2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
27 * At the moment the locking rules of the TNC tree are quite simple and
28 * straightforward. We just have a mutex and lock it when we traverse the
29 * tree. If a znode is not in memory, we read it from flash while still having
33 #include <linux/crc32.h>
34 #include <linux/slab.h>
38 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
39 * @NAME_LESS: name corresponding to the first argument is less than second
40 * @NAME_MATCHES: names match
41 * @NAME_GREATER: name corresponding to the second argument is greater than
43 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
45 * These constants were introduce to improve readability.
55 * insert_old_idx - record an index node obsoleted since the last commit start.
56 * @c: UBIFS file-system description object
57 * @lnum: LEB number of obsoleted index node
58 * @offs: offset of obsoleted index node
60 * Returns %0 on success, and a negative error code on failure.
62 * For recovery, there must always be a complete intact version of the index on
63 * flash at all times. That is called the "old index". It is the index as at the
64 * time of the last successful commit. Many of the index nodes in the old index
65 * may be dirty, but they must not be erased until the next successful commit
66 * (at which point that index becomes the old index).
68 * That means that the garbage collection and the in-the-gaps method of
69 * committing must be able to determine if an index node is in the old index.
70 * Most of the old index nodes can be found by looking up the TNC using the
71 * 'lookup_znode()' function. However, some of the old index nodes may have
72 * been deleted from the current index or may have been changed so much that
73 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
74 * That is what this function does. The RB-tree is ordered by LEB number and
75 * offset because they uniquely identify the old index node.
77 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
79 struct ubifs_old_idx *old_idx, *o;
80 struct rb_node **p, *parent = NULL;
82 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
83 if (unlikely(!old_idx))
88 p = &c->old_idx.rb_node;
91 o = rb_entry(parent, struct ubifs_old_idx, rb);
94 else if (lnum > o->lnum)
96 else if (offs < o->offs)
98 else if (offs > o->offs)
101 ubifs_err("old idx added twice!");
106 rb_link_node(&old_idx->rb, parent, p);
107 rb_insert_color(&old_idx->rb, &c->old_idx);
112 * insert_old_idx_znode - record a znode obsoleted since last commit start.
113 * @c: UBIFS file-system description object
114 * @znode: znode of obsoleted index node
116 * Returns %0 on success, and a negative error code on failure.
118 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
121 struct ubifs_zbranch *zbr;
123 zbr = &znode->parent->zbranch[znode->iip];
125 return insert_old_idx(c, zbr->lnum, zbr->offs);
128 return insert_old_idx(c, c->zroot.lnum,
134 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
135 * @c: UBIFS file-system description object
136 * @znode: znode of obsoleted index node
138 * Returns %0 on success, and a negative error code on failure.
140 static int ins_clr_old_idx_znode(struct ubifs_info *c,
141 struct ubifs_znode *znode)
146 struct ubifs_zbranch *zbr;
148 zbr = &znode->parent->zbranch[znode->iip];
150 err = insert_old_idx(c, zbr->lnum, zbr->offs);
159 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
170 * destroy_old_idx - destroy the old_idx RB-tree.
171 * @c: UBIFS file-system description object
173 * During start commit, the old_idx RB-tree is used to avoid overwriting index
174 * nodes that were in the index last commit but have since been deleted. This
175 * is necessary for recovery i.e. the old index must be kept intact until the
176 * new index is successfully written. The old-idx RB-tree is used for the
177 * in-the-gaps method of writing index nodes and is destroyed every commit.
179 void destroy_old_idx(struct ubifs_info *c)
181 struct rb_node *this = c->old_idx.rb_node;
182 struct ubifs_old_idx *old_idx;
186 this = this->rb_left;
188 } else if (this->rb_right) {
189 this = this->rb_right;
192 old_idx = rb_entry(this, struct ubifs_old_idx, rb);
193 this = rb_parent(this);
195 if (this->rb_left == &old_idx->rb)
196 this->rb_left = NULL;
198 this->rb_right = NULL;
202 c->old_idx = RB_ROOT;
206 * copy_znode - copy a dirty znode.
