2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
27 * At the moment the locking rules of the TNC tree are quite simple and
28 * straightforward. We just have a mutex and lock it when we traverse the
29 * tree. If a znode is not in memory, we read it from flash while still having
33 #include <linux/crc32.h>
37 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
38 * @NAME_LESS: name corresponding to the first argument is less than second
39 * @NAME_MATCHES: names match
40 * @NAME_GREATER: name corresponding to the second argument is greater than
42 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
44 * These constants were introduce to improve readability.
54 * insert_old_idx - record an index node obsoleted since the last commit start.
55 * @c: UBIFS file-system description object
56 * @lnum: LEB number of obsoleted index node
57 * @offs: offset of obsoleted index node
59 * Returns %0 on success, and a negative error code on failure.
61 * For recovery, there must always be a complete intact version of the index on
62 * flash at all times. That is called the "old index". It is the index as at the
63 * time of the last successful commit. Many of the index nodes in the old index
64 * may be dirty, but they must not be erased until the next successful commit
65 * (at which point that index becomes the old index).
67 * That means that the garbage collection and the in-the-gaps method of
68 * committing must be able to determine if an index node is in the old index.
69 * Most of the old index nodes can be found by looking up the TNC using the
70 * 'lookup_znode()' function. However, some of the old index nodes may have
71 * been deleted from the current index or may have been changed so much that
72 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
73 * That is what this function does. The RB-tree is ordered by LEB number and
74 * offset because they uniquely identify the old index node.
76 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
78 struct ubifs_old_idx *old_idx, *o;
79 struct rb_node **p, *parent = NULL;
81 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
82 if (unlikely(!old_idx))
87 p = &c->old_idx.rb_node;
90 o = rb_entry(parent, struct ubifs_old_idx, rb);
93 else if (lnum > o->lnum)
95 else if (offs < o->offs)
97 else if (offs > o->offs)
100 ubifs_err("old idx added twice!");
105 rb_link_node(&old_idx->rb, parent, p);
106 rb_insert_color(&old_idx->rb, &c->old_idx);
111 * insert_old_idx_znode - record a znode obsoleted since last commit start.
112 * @c: UBIFS file-system description object
113 * @znode: znode of obsoleted index node
115 * Returns %0 on success, and a negative error code on failure.
117 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
120 struct ubifs_zbranch *zbr;
122 zbr = &znode->parent->zbranch[znode->iip];
124 return insert_old_idx(c, zbr->lnum, zbr->offs);
127 return insert_old_idx(c, c->zroot.lnum,
133 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
134 * @c: UBIFS file-system description object
135 * @znode: znode of obsoleted index node
137 * Returns %0 on success, and a negative error code on failure.
139 static int ins_clr_old_idx_znode(struct ubifs_info *c,
140 struct ubifs_znode *znode)
145 struct ubifs_zbranch *zbr;
147 zbr = &znode->parent->zbranch[znode->iip];
149 err = insert_old_idx(c, zbr->lnum, zbr->offs);
158 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
169 * destroy_old_idx - destroy the old_idx RB-tree.
170 * @c: UBIFS file-system description object
172 * During start commit, the old_idx RB-tree is used to avoid overwriting index
173 * nodes that were in the index last commit but have since been deleted. This
174 * is necessary for recovery i.e. the old index must be kept intact until the
175 * new index is successfully written. The old-idx RB-tree is used for the
176 * in-the-gaps method of writing index nodes and is destroyed every commit.
178 void destroy_old_idx(struct ubifs_info *c)
180 struct rb_node *this = c->old_idx.rb_node;
181 struct ubifs_old_idx *old_idx;
185 this = this->rb_left;
187 } else if (this->rb_right) {
188 this = this->rb_right;
191 old_idx = rb_entry(this, struct ubifs_old_idx, rb);
192 this = rb_parent(this);
194 if (this->rb_left == &old_idx->rb)
195 this->rb_left = NULL;
197 this->rb_right = NULL;
201 c->old_idx = RB_ROOT;
205 * copy_znode - copy a dirty znode.
206 * @c: UBIFS file-system description object
207 * @znode: znode to copy
209 * A dirty znode being committed may not be changed, so it is copied.
211 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
212 struct ubifs_znode *znode)
214 struct ubifs_znode *zn;
216 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
218 return ERR_PTR(-ENOMEM);
220 memcpy(zn, znode, c->max_znode_sz);
222 __set_bit(DIRTY_ZNODE, &zn->flags);
223 __clear_bit(COW_ZNODE, &zn->flags);
225 ubifs_assert(!test_bit(OBSOLETE_ZNODE, &znode->flags));
226 __set_bit(OBSOLETE_ZNODE, &znode->flags);
228 if (znode->level != 0) {
230 const int n = zn->child_cnt;
232 /* The children now have new parent */
233 for (i = 0; i < n; i++) {
234 struct ubifs_zbranch *zbr = &zn->zbranch[i];
237 zbr->znode->parent = zn;
241 atomic_long_inc(&c->dirty_zn_cnt);
246 * add_idx_dirt - add dirt due to a dirty znode.
247 * @c: UBIFS file-system description object
248 * @lnum: LEB number of index node
249 * @dirt: size of index node
251 * This function updates lprops dirty space and the new size of the index.
253 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
255 c->calc_idx_sz -= ALIGN(dirt, 8);
256 return ubifs_add_dirt(c, lnum, dirt);
260 * dirty_cow_znode - ensure a znode is not being committed.
261 * @c: UBIFS file-system description object
262 * @zbr: branch of znode to check
264 * Returns dirtied znode on success or negative error code on failure.
266 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
267 struct ubifs_zbranch *zbr)
269 struct ubifs_znode *znode = zbr->znode;
270 struct ubifs_znode *zn;
273 if (!test_bit(COW_ZNODE, &znode->flags)) {
274 /* znode is not being committed */
275 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
276 atomic_long_inc(&c->dirty_zn_cnt);
277 atomic_long_dec(&c->clean_zn_cnt);
278 atomic_long_dec(&ubifs_clean_zn_cnt);
279 err = add_idx_dirt(c, zbr->lnum, zbr->len);
286 zn = copy_znode(c, znode);
291 err = insert_old_idx(c, zbr->lnum, zbr->offs);
294 err = add_idx_dirt(c, zbr->lnum, zbr->len);
309 * lnc_add - add a leaf node to the leaf node cache.
310 * @c: UBIFS file-system description object
311 * @zbr: zbranch of leaf node
314 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
315 * purpose of the leaf node cache is to save re-reading the same leaf node over
316 * and over again. Most things are cached by VFS, however the file system must
317 * cache directory entries for readdir and for resolving hash collisions. The
318 * present implementation of the leaf node cache is extremely simple, and
319 * allows for error returns that are not used but that may be needed if a more
320 * complex implementation is created.
322 * Note, this function does not add the @node object to LNC directly, but
323 * allocates a copy of the object and adds the copy to LNC. The reason for this
324 * is that @node has been allocated outside of the TNC subsystem and will be
325 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
326 * may be changed at any time, e.g. freed by the shrinker.
328 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
333 const struct ubifs_dent_node *dent = node;
335 ubifs_assert(!zbr->leaf);
336 ubifs_assert(zbr->len != 0);
337 ubifs_assert(is_hash_key(c, &zbr->key));
339 err = ubifs_validate_entry(c, dent);
342 dbg_dump_node(c, dent);
346 lnc_node = kmalloc(zbr->len, GFP_NOFS);
348 /* We don't have to have the cache, so no error */
351 memcpy(lnc_node, node, zbr->len);
352 zbr->leaf = lnc_node;
357 * lnc_add_directly - add a leaf node to the leaf-node-cache.
358 * @c: UBIFS file-system description object
359 * @zbr: zbranch of leaf node
362 * This function is similar to 'lnc_add()', but it does not create a copy of
363 * @node but inserts @node to TNC directly.
365 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
370 ubifs_assert(!zbr->leaf);
371 ubifs_assert(zbr->len != 0);
373 err = ubifs_validate_entry(c, node);
376 dbg_dump_node(c, node);
385 * lnc_free - remove a leaf node from the leaf node cache.
386 * @zbr: zbranch of leaf node
389 static void lnc_free(struct ubifs_zbranch *zbr)
398 * tnc_read_node_nm - read a "hashed" leaf node.
399 * @c: UBIFS file-system description object
400 * @zbr: key and position of the node
401 * @node: node is returned here
403 * This function reads a "hashed" node defined by @zbr from the leaf node cache
404 * (in it is there) or from the hash media, in which case the node is also
405 * added to LNC. Returns zero in case of success or a negative negative error
406 * code in case of failure.
