigned long time = get_seconds();
1297 dbg_tnck(key, "search and dirty key ");
1299 znode = c->zroot.znode;
1300 if (unlikely(!znode)) {
1301 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1303 return PTR_ERR(znode);
1306 znode = dirty_cow_znode(c, &c->zroot);
1308 return PTR_ERR(znode);
1313 struct ubifs_zbranch *zbr;
1315 exact = ubifs_search_zbranch(c, znode, key, n);
1317 if (znode->level == 0)
1322 zbr = &znode->zbranch[*n];
1326 znode = dirty_cow_znode(c, zbr);
1328 return PTR_ERR(znode);
1332 /* znode is not in TNC cache, load it from the media */
1333 znode = ubifs_load_znode(c, zbr, znode, *n);
1335 return PTR_ERR(znode);
1336 znode = dirty_cow_znode(c, zbr);
1338 return PTR_ERR(znode);
1342 if (exact || !is_hash_key(c, key) || *n != -1) {
1343 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1348 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1351 err = tnc_prev(c, &znode, n);
1352 if (err == -ENOENT) {
1354 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1357 if (unlikely(err < 0))
1359 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1361 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1365 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1366 znode = dirty_cow_bottom_up(c, znode);
1368 return PTR_ERR(znode);
1371 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1377 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1378 * @c: UBIFS file-system description object
1380 * @gc_seq1: garbage collection sequence number
1382 * This function determines if @lnum may have been garbage collected since
1383 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1386 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1389 int gc_seq2, gced_lnum;
1391 gced_lnum = c->gced_lnum;
1393 gc_seq2 = c->gc_seq;
1394 /* Same seq means no GC */
1395 if (gc_seq1 == gc_seq2)
1397 /* Different by more than 1 means we don't know */
1398 if (gc_seq1 + 1 != gc_seq2)
1401 * We have seen the sequence number has increased by 1. Now we need to
1402 * be sure we read the right LEB number, so read it again.
1405 if (gced_lnum != c->gced_lnum)
1407 /* Finally we can check lnum */
1408 if (gced_lnum == lnum)
1411 /* No garbage collection in the read-only U-Boot implementation */
1417 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1418 * @c: UBIFS file-system description object
1419 * @key: node key to lookup
1420 * @node: the node is returned here
1421 * @lnum: LEB number is returned here
1422 * @offs: offset is returned here
1424 * This function looks up and reads node with key @key. The caller has to make
1425 * sure the @node buffer is large enough to fit the node. Returns zero in case
1426 * of success, %-ENOENT if the node was not found, and a negative error code in
1427 * case of failure. The node location can be returned in @lnum and @offs.
1429 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1430 void *node, int *lnum, int *offs)
1432 int found, n, err, safely = 0, gc_seq1;
1433 struct ubifs_znode *znode;
1434 struct ubifs_zbranch zbr, *zt;
1437 mutex_lock(&c->tnc_mutex);
1438 found = ubifs_lookup_level0(c, key, &znode, &n);
1442 } else if (found < 0) {
1446 zt = &znode->zbranch[n];
1451 if (is_hash_key(c, key)) {
1453 * In this case the leaf node cache gets used, so we pass the
1454 * address of the zbranch and keep the mutex locked
1456 err = tnc_read_node_nm(c, zt, node);
1460 err = ubifs_tnc_read_node(c, zt, node);
1463 /* Drop the TNC mutex prematurely and race with garbage collection */
1464 zbr = znode->zbranch[n];
1465 gc_seq1 = c->gc_seq;
1466 mutex_unlock(&c->tnc_mutex);
1468 if (ubifs_get_wbuf(c, zbr.lnum)) {
1469 /* We do not GC journal heads */
1470 err = ubifs_tnc_read_node(c, &zbr, node);
1474 err = fallible_read_node(c, key, &zbr, node);
1475 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1477 * The node may have been GC'ed out from under us so try again
1478 * while keeping the TNC mutex locked.
1486 mutex_unlock(&c->tnc_mutex);
1491 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1492 * @c: UBIFS file-system description object
1493 * @bu: bulk-read parameters and results
1495 * Lookup consecutive data node keys for the same inode that reside
1496 * consecutively in the same LEB. This function returns zero in case of success
1497 * and a negative error code in case of failure.
1499 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1500 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1501 * maximum possible amount of nodes for bulk-read.