207 * @c: UBIFS file-system description object
208 * @znode: znode to copy
210 * A dirty znode being committed may not be changed, so it is copied.
212 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
213 struct ubifs_znode *znode)
215 struct ubifs_znode *zn;
217 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
219 return ERR_PTR(-ENOMEM);
221 memcpy(zn, znode, c->max_znode_sz);
223 __set_bit(DIRTY_ZNODE, &zn->flags);
224 __clear_bit(COW_ZNODE, &zn->flags);
226 ubifs_assert(!ubifs_zn_obsolete(znode));
227 __set_bit(OBSOLETE_ZNODE, &znode->flags);
229 if (znode->level != 0) {
231 const int n = zn->child_cnt;
233 /* The children now have new parent */
234 for (i = 0; i < n; i++) {
235 struct ubifs_zbranch *zbr = &zn->zbranch[i];
238 zbr->znode->parent = zn;
242 atomic_long_inc(&c->dirty_zn_cnt);
247 * add_idx_dirt - add dirt due to a dirty znode.
248 * @c: UBIFS file-system description object
249 * @lnum: LEB number of index node
250 * @dirt: size of index node
252 * This function updates lprops dirty space and the new size of the index.
254 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
256 c->calc_idx_sz -= ALIGN(dirt, 8);
257 return ubifs_add_dirt(c, lnum, dirt);
261 * dirty_cow_znode - ensure a znode is not being committed.
262 * @c: UBIFS file-system description object
263 * @zbr: branch of znode to check
265 * Returns dirtied znode on success or negative error code on failure.
267 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
268 struct ubifs_zbranch *zbr)
270 struct ubifs_znode *znode = zbr->znode;
271 struct ubifs_znode *zn;
274 if (!ubifs_zn_cow(znode)) {
275 /* znode is not being committed */
276 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
277 atomic_long_inc(&c->dirty_zn_cnt);
278 atomic_long_dec(&c->clean_zn_cnt);
279 atomic_long_dec(&ubifs_clean_zn_cnt);
280 err = add_idx_dirt(c, zbr->lnum, zbr->len);
287 zn = copy_znode(c, znode);
292 err = insert_old_idx(c, zbr->lnum, zbr->offs);
295 err = add_idx_dirt(c, zbr->lnum, zbr->len);
310 * lnc_add - add a leaf node to the leaf node cache.
311 * @c: UBIFS file-system description object
312 * @zbr: zbranch of leaf node
315 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
316 * purpose of the leaf node cache is to save re-reading the same leaf node over
317 * and over again. Most things are cached by VFS, however the file system must
318 * cache directory entries for readdir and for resolving hash collisions. The
319 * present implementation of the leaf node cache is extremely simple, and
320 * allows for error returns that are not used but that may be needed if a more
321 * complex implementation is created.
323 * Note, this function does not add the @node object to LNC directly, but
324 * allocates a copy of the object and adds the copy to LNC. The reason for this
325 * is that @node has been allocated outside of the TNC subsystem and will be
326 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
327 * may be changed at any time, e.g. freed by the shrinker.
329 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
334 const struct ubifs_dent_node *dent = node;
336 ubifs_assert(!zbr->leaf);
337 ubifs_assert(zbr->len != 0);
338 ubifs_assert(is_hash_key(c, &zbr->key));
340 err = ubifs_validate_entry(c, dent);
343 ubifs_dump_node(c, dent);
347 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
349 /* We don't have to have the cache, so no error */
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 ubifs_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 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
448 * is true (it is controlled by corresponding mount option). However, if
449 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
450 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
451 * because during mounting or re-mounting from R/O mode to R/W mode we may read
452 * journal nodes (when replying the journal or doing the recovery) and the
453 * journal nodes may potentially be corrupted, so checking is required.