408 static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
413 ubifs_assert(is_hash_key(c, &zbr->key));
416 /* Read from the leaf node cache */
417 ubifs_assert(zbr->len != 0);
418 memcpy(node, zbr->leaf, zbr->len);
422 err = ubifs_tnc_read_node(c, zbr, node);
426 /* Add the node to the leaf node cache */
427 err = lnc_add(c, zbr, node);
432 * try_read_node - read a node if it is a node.
433 * @c: UBIFS file-system description object
434 * @buf: buffer to read to
436 * @len: node length (not aligned)
437 * @lnum: LEB number of node to read
438 * @offs: offset of node to read
440 * This function tries to read a node of known type and length, checks it and
441 * stores it in @buf. This function returns %1 if a node is present and %0 if
442 * a node is not present. A negative error code is returned for I/O errors.
443 * This function performs that same function as ubifs_read_node except that
444 * it does not require that there is actually a node present and instead
445 * the return code indicates if a node was read.
447 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
448 int len, int lnum, int offs)
451 struct ubifs_ch *ch = buf;
452 uint32_t crc, node_crc;
454 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
456 err = ubi_read(c->ubi, lnum, buf, offs, len);
458 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
459 type, lnum, offs, err);
463 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
466 if (ch->node_type != type)
469 node_len = le32_to_cpu(ch->len);
473 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
474 node_crc = le32_to_cpu(ch->crc);
482 * fallible_read_node - try to read a leaf node.
483 * @c: UBIFS file-system description object
484 * @key: key of node to read
485 * @zbr: position of node
486 * @node: node returned
488 * This function tries to read a node and returns %1 if the node is read, %0
489 * if the node is not present, and a negative error code in the case of error.
491 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
492 struct ubifs_zbranch *zbr, void *node)
496 dbg_tnc("LEB %d:%d, key %s", zbr->lnum, zbr->offs, DBGKEY(key));
498 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
501 union ubifs_key node_key;
502 struct ubifs_dent_node *dent = node;
504 /* All nodes have key in the same place */
505 key_read(c, &dent->key, &node_key);
506 if (keys_cmp(c, key, &node_key) != 0)
509 if (ret == 0 && c->replaying)
510 dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
511 zbr->lnum, zbr->offs, zbr->len, DBGKEY(key));
516 * matches_name - determine if a direntry or xattr entry matches a given name.
517 * @c: UBIFS file-system description object
518 * @zbr: zbranch of dent
521 * This function checks if xentry/direntry referred by zbranch @zbr matches name
522 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
523 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
524 * of failure, a negative error code is returned.
526 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
527 const struct qstr *nm)
529 struct ubifs_dent_node *dent;
532 /* If possible, match against the dent in the leaf node cache */
534 dent = kmalloc(zbr->len, GFP_NOFS);
538 err = ubifs_tnc_read_node(c, zbr, dent);
542 /* Add the node to the leaf node cache */
543 err = lnc_add_directly(c, zbr, dent);
549 nlen = le16_to_cpu(dent->nlen);
550 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
554 else if (nlen < nm->len)
569 * get_znode - get a TNC znode that may not be loaded yet.
570 * @c: UBIFS file-system description object
571 * @znode: parent znode
572 * @n: znode branch slot number
574 * This function returns the znode or a negative error code.
576 static struct ubifs_znode *get_znode(struct ubifs_info *c,
577 struct ubifs_znode *znode, int n)
579 struct ubifs_zbranch *zbr;
581 zbr = &znode->zbranch[n];
585 znode = ubifs_load_znode(c, zbr, znode, n);
590 * tnc_next - find next TNC entry.
591 * @c: UBIFS file-system description object
592 * @zn: znode is passed and returned here
593 * @n: znode branch slot number is passed and returned here
595 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
596 * no next entry, or a negative error code otherwise.
598 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
600 struct ubifs_znode *znode = *zn;
604 if (nn < znode->child_cnt) {
609 struct ubifs_znode *zp;
616 if (nn < znode->child_cnt) {
617 znode = get_znode(c, znode, nn);
619 return PTR_ERR(znode);
620 while (znode->level != 0) {
621 znode = get_znode(c, znode, 0);
623 return PTR_ERR(znode);
635 * tnc_prev - find previous TNC entry.
636 * @c: UBIFS file-system description object
637 * @zn: znode is returned here
638 * @n: znode branch slot number is passed and returned here
640 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
641 * there is no next entry, or a negative error code otherwise.
643 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
645 struct ubifs_znode *znode = *zn;
653 struct ubifs_znode *zp;
661 znode = get_znode(c, znode, nn);
663 return PTR_ERR(znode);
664 while (znode->level != 0) {
665 nn = znode->child_cnt - 1;
666 znode = get_znode(c, znode, nn);
668 return PTR_ERR(znode);
670 nn = znode->child_cnt - 1;
680 * resolve_collision - resolve a collision.
681 * @c: UBIFS file-system description object
682 * @key: key of a directory or extended attribute entry
683 * @zn: znode is returned here
684 * @n: zbranch number is passed and returned here
685 * @nm: name of the entry
687 * This function is called for "hashed" keys to make sure that the found key
688 * really corresponds to the looked up node (directory or extended attribute
689 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
690 * %0 is returned if @nm is not found and @zn and @n are set to the previous
691 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
692 * This means that @n may be set to %-1 if the leftmost key in @zn is the
693 * previous one. A negative error code is returned on failures.
695 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
696 struct ubifs_znode **zn, int *n,
697 const struct qstr *nm)
701 err = matches_name(c, &(*zn)->zbranch[*n], nm);
702 if (unlikely(err < 0))
704 if (err == NAME_MATCHES)
707 if (err == NAME_GREATER) {
710 err = tnc_prev(c, zn, n);
711 if (err == -ENOENT) {
712 ubifs_assert(*n == 0);
718 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
720 * We have found the branch after which we would
721 * like to insert, but inserting in this znode
722 * may still be wrong. Consider the following 3
723 * znodes, in the case where we are resolving a
724 * collision with Key2.
727 * ----------------------
728 * level 1 | Key0 | Key1 |
729 * -----------------------
731 * znode za | | znode zb
732 * ------------ ------------
733 * level 0 | Key0 | | Key2 |
734 * ------------ ------------
736 * The lookup finds Key2 in znode zb. Lets say
737 * there is no match and the name is greater so
738 * we look left. When we find Key0, we end up
739 * here. If we return now, we will insert into
740 * znode za at slot n = 1. But that is invalid
741 * according to the parent's keys. Key2 must
742 * be inserted into znode zb.
744 * Note, this problem is not relevant for the
745 * case when we go right, because
746 * 'tnc_insert()' would correct the parent key.
748 if (*n == (*zn)->child_cnt - 1) {
749 err = tnc_next(c, zn, n);
751 /* Should be impossible */
757 ubifs_assert(*n == 0);
762 err = matches_name(c, &(*zn)->zbranch[*n], nm);
765 if (err == NAME_LESS)
767 if (err == NAME_MATCHES)
769 ubifs_assert(err == NAME_GREATER);
773 struct ubifs_znode *znode = *zn;
777 err = tnc_next(c, &znode, &nn);
782 if (keys_cmp(c, &znode->zbranch[nn].key, key))
784 err = matches_name(c, &znode->zbranch[nn], nm);
787 if (err == NAME_GREATER)
791 if (err == NAME_MATCHES)
793 ubifs_assert(err == NAME_LESS);
799 * fallible_matches_name - determine if a dent matches a given name.
800 * @c: UBIFS file-system description object
801 * @zbr: zbranch of dent
804 * This is a "fallible" version of 'matches_name()' function which does not
805 * panic if the direntry/xentry referred by @zbr does not exist on the media.
807 * This function checks if xentry/direntry referred by zbranch @zbr matches name
808 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
809 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
810 * if xentry/direntry referred by @zbr does not exist on the media. A negative
811 * error code is returned in case of failure.
813 static int fallible_matches_name(struct ubifs_info *c,
814 struct ubifs_zbranch *zbr,
815 const struct qstr *nm)
817 struct ubifs_dent_node *dent;
820 /* If possible, match against the dent in the leaf node cache */
822 dent = kmalloc(zbr->len, GFP_NOFS);
826 err = fallible_read_node(c, &zbr->key, zbr, dent);
830 /* The node was not present */
834 ubifs_assert(err == 1);
836 err = lnc_add_directly(c, zbr, dent);
842 nlen = le16_to_cpu(dent->nlen);
843 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
847 else if (nlen < nm->len)
862 * fallible_resolve_collision - resolve a collision even if nodes are missing.