1503 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1505 int n, err = 0, lnum = -1, uninitialized_var(offs);
1506 int uninitialized_var(len);
1507 unsigned int block = key_block(c, &bu->key);
1508 struct ubifs_znode *znode;
1514 mutex_lock(&c->tnc_mutex);
1515 /* Find first key */
1516 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1521 len = znode->zbranch[n].len;
1522 /* The buffer must be big enough for at least 1 node */
1523 if (len > bu->buf_len) {
1528 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1530 lnum = znode->zbranch[n].lnum;
1531 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1534 struct ubifs_zbranch *zbr;
1535 union ubifs_key *key;
1536 unsigned int next_block;
1539 err = tnc_next(c, &znode, &n);
1542 zbr = &znode->zbranch[n];
1544 /* See if there is another data key for this file */
1545 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1546 key_type(c, key) != UBIFS_DATA_KEY) {
1551 /* First key found */
1553 offs = ALIGN(zbr->offs + zbr->len, 8);
1555 if (len > bu->buf_len) {
1561 * The data nodes must be in consecutive positions in
1564 if (zbr->lnum != lnum || zbr->offs != offs)
1566 offs += ALIGN(zbr->len, 8);
1567 len = ALIGN(len, 8) + zbr->len;
1568 /* Must not exceed buffer length */
1569 if (len > bu->buf_len)
1572 /* Allow for holes */
1573 next_block = key_block(c, key);
1574 bu->blk_cnt += (next_block - block - 1);
1575 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1579 bu->zbranch[bu->cnt++] = *zbr;
1581 /* See if we have room for more */
1582 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1584 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1588 if (err == -ENOENT) {
1592 bu->gc_seq = c->gc_seq;
1593 mutex_unlock(&c->tnc_mutex);
1597 * An enormous hole could cause bulk-read to encompass too many
1598 * page cache pages, so limit the number here.
1600 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1601 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1603 * Ensure that bulk-read covers a whole number of page cache
1606 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1607 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1610 /* At the end of file we can round up */
1611 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1614 /* Exclude data nodes that do not make up a whole page cache page */
1615 block = key_block(c, &bu->key) + bu->blk_cnt;
1616 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1618 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1626 * read_wbuf - bulk-read from a LEB with a wbuf.
1627 * @wbuf: wbuf that may overlap the read
1628 * @buf: buffer into which to read
1630 * @lnum: LEB number from which to read
1631 * @offs: offset from which to read
1633 * This functions returns %0 on success or a negative error code on failure.
1635 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1638 const struct ubifs_info *c = wbuf->c;
1641 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1642 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1643 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1644 ubifs_assert(offs + len <= c->leb_size);
1646 spin_lock(&wbuf->lock);
1647 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1649 /* We may safely unlock the write-buffer and read the data */
1650 spin_unlock(&wbuf->lock);
1651 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1654 /* Don't read under wbuf */
1655 rlen = wbuf->offs - offs;
1659 /* Copy the rest from the write-buffer */
1660 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1661 spin_unlock(&wbuf->lock);
1664 /* Read everything that goes before write-buffer */
1665 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1671 * validate_data_node - validate data nodes for bulk-read.
1672 * @c: UBIFS file-system description object
1673 * @buf: buffer containing data node to validate
1674 * @zbr: zbranch of data node to validate
1676 * This functions returns %0 on success or a negative error code on failure.
1678 static int validate_data_node(struct ubifs_info *c, void *buf,
1679 struct ubifs_zbranch *zbr)
1681 union ubifs_key key1;
1682 struct ubifs_ch *ch = buf;
1685 if (ch->node_type != UBIFS_DATA_NODE) {
1686 ubifs_err("bad node type (%d but expected %d)",
1687 ch->node_type, UBIFS_DATA_NODE);
1691 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1693 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1697 len = le32_to_cpu(ch->len);
1698 if (len != zbr->len) {
1699 ubifs_err("bad node length %d, expected %d", len, zbr->len);
1703 /* Make sure the key of the read node is correct */
1704 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1705 if (!keys_eq(c, &zbr->key, &key1)) {
1706 ubifs_err("bad key in node at LEB %d:%d",
1707 zbr->lnum, zbr->offs);
1708 dbg_tnck(&zbr->key, "looked for key ");
1709 dbg_tnck(&key1, "found node's key ");
1718 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1719 ubifs_dump_node(c, buf);
1725 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1726 * @c: UBIFS file-system description object
1727 * @bu: bulk-read parameters and results
1729 * This functions reads and validates the data nodes that were identified by the
1730 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1731 * -EAGAIN to indicate a race with GC, or another negative error code on
1734 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1736 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1737 struct ubifs_wbuf *wbuf;
1740 len = bu->zbranch[bu->cnt - 1].offs;
1741 len += bu->zbranch[bu->cnt - 1].len - offs;
1742 if (len > bu->buf_len) {
1743 ubifs_err("buffer too small %d vs %d", bu->buf_len, len);
1748 wbuf = ubifs_get_wbuf(c, lnum);
1750 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1752 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1754 /* Check for a race with GC */
1755 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1758 if (err && err != -EBADMSG) {
1759 ubifs_err("failed to read from LEB %d:%d, error %d",
1762 dbg_tnck(&bu->key, "key ");
1766 /* Validate the nodes read */
1768 for (i = 0; i < bu->cnt; i++) {
1769 err = validate_data_node(c, buf, &bu->zbranch[i]);
1772 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1779 * do_lookup_nm- look up a "hashed" node.