455 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
456 int len, int lnum, int offs)
459 struct ubifs_ch *ch = buf;
460 uint32_t crc, node_crc;
462 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
464 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
466 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
467 type, lnum, offs, err);
471 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
474 if (ch->node_type != type)
477 node_len = le32_to_cpu(ch->len);
481 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
485 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
486 node_crc = le32_to_cpu(ch->crc);
494 * fallible_read_node - try to read a leaf node.
495 * @c: UBIFS file-system description object
496 * @key: key of node to read
497 * @zbr: position of node
498 * @node: node returned
500 * This function tries to read a node and returns %1 if the node is read, %0
501 * if the node is not present, and a negative error code in the case of error.
503 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
504 struct ubifs_zbranch *zbr, void *node)
508 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
510 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
513 union ubifs_key node_key;
514 struct ubifs_dent_node *dent = node;
516 /* All nodes have key in the same place */
517 key_read(c, &dent->key, &node_key);
518 if (keys_cmp(c, key, &node_key) != 0)
521 if (ret == 0 && c->replaying)
522 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
523 zbr->lnum, zbr->offs, zbr->len);
528 * matches_name - determine if a direntry or xattr entry matches a given name.
529 * @c: UBIFS file-system description object
530 * @zbr: zbranch of dent
533 * This function checks if xentry/direntry referred by zbranch @zbr matches name
534 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
535 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
536 * of failure, a negative error code is returned.
538 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
539 const struct qstr *nm)
541 struct ubifs_dent_node *dent;
544 /* If possible, match against the dent in the leaf node cache */
546 dent = kmalloc(zbr->len, GFP_NOFS);
550 err = ubifs_tnc_read_node(c, zbr, dent);
554 /* Add the node to the leaf node cache */
555 err = lnc_add_directly(c, zbr, dent);
561 nlen = le16_to_cpu(dent->nlen);
562 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
566 else if (nlen < nm->len)
581 * get_znode - get a TNC znode that may not be loaded yet.
582 * @c: UBIFS file-system description object
583 * @znode: parent znode
584 * @n: znode branch slot number
586 * This function returns the znode or a negative error code.
588 static struct ubifs_znode *get_znode(struct ubifs_info *c,
589 struct ubifs_znode *znode, int n)
591 struct ubifs_zbranch *zbr;
593 zbr = &znode->zbranch[n];
597 znode = ubifs_load_znode(c, zbr, znode, n);
602 * tnc_next - find next TNC entry.
603 * @c: UBIFS file-system description object
604 * @zn: znode is passed and returned here
605 * @n: znode branch slot number is passed and returned here
607 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
608 * no next entry, or a negative error code otherwise.
610 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
612 struct ubifs_znode *znode = *zn;
616 if (nn < znode->child_cnt) {
621 struct ubifs_znode *zp;
628 if (nn < znode->child_cnt) {
629 znode = get_znode(c, znode, nn);
631 return PTR_ERR(znode);
632 while (znode->level != 0) {
633 znode = get_znode(c, znode, 0);
635 return PTR_ERR(znode);
647 * tnc_prev - find previous TNC entry.
648 * @c: UBIFS file-system description object
649 * @zn: znode is returned here
650 * @n: znode branch slot number is passed and returned here
652 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
653 * there is no next entry, or a negative error code otherwise.
655 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
657 struct ubifs_znode *znode = *zn;
665 struct ubifs_znode *zp;
673 znode = get_znode(c, znode, nn);
675 return PTR_ERR(znode);
676 while (znode->level != 0) {
677 nn = znode->child_cnt - 1;
678 znode = get_znode(c, znode, nn);
680 return PTR_ERR(znode);
682 nn = znode->child_cnt - 1;
692 * resolve_collision - resolve a collision.