863 * @c: UBIFS file-system description object
865 * @zn: znode is returned here
866 * @n: branch number is passed and returned here
867 * @nm: name of directory entry
868 * @adding: indicates caller is adding a key to the TNC
870 * This is a "fallible" version of the 'resolve_collision()' function which
871 * does not panic if one of the nodes referred to by TNC does not exist on the
872 * media. This may happen when replaying the journal if a deleted node was
873 * Garbage-collected and the commit was not done. A branch that refers to a node
874 * that is not present is called a dangling branch. The following are the return
875 * codes for this function:
876 * o if @nm was found, %1 is returned and @zn and @n are set to the found
878 * o if we are @adding and @nm was not found, %0 is returned;
879 * o if we are not @adding and @nm was not found, but a dangling branch was
880 * found, then %1 is returned and @zn and @n are set to the dangling branch;
881 * o a negative error code is returned in case of failure.
883 static int fallible_resolve_collision(struct ubifs_info *c,
884 const union ubifs_key *key,
885 struct ubifs_znode **zn, int *n,
886 const struct qstr *nm, int adding)
888 struct ubifs_znode *o_znode = NULL, *znode = *zn;
889 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
891 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
892 if (unlikely(cmp < 0))
894 if (cmp == NAME_MATCHES)
896 if (cmp == NOT_ON_MEDIA) {
900 * We are unlucky and hit a dangling branch straight away.
901 * Now we do not really know where to go to find the needed
902 * branch - to the left or to the right. Well, let's try left.
906 unsure = 1; /* Remove a dangling branch wherever it is */
908 if (cmp == NAME_GREATER || unsure) {
911 err = tnc_prev(c, zn, n);
912 if (err == -ENOENT) {
913 ubifs_assert(*n == 0);
919 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
920 /* See comments in 'resolve_collision()' */
921 if (*n == (*zn)->child_cnt - 1) {
922 err = tnc_next(c, zn, n);
924 /* Should be impossible */
930 ubifs_assert(*n == 0);
935 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
938 if (err == NAME_MATCHES)
940 if (err == NOT_ON_MEDIA) {
947 if (err == NAME_LESS)
954 if (cmp == NAME_LESS || unsure) {
959 err = tnc_next(c, &znode, &nn);
964 if (keys_cmp(c, &znode->zbranch[nn].key, key))
966 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
969 if (err == NAME_GREATER)
973 if (err == NAME_MATCHES)
975 if (err == NOT_ON_MEDIA) {
982 /* Never match a dangling branch when adding */
983 if (adding || !o_znode)
986 dbg_mnt("dangling match LEB %d:%d len %d %s",
987 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
988 o_znode->zbranch[o_n].len, DBGKEY(key));
995 * matches_position - determine if a zbranch matches a given position.
996 * @zbr: zbranch of dent
997 * @lnum: LEB number of dent to match
998 * @offs: offset of dent to match
1000 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1002 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1004 if (zbr->lnum == lnum && zbr->offs == offs)
1011 * resolve_collision_directly - resolve a collision directly.
1012 * @c: UBIFS file-system description object
1013 * @key: key of directory entry
1014 * @zn: znode is passed and returned here
1015 * @n: zbranch number is passed and returned here
1016 * @lnum: LEB number of dent node to match
1017 * @offs: offset of dent node to match
1019 * This function is used for "hashed" keys to make sure the found directory or
1020 * extended attribute entry node is what was looked for. It is used when the
1021 * flash address of the right node is known (@lnum:@offs) which makes it much
1022 * easier to resolve collisions (no need to read entries and match full
1023 * names). This function returns %1 and sets @zn and @n if the collision is
1024 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1025 * previous directory entry. Otherwise a negative error code is returned.
1027 static int resolve_collision_directly(struct ubifs_info *c,
1028 const union ubifs_key *key,
1029 struct ubifs_znode **zn, int *n,
1032 struct ubifs_znode *znode;
1037 if (matches_position(&znode->zbranch[nn], lnum, offs))
1042 err = tnc_prev(c, &znode, &nn);
1047 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1049 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1060 err = tnc_next(c, &znode, &nn);
1065 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1069 if (matches_position(&znode->zbranch[nn], lnum, offs))
1075 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1076 * @c: UBIFS file-system description object
1077 * @znode: znode to dirty
1079 * If we do not have a unique key that resides in a znode, then we cannot
1080 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1081 * This function records the path back to the last dirty ancestor, and then
1082 * dirties the znodes on that path.
1084 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1085 struct ubifs_znode *znode)
1087 struct ubifs_znode *zp;
1088 int *path = c->bottom_up_buf, p = 0;
1090 ubifs_assert(c->zroot.znode);
1091 ubifs_assert(znode);
1092 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1093 kfree(c->bottom_up_buf);
1094 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1096 if (!c->bottom_up_buf)
1097 return ERR_PTR(-ENOMEM);
1098 path = c->bottom_up_buf;
1100 if (c->zroot.znode->level) {
1101 /* Go up until parent is dirty */
1109 ubifs_assert(p < c->zroot.znode->level);
1111 if (!zp->cnext && ubifs_zn_dirty(znode))
1117 /* Come back down, dirtying as we go */
1119 struct ubifs_zbranch *zbr;
1123 ubifs_assert(path[p - 1] >= 0);
1124 ubifs_assert(path[p - 1] < zp->child_cnt);
1125 zbr = &zp->zbranch[path[--p]];
1126 znode = dirty_cow_znode(c, zbr);
1128 ubifs_assert(znode == c->zroot.znode);
1129 znode = dirty_cow_znode(c, &c->zroot);
1131 if (IS_ERR(znode) || !p)
1133 ubifs_assert(path[p - 1] >= 0);
1134 ubifs_assert(path[p - 1] < znode->child_cnt);
1135 znode = znode->zbranch[path[p - 1]].znode;
1142 * ubifs_lookup_level0 - search for zero-level znode.
1143 * @c: UBIFS file-system description object
1144 * @key: key to lookup
1145 * @zn: znode is returned here
1146 * @n: znode branch slot number is returned here
1148 * This function looks up the TNC tree and search for zero-level znode which
1149 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1151 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1152 * is returned and slot number of the matched branch is stored in @n;
1153 * o not exact match, which means that zero-level znode does not contain
1154 * @key, then %0 is returned and slot number of the closed branch is stored
1156 * o @key is so small that it is even less than the lowest key of the
1157 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1159 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1160 * function reads corresponding indexing nodes and inserts them to TNC. In
1161 * case of failure, a negative error code is returned.
1163 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1164 struct ubifs_znode **zn, int *n)
1167 struct ubifs_znode *znode;
1168 unsigned long time = get_seconds();
1170 dbg_tnc("search key %s", DBGKEY(key));
1172 znode = c->zroot.znode;
1173 if (unlikely(!znode)) {
1174 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1176 return PTR_ERR(znode);
1182 struct ubifs_zbranch *zbr;
1184 exact = ubifs_search_zbranch(c, znode, key, n);
1186 if (znode->level == 0)
1191 zbr = &znode->zbranch[*n];
1199 /* znode is not in TNC cache, load it from the media */
1200 znode = ubifs_load_znode(c, zbr, znode, *n);
1202 return PTR_ERR(znode);
1206 if (exact || !is_hash_key(c, key) || *n != -1) {
1207 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1212 * Here is a tricky place. We have not found the key and this is a
1213 * "hashed" key, which may collide. The rest of the code deals with
1214 * situations like this:
1218 * | 3 | 5 | | 6 | 7 | (x)
1220 * Or more a complex example:
1224 * | 1 | 3 | | 5 | 8 |
1226 * | 5 | 5 | | 6 | 7 | (x)
1228 * In the examples, if we are looking for key "5", we may reach nodes
1229 * marked with "(x)". In this case what we have do is to look at the
1230 * left and see if there is "5" key there. If there is, we have to
1233 * Note, this whole situation is possible because we allow to have
1234 * elements which are equivalent to the next key in the parent in the
1235 * children of current znode. For example, this happens if we split a
1236 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1240 * | 3 | 5 | | 5 | 6 | 7 |
1242 * And this becomes what is at the first "picture" after key "5" marked
1243 * with "^" is removed. What could be done is we could prohibit
1244 * splitting in the middle of the colliding sequence. Also, when
1245 * removing the leftmost key, we would have to correct the key of the
1246 * parent node, which would introduce additional complications. Namely,
1247 * if we changed the the leftmost key of the parent znode, the garbage
1248 * collector would be unable to find it (GC is doing this when GC'ing
1249 * indexing LEBs). Although we already have an additional RB-tree where
1250 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1251 * after the commit. But anyway, this does not look easy to implement
1252 * so we did not try this.