1780 * @c: UBIFS file-system description object
1781 * @key: node key to lookup
1782 * @node: the node is returned here
1785 * This function look up and reads a node which contains name hash in the key.
1786 * Since the hash may have collisions, there may be many nodes with the same
1787 * key, so we have to sequentially look to all of them until the needed one is
1788 * found. This function returns zero in case of success, %-ENOENT if the node
1789 * was not found, and a negative error code in case of failure.
1791 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1792 void *node, const struct qstr *nm)
1795 struct ubifs_znode *znode;
1797 dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name);
1798 mutex_lock(&c->tnc_mutex);
1799 found = ubifs_lookup_level0(c, key, &znode, &n);
1803 } else if (found < 0) {
1808 ubifs_assert(n >= 0);
1810 err = resolve_collision(c, key, &znode, &n, nm);
1811 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1812 if (unlikely(err < 0))
1819 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1822 mutex_unlock(&c->tnc_mutex);
1827 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1828 * @c: UBIFS file-system description object
1829 * @key: node key to lookup
1830 * @node: the node is returned here
1833 * This function look up and reads a node which contains name hash in the key.
1834 * Since the hash may have collisions, there may be many nodes with the same
1835 * key, so we have to sequentially look to all of them until the needed one is
1836 * found. This function returns zero in case of success, %-ENOENT if the node
1837 * was not found, and a negative error code in case of failure.
1839 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1840 void *node, const struct qstr *nm)
1843 const struct ubifs_dent_node *dent = node;
1846 * We assume that in most of the cases there are no name collisions and
1847 * 'ubifs_tnc_lookup()' returns us the right direntry.
1849 err = ubifs_tnc_lookup(c, key, node);
1853 len = le16_to_cpu(dent->nlen);
1854 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1858 * Unluckily, there are hash collisions and we have to iterate over
1859 * them look at each direntry with colliding name hash sequentially.
1861 return do_lookup_nm(c, key, node, nm);
1865 * correct_parent_keys - correct parent znodes' keys.
1866 * @c: UBIFS file-system description object
1867 * @znode: znode to correct parent znodes for
1869 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1870 * zbranch changes, keys of parent znodes have to be corrected. This helper
1871 * function is called in such situations and corrects the keys if needed.
1873 static void correct_parent_keys(const struct ubifs_info *c,
1874 struct ubifs_znode *znode)
1876 union ubifs_key *key, *key1;
1878 ubifs_assert(znode->parent);
1879 ubifs_assert(znode->iip == 0);
1881 key = &znode->zbranch[0].key;
1882 key1 = &znode->parent->zbranch[0].key;
1884 while (keys_cmp(c, key, key1) < 0) {
1885 key_copy(c, key, key1);
1886 znode = znode->parent;
1888 if (!znode->parent || znode->iip)
1890 key1 = &znode->parent->zbranch[0].key;
1895 * insert_zbranch - insert a zbranch into a znode.
1896 * @znode: znode into which to insert
1897 * @zbr: zbranch to insert
1898 * @n: slot number to insert to
1900 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1901 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1902 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1903 * slot, zbranches starting from @n have to be moved right.
1905 static void insert_zbranch(struct ubifs_znode *znode,
1906 const struct ubifs_zbranch *zbr, int n)
1910 ubifs_assert(ubifs_zn_dirty(znode));
1913 for (i = znode->child_cnt; i > n; i--) {
1914 znode->zbranch[i] = znode->zbranch[i - 1];
1915 if (znode->zbranch[i].znode)
1916 znode->zbranch[i].znode->iip = i;
1919 zbr->znode->iip = n;
1921 for (i = znode->child_cnt; i > n; i--)
1922 znode->zbranch[i] = znode->zbranch[i - 1];
1924 znode->zbranch[n] = *zbr;
1925 znode->child_cnt += 1;
1928 * After inserting at slot zero, the lower bound of the key range of
1929 * this znode may have changed. If this znode is subsequently split
1930 * then the upper bound of the key range may change, and furthermore
1931 * it could change to be lower than the original lower bound. If that
1932 * happens, then it will no longer be possible to find this znode in the
1933 * TNC using the key from the index node on flash. That is bad because
1934 * if it is not found, we will assume it is obsolete and may overwrite
1935 * it. Then if there is an unclean unmount, we will start using the
1936 * old index which will be broken.