693 * @c: UBIFS file-system description object
694 * @key: key of a directory or extended attribute entry
695 * @zn: znode is returned here
696 * @n: zbranch number is passed and returned here
697 * @nm: name of the entry
699 * This function is called for "hashed" keys to make sure that the found key
700 * really corresponds to the looked up node (directory or extended attribute
701 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
702 * %0 is returned if @nm is not found and @zn and @n are set to the previous
703 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
704 * This means that @n may be set to %-1 if the leftmost key in @zn is the
705 * previous one. A negative error code is returned on failures.
707 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
708 struct ubifs_znode **zn, int *n,
709 const struct qstr *nm)
713 err = matches_name(c, &(*zn)->zbranch[*n], nm);
714 if (unlikely(err < 0))
716 if (err == NAME_MATCHES)
719 if (err == NAME_GREATER) {
722 err = tnc_prev(c, zn, n);
723 if (err == -ENOENT) {
724 ubifs_assert(*n == 0);
730 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
732 * We have found the branch after which we would
733 * like to insert, but inserting in this znode
734 * may still be wrong. Consider the following 3
735 * znodes, in the case where we are resolving a
736 * collision with Key2.
739 * ----------------------
740 * level 1 | Key0 | Key1 |
741 * -----------------------
743 * znode za | | znode zb
744 * ------------ ------------
745 * level 0 | Key0 | | Key2 |
746 * ------------ ------------
748 * The lookup finds Key2 in znode zb. Lets say
749 * there is no match and the name is greater so
750 * we look left. When we find Key0, we end up
751 * here. If we return now, we will insert into
752 * znode za at slot n = 1. But that is invalid
753 * according to the parent's keys. Key2 must
754 * be inserted into znode zb.
756 * Note, this problem is not relevant for the
757 * case when we go right, because
758 * 'tnc_insert()' would correct the parent key.
760 if (*n == (*zn)->child_cnt - 1) {
761 err = tnc_next(c, zn, n);
763 /* Should be impossible */
769 ubifs_assert(*n == 0);
774 err = matches_name(c, &(*zn)->zbranch[*n], nm);
777 if (err == NAME_LESS)
779 if (err == NAME_MATCHES)
781 ubifs_assert(err == NAME_GREATER);
785 struct ubifs_znode *znode = *zn;
789 err = tnc_next(c, &znode, &nn);
794 if (keys_cmp(c, &znode->zbranch[nn].key, key))
796 err = matches_name(c, &znode->zbranch[nn], nm);
799 if (err == NAME_GREATER)
803 if (err == NAME_MATCHES)
805 ubifs_assert(err == NAME_LESS);
811 * fallible_matches_name - determine if a dent matches a given name.
812 * @c: UBIFS file-system description object
813 * @zbr: zbranch of dent
816 * This is a "fallible" version of 'matches_name()' function which does not
817 * panic if the direntry/xentry referred by @zbr does not exist on the media.
819 * This function checks if xentry/direntry referred by zbranch @zbr matches name
820 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
821 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
822 * if xentry/direntry referred by @zbr does not exist on the media. A negative
823 * error code is returned in case of failure.
825 static int fallible_matches_name(struct ubifs_info *c,
826 struct ubifs_zbranch *zbr,
827 const struct qstr *nm)
829 struct ubifs_dent_node *dent;
832 /* If possible, match against the dent in the leaf node cache */
834 dent = kmalloc(zbr->len, GFP_NOFS);
838 err = fallible_read_node(c, &zbr->key, zbr, dent);
842 /* The node was not present */
846 ubifs_assert(err == 1);
848 err = lnc_add_directly(c, zbr, dent);
854 nlen = le16_to_cpu(dent->nlen);
855 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
859 else if (nlen < nm->len)
874 * fallible_resolve_collision - resolve a collision even if nodes are missing.