1254 err = tnc_prev(c, &znode, n);
1255 if (err == -ENOENT) {
1256 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1260 if (unlikely(err < 0))
1262 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1263 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1268 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1274 * lookup_level0_dirty - search for zero-level znode dirtying.
1275 * @c: UBIFS file-system description object
1276 * @key: key to lookup
1277 * @zn: znode is returned here
1278 * @n: znode branch slot number is returned here
1280 * This function looks up the TNC tree and search for zero-level znode which
1281 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1283 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1284 * is returned and slot number of the matched branch is stored in @n;
1285 * o not exact match, which means that zero-level znode does not contain @key
1286 * then %0 is returned and slot number of the closed branch is stored in
1288 * o @key is so small that it is even less than the lowest key of the
1289 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1291 * Additionally all znodes in the path from the root to the located zero-level
1292 * znode are marked as dirty.
1294 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1295 * function reads corresponding indexing nodes and inserts them to TNC. In
1296 * case of failure, a negative error code is returned.
1298 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1299 struct ubifs_znode **zn, int *n)
1302 struct ubifs_znode *znode;
1303 unsigned long time = get_seconds();
1305 dbg_tnc("search and dirty key %s", DBGKEY(key));
1307 znode = c->zroot.znode;
1308 if (unlikely(!znode)) {
1309 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1311 return PTR_ERR(znode);
1314 znode = dirty_cow_znode(c, &c->zroot);
1316 return PTR_ERR(znode);
1321 struct ubifs_zbranch *zbr;
1323 exact = ubifs_search_zbranch(c, znode, key, n);
1325 if (znode->level == 0)
1330 zbr = &znode->zbranch[*n];
1334 znode = dirty_cow_znode(c, zbr);
1336 return PTR_ERR(znode);
1340 /* znode is not in TNC cache, load it from the media */
1341 znode = ubifs_load_znode(c, zbr, znode, *n);
1343 return PTR_ERR(znode);
1344 znode = dirty_cow_znode(c, zbr);
1346 return PTR_ERR(znode);
1350 if (exact || !is_hash_key(c, key) || *n != -1) {
1351 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1356 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1359 err = tnc_prev(c, &znode, n);
1360 if (err == -ENOENT) {
1362 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1365 if (unlikely(err < 0))
1367 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1369 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1373 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1374 znode = dirty_cow_bottom_up(c, znode);
1376 return PTR_ERR(znode);
1379 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1385 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1386 * @c: UBIFS file-system description object
1388 * @gc_seq1: garbage collection sequence number
1390 * This function determines if @lnum may have been garbage collected since
1391 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1394 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1396 int gc_seq2, gced_lnum;
1398 gced_lnum = c->gced_lnum;
1400 gc_seq2 = c->gc_seq;
1401 /* Same seq means no GC */
1402 if (gc_seq1 == gc_seq2)
1404 /* Different by more than 1 means we don't know */
1405 if (gc_seq1 + 1 != gc_seq2)
1408 * We have seen the sequence number has increased by 1. Now we need to
1409 * be sure we read the right LEB number, so read it again.
1412 if (gced_lnum != c->gced_lnum)
1414 /* Finally we can check lnum */
1415 if (gced_lnum == lnum)
1421 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1422 * @c: UBIFS file-system description object
1423 * @key: node key to lookup
1424 * @node: the node is returned here
1425 * @lnum: LEB number is returned here
1426 * @offs: offset is returned here
1428 * This function look up and reads node with key @key. The caller has to make
1429 * sure the @node buffer is large enough to fit the node. Returns zero in case
1430 * of success, %-ENOENT if the node was not found, and a negative error code in
1431 * case of failure. The node location can be returned in @lnum and @offs.
1433 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1434 void *node, int *lnum, int *offs)
1436 int found, n, err, safely = 0, gc_seq1;
1437 struct ubifs_znode *znode;
1438 struct ubifs_zbranch zbr, *zt;
1441 mutex_lock(&c->tnc_mutex);
1442 found = ubifs_lookup_level0(c, key, &znode, &n);
1446 } else if (found < 0) {
1450 zt = &znode->zbranch[n];
1455 if (is_hash_key(c, key)) {
1457 * In this case the leaf node cache gets used, so we pass the
1458 * address of the zbranch and keep the mutex locked
1460 err = tnc_read_node_nm(c, zt, node);
1464 err = ubifs_tnc_read_node(c, zt, node);
1467 /* Drop the TNC mutex prematurely and race with garbage collection */
1468 zbr = znode->zbranch[n];
1469 gc_seq1 = c->gc_seq;
1470 mutex_unlock(&c->tnc_mutex);
1472 if (ubifs_get_wbuf(c, zbr.lnum)) {
1473 /* We do not GC journal heads */
1474 err = ubifs_tnc_read_node(c, &zbr, node);
1478 err = fallible_read_node(c, key, &zbr, node);
1479 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1481 * The node may have been GC'ed out from under us so try again
1482 * while keeping the TNC mutex locked.
1490 mutex_unlock(&c->tnc_mutex);
1495 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1496 * @c: UBIFS file-system description object
1497 * @bu: bulk-read parameters and results
1499 * Lookup consecutive data node keys for the same inode that reside
1500 * consecutively in the same LEB.
1502 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1504 int n, err = 0, lnum = -1, uninitialized_var(offs);
1505 int uninitialized_var(len);
1506 unsigned int block = key_block(c, &bu->key);
1507 struct ubifs_znode *znode;
1513 mutex_lock(&c->tnc_mutex);
1514 /* Find first key */
1515 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1520 len = znode->zbranch[n].len;
1521 /* The buffer must be big enough for at least 1 node */
1522 if (len > bu->buf_len) {
1527 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1529 lnum = znode->zbranch[n].lnum;
1530 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1533 struct ubifs_zbranch *zbr;
1534 union ubifs_key *key;
1535 unsigned int next_block;
1538 err = tnc_next(c, &znode, &n);
1541 zbr = &znode->zbranch[n];
1543 /* See if there is another data key for this file */
1544 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1545 key_type(c, key) != UBIFS_DATA_KEY) {
1550 /* First key found */
1552 offs = ALIGN(zbr->offs + zbr->len, 8);
1554 if (len > bu->buf_len) {
1560 * The data nodes must be in consecutive positions in
1563 if (zbr->lnum != lnum || zbr->offs != offs)
1565 offs += ALIGN(zbr->len, 8);
1566 len = ALIGN(len, 8) + zbr->len;
1567 /* Must not exceed buffer length */
1568 if (len > bu->buf_len)
1571 /* Allow for holes */
1572 next_block = key_block(c, key);
1573 bu->blk_cnt += (next_block - block - 1);
1574 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1578 bu->zbranch[bu->cnt++] = *zbr;
1580 /* See if we have room for more */
1581 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1583 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1587 if (err == -ENOENT) {
1591 bu->gc_seq = c->gc_seq;
1592 mutex_unlock(&c->tnc_mutex);
1596 * An enormous hole could cause bulk-read to encompass too many
1597 * page cache pages, so limit the number here.
1599 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1600 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1602 * Ensure that bulk-read covers a whole number of page cache
1605 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1606 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1609 /* At the end of file we can round up */
1610 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1613 /* Exclude data nodes that do not make up a whole page cache page */
1614 block = key_block(c, &bu->key) + bu->blk_cnt;
1615 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1617 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1625 * read_wbuf - bulk-read from a LEB with a wbuf.
1626 * @wbuf: wbuf that may overlap the read
1627 * @buf: buffer into which to read
1629 * @lnum: LEB number from which to read
1630 * @offs: offset from which to read
1632 * This functions returns %0 on success or a negative error code on failure.
1634 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1637 const struct ubifs_info *c = wbuf->c;
1640 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1641 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1642 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1643 ubifs_assert(offs + len <= c->leb_size);
1645 spin_lock(&wbuf->lock);
1646 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1648 /* We may safely unlock the write-buffer and read the data */
1649 spin_unlock(&wbuf->lock);
1650 return ubi_read(c->ubi, lnum, buf, offs, len);
1653 /* Don't read under wbuf */
1654 rlen = wbuf->offs - offs;
1658 /* Copy the rest from the write-buffer */
1659 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1660 spin_unlock(&wbuf->lock);
1663 /* Read everything that goes before write-buffer */
1664 return ubi_read(c->ubi, lnum, buf, offs, rlen);
1670 * validate_data_node - validate data nodes for bulk-read.