1938 * So we first mark znodes that have insertions at slot zero, and then
1939 * if they are split we add their lnum/offs to the old_idx tree.
1946 * tnc_insert - insert a node into TNC.
1947 * @c: UBIFS file-system description object
1948 * @znode: znode to insert into
1949 * @zbr: branch to insert
1950 * @n: slot number to insert new zbranch to
1952 * This function inserts a new node described by @zbr into znode @znode. If
1953 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1954 * are splat as well if needed. Returns zero in case of success or a negative
1955 * error code in case of failure.
1957 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1958 struct ubifs_zbranch *zbr, int n)
1960 struct ubifs_znode *zn, *zi, *zp;
1961 int i, keep, move, appending = 0;
1962 union ubifs_key *key = &zbr->key, *key1;
1964 ubifs_assert(n >= 0 && n <= c->fanout);
1966 /* Implement naive insert for now */
1969 if (znode->child_cnt < c->fanout) {
1970 ubifs_assert(n != c->fanout);
1971 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
1973 insert_zbranch(znode, zbr, n);
1975 /* Ensure parent's key is correct */
1976 if (n == 0 && zp && znode->iip == 0)
1977 correct_parent_keys(c, znode);
1983 * Unfortunately, @znode does not have more empty slots and we have to
1986 dbg_tnck(key, "splitting level %d, key ", znode->level);
1990 * We can no longer be sure of finding this znode by key, so we
1991 * record it in the old_idx tree.
1993 ins_clr_old_idx_znode(c, znode);
1995 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
1999 zn->level = znode->level;
2001 /* Decide where to split */
2002 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2003 /* Try not to split consecutive data keys */
2004 if (n == c->fanout) {
2005 key1 = &znode->zbranch[n - 1].key;
2006 if (key_inum(c, key1) == key_inum(c, key) &&
2007 key_type(c, key1) == UBIFS_DATA_KEY)
2011 } else if (appending && n != c->fanout) {
2012 /* Try not to split consecutive data keys */
2015 if (n >= (c->fanout + 1) / 2) {
2016 key1 = &znode->zbranch[0].key;
2017 if (key_inum(c, key1) == key_inum(c, key) &&
2018 key_type(c, key1) == UBIFS_DATA_KEY) {
2019 key1 = &znode->zbranch[n].key;
2020 if (key_inum(c, key1) != key_inum(c, key) ||
2021 key_type(c, key1) != UBIFS_DATA_KEY) {
2023 move = c->fanout - keep;
2035 keep = (c->fanout + 1) / 2;
2036 move = c->fanout - keep;
2040 * Although we don't at present, we could look at the neighbors and see
2041 * if we can move some zbranches there.
2045 /* Insert into existing znode */
2050 /* Insert into new znode */
2055 zbr->znode->parent = zn;
2060 __set_bit(DIRTY_ZNODE, &zn->flags);
2061 atomic_long_inc(&c->dirty_zn_cnt);
2063 zn->child_cnt = move;
2064 znode->child_cnt = keep;
2066 dbg_tnc("moving %d, keeping %d", move, keep);
2069 for (i = 0; i < move; i++) {
2070 zn->zbranch[i] = znode->zbranch[keep + i];
2073 if (zn->zbranch[i].znode) {
2074 zn->zbranch[i].znode->parent = zn;
2075 zn->zbranch[i].znode->iip = i;
2079 /* Insert new key and branch */
2080 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2082 insert_zbranch(zi, zbr, n);
2084 /* Insert new znode (produced by spitting) into the parent */
2086 if (n == 0 && zi == znode && znode->iip == 0)
2087 correct_parent_keys(c, znode);
2089 /* Locate insertion point */
2092 /* Tail recursion */
2093 zbr->key = zn->zbranch[0].key;
2103 /* We have to split root znode */
2104 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2106 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2111 zi->level = znode->level + 1;
2113 __set_bit(DIRTY_ZNODE, &zi->flags);
2114 atomic_long_inc(&c->dirty_zn_cnt);
2116 zi->zbranch[0].key = znode->zbranch[0].key;
2117 zi->zbranch[0].znode = znode;
2118 zi->zbranch[0].lnum = c->zroot.lnum;
2119 zi->zbranch[0].offs = c->zroot.offs;
2120 zi->zbranch[0].len = c->zroot.len;
2121 zi->zbranch[1].key = zn->zbranch[0].key;
2122 zi->zbranch[1].znode = zn;
2127 c->zroot.znode = zi;
2138 * ubifs_tnc_add - add a node to TNC.