875 * @c: UBIFS file-system description object
877 * @zn: znode is returned here
878 * @n: branch number is passed and returned here
879 * @nm: name of directory entry
880 * @adding: indicates caller is adding a key to the TNC
882 * This is a "fallible" version of the 'resolve_collision()' function which
883 * does not panic if one of the nodes referred to by TNC does not exist on the
884 * media. This may happen when replaying the journal if a deleted node was
885 * Garbage-collected and the commit was not done. A branch that refers to a node
886 * that is not present is called a dangling branch. The following are the return
887 * codes for this function:
888 * o if @nm was found, %1 is returned and @zn and @n are set to the found
890 * o if we are @adding and @nm was not found, %0 is returned;
891 * o if we are not @adding and @nm was not found, but a dangling branch was
892 * found, then %1 is returned and @zn and @n are set to the dangling branch;
893 * o a negative error code is returned in case of failure.
895 static int fallible_resolve_collision(struct ubifs_info *c,
896 const union ubifs_key *key,
897 struct ubifs_znode **zn, int *n,
898 const struct qstr *nm, int adding)
900 struct ubifs_znode *o_znode = NULL, *znode = *zn;
901 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
903 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
904 if (unlikely(cmp < 0))
906 if (cmp == NAME_MATCHES)
908 if (cmp == NOT_ON_MEDIA) {
912 * We are unlucky and hit a dangling branch straight away.
913 * Now we do not really know where to go to find the needed
914 * branch - to the left or to the right. Well, let's try left.
918 unsure = 1; /* Remove a dangling branch wherever it is */
920 if (cmp == NAME_GREATER || unsure) {
923 err = tnc_prev(c, zn, n);
924 if (err == -ENOENT) {
925 ubifs_assert(*n == 0);
931 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
932 /* See comments in 'resolve_collision()' */
933 if (*n == (*zn)->child_cnt - 1) {
934 err = tnc_next(c, zn, n);
936 /* Should be impossible */
942 ubifs_assert(*n == 0);
947 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
950 if (err == NAME_MATCHES)
952 if (err == NOT_ON_MEDIA) {
959 if (err == NAME_LESS)
966 if (cmp == NAME_LESS || unsure) {
971 err = tnc_next(c, &znode, &nn);
976 if (keys_cmp(c, &znode->zbranch[nn].key, key))
978 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
981 if (err == NAME_GREATER)
985 if (err == NAME_MATCHES)
987 if (err == NOT_ON_MEDIA) {
994 /* Never match a dangling branch when adding */
995 if (adding || !o_znode)
998 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
999 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1000 o_znode->zbranch[o_n].len);
1007 * matches_position - determine if a zbranch matches a given position.
1008 * @zbr: zbranch of dent
1009 * @lnum: LEB number of dent to match
1010 * @offs: offset of dent to match
1012 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1014 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1016 if (zbr->lnum == lnum && zbr->offs == offs)
1023 * resolve_collision_directly - resolve a collision directly.
1024 * @c: UBIFS file-system description object
1025 * @key: key of directory entry
1026 * @zn: znode is passed and returned here
1027 * @n: zbranch number is passed and returned here
1028 * @lnum: LEB number of dent node to match
1029 * @offs: offset of dent node to match
1031 * This function is used for "hashed" keys to make sure the found directory or
1032 * extended attribute entry node is what was looked for. It is used when the
1033 * flash address of the right node is known (@lnum:@offs) which makes it much
1034 * easier to resolve collisions (no need to read entries and match full
1035 * names). This function returns %1 and sets @zn and @n if the collision is
1036 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1037 * previous directory entry. Otherwise a negative error code is returned.
1039 static int resolve_collision_directly(struct ubifs_info *c,
1040 const union ubifs_key *key,
1041 struct ubifs_znode **zn, int *n,
1044 struct ubifs_znode *znode;
1049 if (matches_position(&znode->zbranch[nn], lnum, offs))
1054 err = tnc_prev(c, &znode, &nn);
1059 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1061 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1072 err = tnc_next(c, &znode, &nn);
1077 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1081 if (matches_position(&znode->zbranch[nn], lnum, offs))
1087 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1088 * @c: UBIFS file-system description object
1089 * @znode: znode to dirty
1091 * If we do not have a unique key that resides in a znode, then we cannot
1092 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1093 * This function records the path back to the last dirty ancestor, and then
1094 * dirties the znodes on that path.