1671 * @c: UBIFS file-system description object
1672 * @buf: buffer containing data node to validate
1673 * @zbr: zbranch of data node to validate
1675 * This functions returns %0 on success or a negative error code on failure.
1677 static int validate_data_node(struct ubifs_info *c, void *buf,
1678 struct ubifs_zbranch *zbr)
1680 union ubifs_key key1;
1681 struct ubifs_ch *ch = buf;
1684 if (ch->node_type != UBIFS_DATA_NODE) {
1685 ubifs_err("bad node type (%d but expected %d)",
1686 ch->node_type, UBIFS_DATA_NODE);
1690 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0);
1692 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1696 len = le32_to_cpu(ch->len);
1697 if (len != zbr->len) {
1698 ubifs_err("bad node length %d, expected %d", len, zbr->len);
1702 /* Make sure the key of the read node is correct */
1703 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1704 if (!keys_eq(c, &zbr->key, &key1)) {
1705 ubifs_err("bad key in node at LEB %d:%d",
1706 zbr->lnum, zbr->offs);
1707 dbg_tnc("looked for key %s found node's key %s",
1708 DBGKEY(&zbr->key), DBGKEY1(&key1));
1717 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1718 dbg_dump_node(c, buf);
1724 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1725 * @c: UBIFS file-system description object
1726 * @bu: bulk-read parameters and results
1728 * This functions reads and validates the data nodes that were identified by the
1729 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1730 * -EAGAIN to indicate a race with GC, or another negative error code on
1733 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1735 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1736 struct ubifs_wbuf *wbuf;
1739 len = bu->zbranch[bu->cnt - 1].offs;
1740 len += bu->zbranch[bu->cnt - 1].len - offs;
1741 if (len > bu->buf_len) {
1742 ubifs_err("buffer too small %d vs %d", bu->buf_len, len);
1747 wbuf = ubifs_get_wbuf(c, lnum);
1749 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1751 err = ubi_read(c->ubi, lnum, bu->buf, offs, len);
1753 /* Check for a race with GC */
1754 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1757 if (err && err != -EBADMSG) {
1758 ubifs_err("failed to read from LEB %d:%d, error %d",
1761 dbg_tnc("key %s", DBGKEY(&bu->key));
1765 /* Validate the nodes read */
1767 for (i = 0; i < bu->cnt; i++) {
1768 err = validate_data_node(c, buf, &bu->zbranch[i]);
1771 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1778 * do_lookup_nm- look up a "hashed" node.
1779 * @c: UBIFS file-system description object
1780 * @key: node key to lookup
1781 * @node: the node is returned here
1784 * This function look up and reads a node which contains name hash in the key.
1785 * Since the hash may have collisions, there may be many nodes with the same
1786 * key, so we have to sequentially look to all of them until the needed one is
1787 * found. This function returns zero in case of success, %-ENOENT if the node
1788 * was not found, and a negative error code in case of failure.
1790 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1791 void *node, const struct qstr *nm)
1794 struct ubifs_znode *znode;
1796 dbg_tnc("name '%.*s' key %s", nm->len, nm->name, DBGKEY(key));
1797 mutex_lock(&c->tnc_mutex);
1798 found = ubifs_lookup_level0(c, key, &znode, &n);
1802 } else if (found < 0) {
1807 ubifs_assert(n >= 0);
1809 err = resolve_collision(c, key, &znode, &n, nm);
1810 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1811 if (unlikely(err < 0))
1818 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1821 mutex_unlock(&c->tnc_mutex);
1826 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1827 * @c: UBIFS file-system description object
1828 * @key: node key to lookup
1829 * @node: the node is returned here
1832 * This function look up and reads a node which contains name hash in the key.
1833 * Since the hash may have collisions, there may be many nodes with the same
1834 * key, so we have to sequentially look to all of them until the needed one is
1835 * found. This function returns zero in case of success, %-ENOENT if the node
1836 * was not found, and a negative error code in case of failure.
1838 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1839 void *node, const struct qstr *nm)
1842 const struct ubifs_dent_node *dent = node;
1845 * We assume that in most of the cases there are no name collisions and
1846 * 'ubifs_tnc_lookup()' returns us the right direntry.
1848 err = ubifs_tnc_lookup(c, key, node);
1852 len = le16_to_cpu(dent->nlen);
1853 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1857 * Unluckily, there are hash collisions and we have to iterate over
1858 * them look at each direntry with colliding name hash sequentially.
1860 return do_lookup_nm(c, key, node, nm);
1864 * correct_parent_keys - correct parent znodes' keys.
1865 * @c: UBIFS file-system description object
1866 * @znode: znode to correct parent znodes for
1868 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1869 * zbranch changes, keys of parent znodes have to be corrected. This helper
1870 * function is called in such situations and corrects the keys if needed.
1872 static void correct_parent_keys(const struct ubifs_info *c,
1873 struct ubifs_znode *znode)
1875 union ubifs_key *key, *key1;
1877 ubifs_assert(znode->parent);
1878 ubifs_assert(znode->iip == 0);
1880 key = &znode->zbranch[0].key;
1881 key1 = &znode->parent->zbranch[0].key;
1883 while (keys_cmp(c, key, key1) < 0) {
1884 key_copy(c, key, key1);
1885 znode = znode->parent;
1887 if (!znode->parent || znode->iip)
1889 key1 = &znode->parent->zbranch[0].key;
1894 * insert_zbranch - insert a zbranch into a znode.
1895 * @znode: znode into which to insert
1896 * @zbr: zbranch to insert
1897 * @n: slot number to insert to
1899 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1900 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1901 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1902 * slot, zbranches starting from @n have to be moved right.
1904 static void insert_zbranch(struct ubifs_znode *znode,
1905 const struct ubifs_zbranch *zbr, int n)
1909 ubifs_assert(ubifs_zn_dirty(znode));
1912 for (i = znode->child_cnt; i > n; i--) {
1913 znode->zbranch[i] = znode->zbranch[i - 1];
1914 if (znode->zbranch[i].znode)
1915 znode->zbranch[i].znode->iip = i;
1918 zbr->znode->iip = n;
1920 for (i = znode->child_cnt; i > n; i--)
1921 znode->zbranch[i] = znode->zbranch[i - 1];
1923 znode->zbranch[n] = *zbr;
1924 znode->child_cnt += 1;
1927 * After inserting at slot zero, the lower bound of the key range of
1928 * this znode may have changed. If this znode is subsequently split
1929 * then the upper bound of the key range may change, and furthermore
1930 * it could change to be lower than the original lower bound. If that
1931 * happens, then it will no longer be possible to find this znode in the
1932 * TNC using the key from the index node on flash. That is bad because
1933 * if it is not found, we will assume it is obsolete and may overwrite
1934 * it. Then if there is an unclean unmount, we will start using the
1935 * old index which will be broken.
1937 * So we first mark znodes that have insertions at slot zero, and then
1938 * if they are split we add their lnum/offs to the old_idx tree.
1945 * tnc_insert - insert a node into TNC.
1946 * @c: UBIFS file-system description object
1947 * @znode: znode to insert into
1948 * @zbr: branch to insert
1949 * @n: slot number to insert new zbranch to
1951 * This function inserts a new node described by @zbr into znode @znode. If
1952 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1953 * are splat as well if needed. Returns zero in case of success or a negative
1954 * error code in case of failure.
1956 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1957 struct ubifs_zbranch *zbr, int n)
1959 struct ubifs_znode *zn, *zi, *zp;
1960 int i, keep, move, appending = 0;
1961 union ubifs_key *key = &zbr->key;
1963 ubifs_assert(n >= 0 && n <= c->fanout);
1965 /* Implement naive insert for now */
1968 if (znode->child_cnt < c->fanout) {
1969 ubifs_assert(n != c->fanout);
1970 dbg_tnc("inserted at %d level %d, key %s", 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_tnc("splitting level %d, key %s", znode->level, DBGKEY(key));
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 && n == c->fanout &&
2003 key_type(c, key) == UBIFS_DATA_KEY) {
2004 union ubifs_key *key1;
2007 * If this is an inode which is being appended - do not split
2008 * it because no other zbranches can be inserted between
2009 * zbranches of consecutive data nodes anyway.
2011 key1 = &znode->zbranch[n - 1].key;
2012 if (key_inum(c, key1) == key_inum(c, key) &&
2013 key_type(c, key1) == UBIFS_DATA_KEY &&
2014 key_block(c, key1) == key_block(c, key) - 1)
2022 keep = (c->fanout + 1) / 2;
2023 move = c->fanout - keep;
2027 * Although we don't at present, we could look at the neighbors and see
2028 * if we can move some zbranches there.