2139 * @c: UBIFS file-system description object
2141 * @lnum: LEB number of node
2142 * @offs: node offset
2145 * This function adds a node with key @key to TNC. The node may be new or it may
2146 * obsolete some existing one. Returns %0 on success or negative error code on
2149 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2152 int found, n, err = 0;
2153 struct ubifs_znode *znode;
2155 mutex_lock(&c->tnc_mutex);
2156 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2157 found = lookup_level0_dirty(c, key, &znode, &n);
2159 struct ubifs_zbranch zbr;
2165 key_copy(c, key, &zbr.key);
2166 err = tnc_insert(c, znode, &zbr, n + 1);
2167 } else if (found == 1) {
2168 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2171 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2178 err = dbg_check_tnc(c, 0);
2179 mutex_unlock(&c->tnc_mutex);
2185 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2186 * @c: UBIFS file-system description object
2188 * @old_lnum: LEB number of old node
2189 * @old_offs: old node offset
2190 * @lnum: LEB number of node
2191 * @offs: node offset
2194 * This function replaces a node with key @key in the TNC only if the old node
2195 * is found. This function is called by garbage collection when node are moved.
2196 * Returns %0 on success or negative error code on failure.
2198 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2199 int old_lnum, int old_offs, int lnum, int offs, int len)
2201 int found, n, err = 0;
2202 struct ubifs_znode *znode;
2204 mutex_lock(&c->tnc_mutex);
2205 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2206 old_offs, lnum, offs, len);
2207 found = lookup_level0_dirty(c, key, &znode, &n);
2214 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2217 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2219 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2226 } else if (is_hash_key(c, key)) {
2227 found = resolve_collision_directly(c, key, &znode, &n,
2228 old_lnum, old_offs);
2229 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2230 found, znode, n, old_lnum, old_offs);
2237 /* Ensure the znode is dirtied */
2238 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2239 znode = dirty_cow_bottom_up(c, znode);
2240 if (IS_ERR(znode)) {
2241 err = PTR_ERR(znode);
2245 zbr = &znode->zbranch[n];
2247 err = ubifs_add_dirt(c, zbr->lnum,
2259 err = ubifs_add_dirt(c, lnum, len);
2262 err = dbg_check_tnc(c, 0);
2265 mutex_unlock(&c->tnc_mutex);
2270 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2271 * @c: UBIFS file-system description object
2273 * @lnum: LEB number of node
2274 * @offs: node offset
2278 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2279 * may have collisions, like directory entry keys.
2281 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2282 int lnum, int offs, int len, const struct qstr *nm)
2284 int found, n, err = 0;
2285 struct ubifs_znode *znode;
2287 mutex_lock(&c->tnc_mutex);
2288 dbg_tnck(key, "LEB %d:%d, name '%.*s', key ",
2289 lnum, offs, nm->len, nm->name);
2290 found = lookup_level0_dirty(c, key, &znode, &n);
2298 found = fallible_resolve_collision(c, key, &znode, &n,
2301 found = resolve_collision(c, key, &znode, &n, nm);
2302 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2308 /* Ensure the znode is dirtied */
2309 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2310 znode = dirty_cow_bottom_up(c, znode);
2311 if (IS_ERR(znode)) {
2312 err = PTR_ERR(znode);
2318 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2321 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2330 struct ubifs_zbranch zbr;
2336 key_copy(c, key, &zbr.key);
2337 err = tnc_insert(c, znode, &zbr, n + 1);
2342 * We did not find it in the index so there may be a
2343 * dangling branch still in the index. So we remove it
2344 * by passing 'ubifs_tnc_remove_nm()' the same key but
2345 * an unmatchable name.
2347 struct qstr noname = { .name = "" };
2349 err = dbg_check_tnc(c, 0);
2350 mutex_unlock(&c->tnc_mutex);
2353 return ubifs_tnc_remove_nm(c, key, &noname);
2359 err = dbg_check_tnc(c, 0);
2360 mutex_unlock(&c->tnc_mutex);
2365 * tnc_delete - delete a znode form TNC.
2366 * @c: UBIFS file-system description object
2367 * @znode: znode to delete from
2368 * @n: zbranch slot number to delete
2370 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2371 * case of success and a negative error code in case of failure.
2373 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2375 struct ubifs_zbranch *zbr;
2376 struct ubifs_znode *zp;
2379 /* Delete without merge for now */
2380 ubifs_assert(znode->level == 0);
2381 ubifs_assert(n >= 0 && n < c->fanout);
2382 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2384 zbr = &znode->zbranch[n];
2387 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2389 ubifs_dump_znode(c, znode);
2393 /* We do not "gap" zbranch slots */
2394 for (i = n; i < znode->child_cnt - 1; i++)
2395 znode->zbranch[i] = znode->zbranch[i + 1];
2396 znode->child_cnt -= 1;
2398 if (znode->child_cnt > 0)
2402 * This was the last zbranch, we have to delete this znode from the
2407 ubifs_assert(!ubifs_zn_obsolete(znode));
2408 ubifs_assert(ubifs_zn_dirty(znode));
2413 atomic_long_dec(&c->dirty_zn_cnt);
2415 err = insert_old_idx_znode(c, znode);
2420 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2421 atomic_long_inc(&c->clean_zn_cnt);
2422 atomic_long_inc(&ubifs_clean_zn_cnt);
2426 } while (znode->child_cnt == 1); /* while removing last child */
2428 /* Remove from znode, entry n - 1 */
2429 znode->child_cnt -= 1;
2430 ubifs_assert(znode->level != 0);
2431 for (i = n; i < znode->child_cnt; i++) {
2432 znode->zbranch[i] = znode->zbranch[i + 1];
2433 if (znode->zbranch[i].znode)
2434 znode->zbranch[i].znode->iip = i;
2438 * If this is the root and it has only 1 child then
2439 * collapse the tree.
2441 if (!znode->parent) {
2442 while (znode->child_cnt == 1 && znode->level != 0) {
2444 zbr = &znode->zbranch[0];
2445 znode = get_znode(c, znode, 0);
2447 return PTR_ERR(znode);
2448 znode = dirty_cow_znode(c, zbr);
2450 return PTR_ERR(znode);
2451 znode->parent = NULL;
2454 err = insert_old_idx(c, c->zroot.lnum,
2459 c->zroot.lnum = zbr->lnum;
2460 c->zroot.offs = zbr->offs;
2461 c->zroot.len = zbr->len;
2462 c->zroot.znode = znode;
2463 ubifs_assert(!ubifs_zn_obsolete(zp));
2464 ubifs_assert(ubifs_zn_dirty(zp));
2465 atomic_long_dec(&c->dirty_zn_cnt);
2468 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2469 atomic_long_inc(&c->clean_zn_cnt);
2470 atomic_long_inc(&ubifs_clean_zn_cnt);
2480 * ubifs_tnc_remove - remove an index entry of a node.
2481 * @c: UBIFS file-system description object
2484 * Returns %0 on success or negative error code on failure.
2486 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2488 int found, n, err = 0;
2489 struct ubifs_znode *znode;
2491 mutex_lock(&c->tnc_mutex);
2492 dbg_tnck(key, "key ");
2493 found = lookup_level0_dirty(c, key, &znode, &n);
2499 err = tnc_delete(c, znode, n);
2501 err = dbg_check_tnc(c, 0);
2504 mutex_unlock(&c->tnc_mutex);
2509 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2510 * @c: UBIFS file-system description object
2512 * @nm: directory entry name
2514 * Returns %0 on success or negative error code on failure.
2516 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2517 const struct qstr *nm)
2520 struct ubifs_znode *znode;
2522 mutex_lock(&c->tnc_mutex);
2523 dbg_tnck(key, "%.*s, key ", nm->len, nm->name);
2524 err = lookup_level0_dirty(c, key, &znode, &n);
2530 err = fallible_resolve_collision(c, key, &znode, &n,
2533 err = resolve_collision(c, key, &znode, &n, nm);
2534 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2538 /* Ensure the znode is dirtied */
2539 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2540 znode = dirty_cow_bottom_up(c, znode);
2541 if (IS_ERR(znode)) {
2542 err = PTR_ERR(znode);
2546 err = tnc_delete(c, znode, n);
2552 err = dbg_check_tnc(c, 0);
2553 mutex_unlock(&c->tnc_mutex);
2558 * key_in_range - determine if a key falls within a range of keys.
2559 * @c: UBIFS file-system description object
2560 * @key: key to check
2561 * @from_key: lowest key in range
2562 * @to_key: highest key in range
2564 * This function returns %1 if the key is in range and %0 otherwise.
2566 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2567 union ubifs_key *from_key, union ubifs_key *to_key)
2569 if (keys_cmp(c, key, from_key) < 0)
2571 if (keys_cmp(c, key, to_key) > 0)
2577 * ubifs_tnc_remove_range - remove index entries in range.
2578 * @c: UBIFS file-system description object
2579 * @from_key: lowest key to remove
2580 * @to_key: highest key to remove
2582 * This function removes index entries starting at @from_key and ending at
2583 * @to_key. This function returns zero in case of success and a negative error
2584 * code in case of failure.