2032 /* Insert into existing znode */
2037 /* Insert into new znode */
2042 zbr->znode->parent = zn;
2045 __set_bit(DIRTY_ZNODE, &zn->flags);
2046 atomic_long_inc(&c->dirty_zn_cnt);
2048 zn->child_cnt = move;
2049 znode->child_cnt = keep;
2051 dbg_tnc("moving %d, keeping %d", move, keep);
2054 for (i = 0; i < move; i++) {
2055 zn->zbranch[i] = znode->zbranch[keep + i];
2058 if (zn->zbranch[i].znode) {
2059 zn->zbranch[i].znode->parent = zn;
2060 zn->zbranch[i].znode->iip = i;
2064 /* Insert new key and branch */
2065 dbg_tnc("inserting at %d level %d, key %s", n, zn->level, DBGKEY(key));
2067 insert_zbranch(zi, zbr, n);
2069 /* Insert new znode (produced by spitting) into the parent */
2072 /* Locate insertion point */
2074 if (appending && n != c->fanout)
2077 if (i == 0 && zi == znode && znode->iip == 0)
2078 correct_parent_keys(c, znode);
2080 /* Tail recursion */
2081 zbr->key = zn->zbranch[0].key;
2091 /* We have to split root znode */
2092 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2094 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2099 zi->level = znode->level + 1;
2101 __set_bit(DIRTY_ZNODE, &zi->flags);
2102 atomic_long_inc(&c->dirty_zn_cnt);
2104 zi->zbranch[0].key = znode->zbranch[0].key;
2105 zi->zbranch[0].znode = znode;
2106 zi->zbranch[0].lnum = c->zroot.lnum;
2107 zi->zbranch[0].offs = c->zroot.offs;
2108 zi->zbranch[0].len = c->zroot.len;
2109 zi->zbranch[1].key = zn->zbranch[0].key;
2110 zi->zbranch[1].znode = zn;
2115 c->zroot.znode = zi;
2126 * ubifs_tnc_add - add a node to TNC.
2127 * @c: UBIFS file-system description object
2129 * @lnum: LEB number of node
2130 * @offs: node offset
2133 * This function adds a node with key @key to TNC. The node may be new or it may
2134 * obsolete some existing one. Returns %0 on success or negative error code on
2137 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2140 int found, n, err = 0;
2141 struct ubifs_znode *znode;
2143 mutex_lock(&c->tnc_mutex);
2144 dbg_tnc("%d:%d, len %d, key %s", lnum, offs, len, DBGKEY(key));
2145 found = lookup_level0_dirty(c, key, &znode, &n);
2147 struct ubifs_zbranch zbr;
2153 key_copy(c, key, &zbr.key);
2154 err = tnc_insert(c, znode, &zbr, n + 1);
2155 } else if (found == 1) {
2156 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2159 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2166 err = dbg_check_tnc(c, 0);
2167 mutex_unlock(&c->tnc_mutex);
2173 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2174 * @c: UBIFS file-system description object
2176 * @old_lnum: LEB number of old node
2177 * @old_offs: old node offset
2178 * @lnum: LEB number of node
2179 * @offs: node offset
2182 * This function replaces a node with key @key in the TNC only if the old node
2183 * is found. This function is called by garbage collection when node are moved.
2184 * Returns %0 on success or negative error code on failure.
2186 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2187 int old_lnum, int old_offs, int lnum, int offs, int len)
2189 int found, n, err = 0;
2190 struct ubifs_znode *znode;
2192 mutex_lock(&c->tnc_mutex);
2193 dbg_tnc("old LEB %d:%d, new LEB %d:%d, len %d, key %s", old_lnum,
2194 old_offs, lnum, offs, len, DBGKEY(key));
2195 found = lookup_level0_dirty(c, key, &znode, &n);
2202 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2205 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2207 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2214 } else if (is_hash_key(c, key)) {
2215 found = resolve_collision_directly(c, key, &znode, &n,
2216 old_lnum, old_offs);
2217 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2218 found, znode, n, old_lnum, old_offs);
2225 /* Ensure the znode is dirtied */
2226 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2227 znode = dirty_cow_bottom_up(c,
2229 if (IS_ERR(znode)) {
2230 err = PTR_ERR(znode);
2234 zbr = &znode->zbranch[n];
2236 err = ubifs_add_dirt(c, zbr->lnum,
2248 err = ubifs_add_dirt(c, lnum, len);
2251 err = dbg_check_tnc(c, 0);
2254 mutex_unlock(&c->tnc_mutex);
2259 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2260 * @c: UBIFS file-system description object
2262 * @lnum: LEB number of node
2263 * @offs: node offset
2267 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2268 * may have collisions, like directory entry keys.
2270 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2271 int lnum, int offs, int len, const struct qstr *nm)
2273 int found, n, err = 0;
2274 struct ubifs_znode *znode;
2276 mutex_lock(&c->tnc_mutex);
2277 dbg_tnc("LEB %d:%d, name '%.*s', key %s", lnum, offs, nm->len, nm->name,
2279 found = lookup_level0_dirty(c, key, &znode, &n);
2287 found = fallible_resolve_collision(c, key, &znode, &n,
2290 found = resolve_collision(c, key, &znode, &n, nm);
2291 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2297 /* Ensure the znode is dirtied */
2298 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2299 znode = dirty_cow_bottom_up(c, znode);
2300 if (IS_ERR(znode)) {
2301 err = PTR_ERR(znode);
2307 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2310 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2319 struct ubifs_zbranch zbr;
2325 key_copy(c, key, &zbr.key);
2326 err = tnc_insert(c, znode, &zbr, n + 1);
2331 * We did not find it in the index so there may be a
2332 * dangling branch still in the index. So we remove it
2333 * by passing 'ubifs_tnc_remove_nm()' the same key but
2334 * an unmatchable name.
2336 struct qstr noname = { .len = 0, .name = "" };
2338 err = dbg_check_tnc(c, 0);
2339 mutex_unlock(&c->tnc_mutex);
2342 return ubifs_tnc_remove_nm(c, key, &noname);
2348 err = dbg_check_tnc(c, 0);
2349 mutex_unlock(&c->tnc_mutex);
2354 * tnc_delete - delete a znode form TNC.
2355 * @c: UBIFS file-system description object
2356 * @znode: znode to delete from
2357 * @n: zbranch slot number to delete
2359 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2360 * case of success and a negative error code in case of failure.
2362 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2364 struct ubifs_zbranch *zbr;
2365 struct ubifs_znode *zp;
2368 /* Delete without merge for now */
2369 ubifs_assert(znode->level == 0);
2370 ubifs_assert(n >= 0 && n < c->fanout);
2371 dbg_tnc("deleting %s", DBGKEY(&znode->zbranch[n].key));
2373 zbr = &znode->zbranch[n];
2376 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2378 dbg_dump_znode(c, znode);
2382 /* We do not "gap" zbranch slots */
2383 for (i = n; i < znode->child_cnt - 1; i++)
2384 znode->zbranch[i] = znode->zbranch[i + 1];
2385 znode->child_cnt -= 1;
2387 if (znode->child_cnt > 0)
2391 * This was the last zbranch, we have to delete this znode from the
2396 ubifs_assert(!test_bit(OBSOLETE_ZNODE, &znode->flags));
2397 ubifs_assert(ubifs_zn_dirty(znode));
2402 atomic_long_dec(&c->dirty_zn_cnt);
2404 err = insert_old_idx_znode(c, znode);
2409 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2410 atomic_long_inc(&c->clean_zn_cnt);
2411 atomic_long_inc(&ubifs_clean_zn_cnt);
2415 } while (znode->child_cnt == 1); /* while removing last child */
2417 /* Remove from znode, entry n - 1 */
2418 znode->child_cnt -= 1;
2419 ubifs_assert(znode->level != 0);
2420 for (i = n; i < znode->child_cnt; i++) {
2421 znode->zbranch[i] = znode->zbranch[i + 1];
2422 if (znode->zbranch[i].znode)
2423 znode->zbranch[i].znode->iip = i;
2427 * If this is the root and it has only 1 child then
2428 * collapse the tree.