2586 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2587 union ubifs_key *to_key)
2589 int i, n, k, err = 0;
2590 struct ubifs_znode *znode;
2591 union ubifs_key *key;
2593 mutex_lock(&c->tnc_mutex);
2595 /* Find first level 0 znode that contains keys to remove */
2596 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2603 err = tnc_next(c, &znode, &n);
2604 if (err == -ENOENT) {
2610 key = &znode->zbranch[n].key;
2611 if (!key_in_range(c, key, from_key, to_key)) {
2617 /* Ensure the znode is dirtied */
2618 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2619 znode = dirty_cow_bottom_up(c, znode);
2620 if (IS_ERR(znode)) {
2621 err = PTR_ERR(znode);
2626 /* Remove all keys in range except the first */
2627 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2628 key = &znode->zbranch[i].key;
2629 if (!key_in_range(c, key, from_key, to_key))
2631 lnc_free(&znode->zbranch[i]);
2632 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2633 znode->zbranch[i].len);
2635 ubifs_dump_znode(c, znode);
2638 dbg_tnck(key, "removing key ");
2641 for (i = n + 1 + k; i < znode->child_cnt; i++)
2642 znode->zbranch[i - k] = znode->zbranch[i];
2643 znode->child_cnt -= k;
2646 /* Now delete the first */
2647 err = tnc_delete(c, znode, n);
2654 err = dbg_check_tnc(c, 0);
2655 mutex_unlock(&c->tnc_mutex);
2660 * ubifs_tnc_remove_ino - remove an inode from TNC.
2661 * @c: UBIFS file-system description object
2662 * @inum: inode number to remove
2664 * This function remove inode @inum and all the extended attributes associated
2665 * with the anode from TNC and returns zero in case of success or a negative
2666 * error code in case of failure.
2668 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2670 union ubifs_key key1, key2;
2671 struct ubifs_dent_node *xent, *pxent = NULL;
2672 struct qstr nm = { .name = NULL };
2674 dbg_tnc("ino %lu", (unsigned long)inum);
2677 * Walk all extended attribute entries and remove them together with
2678 * corresponding extended attribute inodes.
2680 lowest_xent_key(c, &key1, inum);
2685 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2687 err = PTR_ERR(xent);
2693 xattr_inum = le64_to_cpu(xent->inum);
2694 dbg_tnc("xent '%s', ino %lu", xent->name,
2695 (unsigned long)xattr_inum);
2697 nm.name = xent->name;
2698 nm.len = le16_to_cpu(xent->nlen);
2699 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2705 lowest_ino_key(c, &key1, xattr_inum);
2706 highest_ino_key(c, &key2, xattr_inum);
2707 err = ubifs_tnc_remove_range(c, &key1, &key2);
2715 key_read(c, &xent->key, &key1);
2719 lowest_ino_key(c, &key1, inum);
2720 highest_ino_key(c, &key2, inum);
2722 return ubifs_tnc_remove_range(c, &key1, &key2);
2726 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2727 * @c: UBIFS file-system description object
2728 * @key: key of last entry
2729 * @nm: name of last entry found or %NULL
2731 * This function finds and reads the next directory or extended attribute entry
2732 * after the given key (@key) if there is one. @nm is used to resolve
2735 * If the name of the current entry is not known and only the key is known,
2736 * @nm->name has to be %NULL. In this case the semantics of this function is a
2737 * little bit different and it returns the entry corresponding to this key, not
2738 * the next one. If the key was not found, the closest "right" entry is
2741 * If the fist entry has to be found, @key has to contain the lowest possible
2742 * key value for this inode and @name has to be %NULL.
2744 * This function returns the found directory or extended attribute entry node
2745 * in case of success, %-ENOENT is returned if no entry was found, and a
2746 * negative error code is returned in case of failure.
2748 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2749 union ubifs_key *key,
2750 const struct qstr *nm)
2752 int n, err, type = key_type(c, key);
2753 struct ubifs_znode *znode;
2754 struct ubifs_dent_node *dent;
2755 struct ubifs_zbranch *zbr;
2756 union ubifs_key *dkey;
2758 dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)");
2759 ubifs_assert(is_hash_key(c, key));
2761 mutex_lock(&c->tnc_mutex);
2762 err = ubifs_lookup_level0(c, key, &znode, &n);
2763 if (unlikely(err < 0))
2768 /* Handle collisions */
2769 err = resolve_collision(c, key, &znode, &n, nm);
2770 dbg_tnc("rc returned %d, znode %p, n %d",
2772 if (unlikely(err < 0))
2776 /* Now find next entry */
2777 err = tnc_next(c, &znode, &n);
2782 * The full name of the entry was not given, in which case the
2783 * behavior of this function is a little different and it
2784 * returns current entry, not the next one.