2430 if (!znode->parent) {
2431 while (znode->child_cnt == 1 && znode->level != 0) {
2433 zbr = &znode->zbranch[0];
2434 znode = get_znode(c, znode, 0);
2436 return PTR_ERR(znode);
2437 znode = dirty_cow_znode(c, zbr);
2439 return PTR_ERR(znode);
2440 znode->parent = NULL;
2443 err = insert_old_idx(c, c->zroot.lnum,
2448 c->zroot.lnum = zbr->lnum;
2449 c->zroot.offs = zbr->offs;
2450 c->zroot.len = zbr->len;
2451 c->zroot.znode = znode;
2452 ubifs_assert(!test_bit(OBSOLETE_ZNODE,
2454 ubifs_assert(test_bit(DIRTY_ZNODE, &zp->flags));
2455 atomic_long_dec(&c->dirty_zn_cnt);
2458 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2459 atomic_long_inc(&c->clean_zn_cnt);
2460 atomic_long_inc(&ubifs_clean_zn_cnt);
2470 * ubifs_tnc_remove - remove an index entry of a node.
2471 * @c: UBIFS file-system description object
2474 * Returns %0 on success or negative error code on failure.
2476 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2478 int found, n, err = 0;
2479 struct ubifs_znode *znode;
2481 mutex_lock(&c->tnc_mutex);
2482 dbg_tnc("key %s", DBGKEY(key));
2483 found = lookup_level0_dirty(c, key, &znode, &n);
2489 err = tnc_delete(c, znode, n);
2491 err = dbg_check_tnc(c, 0);
2494 mutex_unlock(&c->tnc_mutex);
2499 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2500 * @c: UBIFS file-system description object
2502 * @nm: directory entry name
2504 * Returns %0 on success or negative error code on failure.
2506 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2507 const struct qstr *nm)
2510 struct ubifs_znode *znode;
2512 mutex_lock(&c->tnc_mutex);
2513 dbg_tnc("%.*s, key %s", nm->len, nm->name, DBGKEY(key));
2514 err = lookup_level0_dirty(c, key, &znode, &n);
2520 err = fallible_resolve_collision(c, key, &znode, &n,
2523 err = resolve_collision(c, key, &znode, &n, nm);
2524 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2528 /* Ensure the znode is dirtied */
2529 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2530 znode = dirty_cow_bottom_up(c, znode);
2531 if (IS_ERR(znode)) {
2532 err = PTR_ERR(znode);
2536 err = tnc_delete(c, znode, n);
2542 err = dbg_check_tnc(c, 0);
2543 mutex_unlock(&c->tnc_mutex);
2548 * key_in_range - determine if a key falls within a range of keys.
2549 * @c: UBIFS file-system description object
2550 * @key: key to check
2551 * @from_key: lowest key in range
2552 * @to_key: highest key in range
2554 * This function returns %1 if the key is in range and %0 otherwise.
2556 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2557 union ubifs_key *from_key, union ubifs_key *to_key)
2559 if (keys_cmp(c, key, from_key) < 0)
2561 if (keys_cmp(c, key, to_key) > 0)
2567 * ubifs_tnc_remove_range - remove index entries in range.
2568 * @c: UBIFS file-system description object
2569 * @from_key: lowest key to remove
2570 * @to_key: highest key to remove
2572 * This function removes index entries starting at @from_key and ending at
2573 * @to_key. This function returns zero in case of success and a negative error
2574 * code in case of failure.
2576 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2577 union ubifs_key *to_key)
2579 int i, n, k, err = 0;
2580 struct ubifs_znode *znode;
2581 union ubifs_key *key;
2583 mutex_lock(&c->tnc_mutex);
2585 /* Find first level 0 znode that contains keys to remove */
2586 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2593 err = tnc_next(c, &znode, &n);
2594 if (err == -ENOENT) {
2600 key = &znode->zbranch[n].key;
2601 if (!key_in_range(c, key, from_key, to_key)) {
2607 /* Ensure the znode is dirtied */
2608 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2609 znode = dirty_cow_bottom_up(c, znode);
2610 if (IS_ERR(znode)) {
2611 err = PTR_ERR(znode);
2616 /* Remove all keys in range except the first */
2617 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2618 key = &znode->zbranch[i].key;
2619 if (!key_in_range(c, key, from_key, to_key))
2621 lnc_free(&znode->zbranch[i]);
2622 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2623 znode->zbranch[i].len);
2625 dbg_dump_znode(c, znode);
2628 dbg_tnc("removing %s", DBGKEY(key));
2631 for (i = n + 1 + k; i < znode->child_cnt; i++)
2632 znode->zbranch[i - k] = znode->zbranch[i];
2633 znode->child_cnt -= k;
2636 /* Now delete the first */
2637 err = tnc_delete(c, znode, n);
2644 err = dbg_check_tnc(c, 0);
2645 mutex_unlock(&c->tnc_mutex);
2650 * ubifs_tnc_remove_ino - remove an inode from TNC.
2651 * @c: UBIFS file-system description object
2652 * @inum: inode number to remove
2654 * This function remove inode @inum and all the extended attributes associated
2655 * with the anode from TNC and returns zero in case of success or a negative
2656 * error code in case of failure.
2658 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2660 union ubifs_key key1, key2;
2661 struct ubifs_dent_node *xent, *pxent = NULL;
2662 struct qstr nm = { .name = NULL };
2664 dbg_tnc("ino %lu", inum);
2667 * Walk all extended attribute entries and remove them together with
2668 * corresponding extended attribute inodes.
2670 lowest_xent_key(c, &key1, inum);
2675 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2677 err = PTR_ERR(xent);
2683 xattr_inum = le64_to_cpu(xent->inum);
2684 dbg_tnc("xent '%s', ino %lu", xent->name, xattr_inum);
2686 nm.name = xent->name;
2687 nm.len = le16_to_cpu(xent->nlen);
2688 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2694 lowest_ino_key(c, &key1, xattr_inum);
2695 highest_ino_key(c, &key2, xattr_inum);
2696 err = ubifs_tnc_remove_range(c, &key1, &key2);
2704 key_read(c, &xent->key, &key1);
2708 lowest_ino_key(c, &key1, inum);
2709 highest_ino_key(c, &key2, inum);
2711 return ubifs_tnc_remove_range(c, &key1, &key2);
2715 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2716 * @c: UBIFS file-system description object
2717 * @key: key of last entry
2718 * @nm: name of last entry found or %NULL
2720 * This function finds and reads the next directory or extended attribute entry
2721 * after the given key (@key) if there is one. @nm is used to resolve
2724 * If the name of the current entry is not known and only the key is known,
2725 * @nm->name has to be %NULL. In this case the semantics of this function is a
2726 * little bit different and it returns the entry corresponding to this key, not
2727 * the next one. If the key was not found, the closest "right" entry is
2730 * If the fist entry has to be found, @key has to contain the lowest possible
2731 * key value for this inode and @name has to be %NULL.
2733 * This function returns the found directory or extended attribute entry node
2734 * in case of success, %-ENOENT is returned if no entry was found, and a
2735 * negative error code is returned in case of failure.
2737 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2738 union ubifs_key *key,
2739 const struct qstr *nm)
2741 int n, err, type = key_type(c, key);
2742 struct ubifs_znode *znode;
2743 struct ubifs_dent_node *dent;
2744 struct ubifs_zbranch *zbr;
2745 union ubifs_key *dkey;
2747 dbg_tnc("%s %s", nm->name ? (char *)nm->name : "(lowest)", DBGKEY(key));
2748 ubifs_assert(is_hash_key(c, key));
2750 mutex_lock(&c->tnc_mutex);
2751 err = ubifs_lookup_level0(c, key, &znode, &n);
2752 if (unlikely(err < 0))
2757 /* Handle collisions */
2758 err = resolve_collision(c, key, &znode, &n, nm);
2759 dbg_tnc("rc returned %d, znode %p, n %d",
2761 if (unlikely(err < 0))
2765 /* Now find next entry */
2766 err = tnc_next(c, &znode, &n);
2771 * The full name of the entry was not given, in which case the
2772 * behavior of this function is a little different and it
2773 * returns current entry, not the next one.
2777 * However, the given key does not exist in the TNC
2778 * tree and @znode/@n variables contain the closest
2779 * "preceding" element. Switch to the next one.
2781 err = tnc_next(c, &znode, &n);
2787 zbr = &znode->zbranch[n];
2788 dent = kmalloc(zbr->len, GFP_NOFS);
2789 if (unlikely(!dent)) {
2795 * The above 'tnc_next()' call could lead us to the next inode, check
2799 if (key_inum(c, dkey) != key_inum(c, key) ||
2800 key_type(c, dkey) != type) {
2805 err = tnc_read_node_nm(c, zbr, dent);
2809 mutex_unlock(&c->tnc_mutex);
2815 mutex_unlock(&c->tnc_mutex);
2816 return ERR_PTR(err);
2820 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2821 * @c: UBIFS file-system description object
2823 * Destroy left-over obsolete znodes from a failed commit.