2788 * However, the given key does not exist in the TNC
2789 * tree and @znode/@n variables contain the closest
2790 * "preceding" element. Switch to the next one.
2792 err = tnc_next(c, &znode, &n);
2798 zbr = &znode->zbranch[n];
2799 dent = kmalloc(zbr->len, GFP_NOFS);
2800 if (unlikely(!dent)) {
2806 * The above 'tnc_next()' call could lead us to the next inode, check
2810 if (key_inum(c, dkey) != key_inum(c, key) ||
2811 key_type(c, dkey) != type) {
2816 err = tnc_read_node_nm(c, zbr, dent);
2820 mutex_unlock(&c->tnc_mutex);
2826 mutex_unlock(&c->tnc_mutex);
2827 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);
2875 * left_znode - get the znode to the left.
2876 * @c: UBIFS file-system description object
2879 * This function returns a pointer to the znode to the left of @znode or NULL if
2880 * there is not one. A negative error code is returned on failure.
2882 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2883 struct ubifs_znode *znode)
2885 int level = znode->level;
2888 int n = znode->iip - 1;
2890 /* Go up until we can go left */
2891 znode = znode->parent;
2895 /* Now go down the rightmost branch to 'level' */
2896 znode = get_znode(c, znode, n);
2899 while (znode->level != level) {
2900 n = znode->child_cnt - 1;
2901 znode = get_znode(c, znode, n);
2912 * right_znode - get the znode to the right.
2913 * @c: UBIFS file-system description object
2916 * This function returns a pointer to the znode to the right of @znode or NULL
2917 * if there is not one. A negative error code is returned on failure.
2919 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2920 struct ubifs_znode *znode)
2922 int level = znode->level;
2925 int n = znode->iip + 1;
2927 /* Go up until we can go right */
2928 znode = znode->parent;
2931 if (n < znode->child_cnt) {
2932 /* Now go down the leftmost branch to 'level' */
2933 znode = get_znode(c, znode, n);
2936 while (znode->level != level) {
2937 znode = get_znode(c, znode, 0);
2948 * lookup_znode - find a particular indexing node from TNC.
2949 * @c: UBIFS file-system description object
2950 * @key: index node key to lookup
2951 * @level: index node level
2952 * @lnum: index node LEB number
2953 * @offs: index node offset
2955 * This function searches an indexing node by its first key @key and its
2956 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2957 * nodes it traverses to TNC. This function is called for indexing nodes which
2958 * were found on the media by scanning, for example when garbage-collecting or
2959 * when doing in-the-gaps commit. This means that the indexing node which is
2960 * looked for does not have to have exactly the same leftmost key @key, because
2961 * the leftmost key may have been changed, in which case TNC will contain a
2962 * dirty znode which still refers the same @lnum:@offs. This function is clever
2963 * enough to recognize such indexing nodes.
2965 * Note, if a znode was deleted or changed too much, then this function will
2966 * not find it. For situations like this UBIFS has the old index RB-tree
2967 * (indexed by @lnum:@offs).
2969 * This function returns a pointer to the znode found or %NULL if it is not
2970 * found. A negative error code is returned on failure.
2972 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2973 union ubifs_key *key, int level,
2976 struct ubifs_znode *znode, *zn;
2979 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
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);
3264 * dbg_check_inode_size - check if inode size is correct.
3265 * @c: UBIFS file-system description object
3266 * @inum: inode number
3269 * This function makes sure that the inode size (@size) is correct and it does
3270 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3271 * if it has a data page beyond @size, and other negative error code in case of
3274 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3278 union ubifs_key from_key, to_key, *key;
3279 struct ubifs_znode *znode;
3282 if (!S_ISREG(inode->i_mode))
3284 if (!dbg_is_chk_gen(c))
3287 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3288 data_key_init(c, &from_key, inode->i_ino, block);
3289 highest_data_key(c, &to_key, inode->i_ino);
3291 mutex_lock(&c->tnc_mutex);
3292 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3302 err = tnc_next(c, &znode, &n);
3303 if (err == -ENOENT) {
3310 ubifs_assert(err == 0);
3311 key = &znode->zbranch[n].key;
3312 if (!key_in_range(c, key, &from_key, &to_key))
3316 block = key_block(c, key);
3317 ubifs_err("inode %lu has size %lld, but there are data at offset %lld",
3318 (unsigned long)inode->i_ino, size,
3319 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3320 mutex_unlock(&c->tnc_mutex);
3321 ubifs_dump_inode(c, inode);
3326 mutex_unlock(&c->tnc_mutex);