2825 static void tnc_destroy_cnext(struct ubifs_info *c)
2827 struct ubifs_znode *cnext;
2831 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2834 struct ubifs_znode *znode = cnext;
2836 cnext = cnext->cnext;
2837 if (test_bit(OBSOLETE_ZNODE, &znode->flags))
2839 } while (cnext && cnext != c->cnext);
2843 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2844 * @c: UBIFS file-system description object
2846 void ubifs_tnc_close(struct ubifs_info *c)
2850 tnc_destroy_cnext(c);
2851 if (c->zroot.znode) {
2852 clean_freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
2853 atomic_long_sub(clean_freed, &ubifs_clean_zn_cnt);
2861 * left_znode - get the znode to the left.
2862 * @c: UBIFS file-system description object
2865 * This function returns a pointer to the znode to the left of @znode or NULL if
2866 * there is not one. A negative error code is returned on failure.
2868 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2869 struct ubifs_znode *znode)
2871 int level = znode->level;
2874 int n = znode->iip - 1;
2876 /* Go up until we can go left */
2877 znode = znode->parent;
2881 /* Now go down the rightmost branch to 'level' */
2882 znode = get_znode(c, znode, n);
2885 while (znode->level != level) {
2886 n = znode->child_cnt - 1;
2887 znode = get_znode(c, znode, n);
2898 * right_znode - get the znode to the right.
2899 * @c: UBIFS file-system description object
2902 * This function returns a pointer to the znode to the right of @znode or NULL
2903 * if there is not one. A negative error code is returned on failure.
2905 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2906 struct ubifs_znode *znode)
2908 int level = znode->level;
2911 int n = znode->iip + 1;
2913 /* Go up until we can go right */
2914 znode = znode->parent;
2917 if (n < znode->child_cnt) {
2918 /* Now go down the leftmost branch to 'level' */
2919 znode = get_znode(c, znode, n);
2922 while (znode->level != level) {
2923 znode = get_znode(c, znode, 0);
2934 * lookup_znode - find a particular indexing node from TNC.
2935 * @c: UBIFS file-system description object
2936 * @key: index node key to lookup
2937 * @level: index node level
2938 * @lnum: index node LEB number
2939 * @offs: index node offset
2941 * This function searches an indexing node by its first key @key and its
2942 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2943 * nodes it traverses to TNC. This function is called fro indexing nodes which
2944 * were found on the media by scanning, for example when garbage-collecting or
2945 * when doing in-the-gaps commit. This means that the indexing node which is
2946 * looked for does not have to have exactly the same leftmost key @key, because
2947 * the leftmost key may have been changed, in which case TNC will contain a
2948 * dirty znode which still refers the same @lnum:@offs. This function is clever
2949 * enough to recognize such indexing nodes.
2951 * Note, if a znode was deleted or changed too much, then this function will
2952 * not find it. For situations like this UBIFS has the old index RB-tree
2953 * (indexed by @lnum:@offs).
2955 * This function returns a pointer to the znode found or %NULL if it is not
2956 * found. A negative error code is returned on failure.
2958 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2959 union ubifs_key *key, int level,
2962 struct ubifs_znode *znode, *zn;
2966 * The arguments have probably been read off flash, so don't assume
2970 return ERR_PTR(-EINVAL);
2972 /* Get the root znode */
2973 znode = c->zroot.znode;
2975 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
2979 /* Check if it is the one we are looking for */
2980 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
2982 /* Descend to the parent level i.e. (level + 1) */
2983 if (level >= znode->level)
2986 ubifs_search_zbranch(c, znode, key, &n);
2989 * We reached a znode where the leftmost key is greater
2990 * than the key we are searching for. This is the same
2991 * situation as the one described in a huge comment at
2992 * the end of the 'ubifs_lookup_level0()' function. And
2993 * for exactly the same reasons we have to try to look
2994 * left before giving up.
2996 znode = left_znode(c, znode);
3001 ubifs_search_zbranch(c, znode, key, &n);
3002 ubifs_assert(n >= 0);
3004 if (znode->level == level + 1)
3006 znode = get_znode(c, znode, n);
3010 /* Check if the child is the one we are looking for */
3011 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3012 return get_znode(c, znode, n);
3013 /* If the key is unique, there is nowhere else to look */
3014 if (!is_hash_key(c, key))
3017 * The key is not unique and so may be also in the znodes to either
3024 /* Move one branch to the left */
3028 znode = left_znode(c, znode);
3033 n = znode->child_cnt - 1;
3036 if (znode->zbranch[n].lnum == lnum &&
3037 znode->zbranch[n].offs == offs)
3038 return get_znode(c, znode, n);
3039 /* Stop if the key is less than the one we are looking for */
3040 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3043 /* Back to the middle */
3048 /* Move one branch to the right */
3049 if (++n >= znode->child_cnt) {
3050 znode = right_znode(c, znode);
3058 if (znode->zbranch[n].lnum == lnum &&
3059 znode->zbranch[n].offs == offs)
3060 return get_znode(c, znode, n);
3061 /* Stop if the key is greater than the one we are looking for */
3062 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3069 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3070 * @c: UBIFS file-system description object
3071 * @key: key of index node
3072 * @level: index node level
3073 * @lnum: LEB number of index node
3074 * @offs: offset of index node
3076 * This function returns %0 if the index node is not referred to in the TNC, %1
3077 * if the index node is referred to in the TNC and the corresponding znode is
3078 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3079 * znode is clean, and a negative error code in case of failure.
3081 * Note, the @key argument has to be the key of the first child. Also note,
3082 * this function relies on the fact that 0:0 is never a valid LEB number and
3083 * offset for a main-area node.
3085 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3088 struct ubifs_znode *znode;
3090 znode = lookup_znode(c, key, level, lnum, offs);
3094 return PTR_ERR(znode);
3096 return ubifs_zn_dirty(znode) ? 1 : 2;
3100 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3101 * @c: UBIFS file-system description object
3103 * @lnum: node LEB number
3104 * @offs: node offset
3106 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3107 * not, and a negative error code in case of failure.
3109 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3110 * and offset for a main-area node.
3112 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3115 struct ubifs_zbranch *zbr;
3116 struct ubifs_znode *znode, *zn;
3117 int n, found, err, nn;
3118 const int unique = !is_hash_key(c, key);
3120 found = ubifs_lookup_level0(c, key, &znode, &n);
3122 return found; /* Error code */
3125 zbr = &znode->zbranch[n];
3126 if (lnum == zbr->lnum && offs == zbr->offs)
3127 return 1; /* Found it */
3131 * Because the key is not unique, we have to look left
3138 err = tnc_prev(c, &znode, &n);
3143 if (keys_cmp(c, key, &znode->zbranch[n].key))
3145 zbr = &znode->zbranch[n];
3146 if (lnum == zbr->lnum && offs == zbr->offs)
3147 return 1; /* Found it */
3153 err = tnc_next(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 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3170 * @c: UBIFS file-system description object
3172 * @level: index node level (if it is an index node)
3173 * @lnum: node LEB number
3174 * @offs: node offset
3175 * @is_idx: non-zero if the node is an index node
3177 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3178 * negative error code in case of failure. For index nodes, @key has to be the
3179 * key of the first child. An index node is considered to be in the TNC only if
3180 * the corresponding znode is clean or has not been loaded.
3182 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3183 int lnum, int offs, int is_idx)
3187 mutex_lock(&c->tnc_mutex);
3189 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3193 /* The index node was found but it was dirty */
3196 /* The index node was found and it was clean */
3201 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3204 mutex_unlock(&c->tnc_mutex);
3209 * ubifs_dirty_idx_node - dirty an index node.
3210 * @c: UBIFS file-system description object
3211 * @key: index node key
3212 * @level: index node level
3213 * @lnum: index node LEB number
3214 * @offs: index node offset
3216 * This function loads and dirties an index node so that it can be garbage
3217 * collected. The @key argument has to be the key of the first child. This
3218 * function relies on the fact that 0:0 is never a valid LEB number and offset
3219 * for a main-area node. Returns %0 on success and a negative error code on
3222 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3225 struct ubifs_znode *znode;
3228 mutex_lock(&c->tnc_mutex);
3229 znode = lookup_znode(c, key, level, lnum, offs);
3232 if (IS_ERR(znode)) {
3233 err = PTR_ERR(znode);
3236 znode = dirty_cow_bottom_up(c, znode);
3237 if (IS_ERR(znode)) {
3238 err = PTR_ERR(znode);
3243 mutex_unlock(&c->tnc_mutex);