1096 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1097 struct ubifs_znode *znode)
1099 struct ubifs_znode *zp;
1100 int *path = c->bottom_up_buf, p = 0;
1102 ubifs_assert(c->zroot.znode);
1103 ubifs_assert(znode);
1104 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1105 kfree(c->bottom_up_buf);
1106 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1108 if (!c->bottom_up_buf)
1109 return ERR_PTR(-ENOMEM);
1110 path = c->bottom_up_buf;
1112 if (c->zroot.znode->level) {
1113 /* Go up until parent is dirty */
1121 ubifs_assert(p < c->zroot.znode->level);
1123 if (!zp->cnext && ubifs_zn_dirty(znode))
1129 /* Come back down, dirtying as we go */
1131 struct ubifs_zbranch *zbr;
1135 ubifs_assert(path[p - 1] >= 0);
1136 ubifs_assert(path[p - 1] < zp->child_cnt);
1137 zbr = &zp->zbranch[path[--p]];
1138 znode = dirty_cow_znode(c, zbr);
1140 ubifs_assert(znode == c->zroot.znode);
1141 znode = dirty_cow_znode(c, &c->zroot);
1143 if (IS_ERR(znode) || !p)
1145 ubifs_assert(path[p - 1] >= 0);
1146 ubifs_assert(path[p - 1] < znode->child_cnt);
1147 znode = znode->zbranch[path[p - 1]].znode;
1154 * ubifs_lookup_level0 - search for zero-level znode.
1155 * @c: UBIFS file-system description object
1156 * @key: key to lookup
1157 * @zn: znode is returned here
1158 * @n: znode branch slot number is returned here
1160 * This function looks up the TNC tree and search for zero-level znode which
1161 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1163 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1164 * is returned and slot number of the matched branch is stored in @n;
1165 * o not exact match, which means that zero-level znode does not contain
1166 * @key, then %0 is returned and slot number of the closest branch is stored
1168 * o @key is so small that it is even less than the lowest key of the
1169 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1171 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1172 * function reads corresponding indexing nodes and inserts them to TNC. In
1173 * case of failure, a negative error code is returned.
1175 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1176 struct ubifs_znode **zn, int *n)
1179 struct ubifs_znode *znode;
1180 unsigned long time = get_seconds();
1182 dbg_tnck(key, "search key ");
1183 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1185 znode = c->zroot.znode;
1186 if (unlikely(!znode)) {
1187 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1189 return PTR_ERR(znode);
1195 struct ubifs_zbranch *zbr;
1197 exact = ubifs_search_zbranch(c, znode, key, n);
1199 if (znode->level == 0)
1204 zbr = &znode->zbranch[*n];
1212 /* znode is not in TNC cache, load it from the media */
1213 znode = ubifs_load_znode(c, zbr, znode, *n);
1215 return PTR_ERR(znode);
1219 if (exact || !is_hash_key(c, key) || *n != -1) {
1220 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1225 * Here is a tricky place. We have not found the key and this is a
1226 * "hashed" key, which may collide. The rest of the code deals with
1227 * situations like this:
1231 * | 3 | 5 | | 6 | 7 | (x)
1233 * Or more a complex example:
1237 * | 1 | 3 | | 5 | 8 |
1239 * | 5 | 5 | | 6 | 7 | (x)
1241 * In the examples, if we are looking for key "5", we may reach nodes
1242 * marked with "(x)". In this case what we have do is to look at the
1243 * left and see if there is "5" key there. If there is, we have to
1246 * Note, this whole situation is possible because we allow to have
1247 * elements which are equivalent to the next key in the parent in the
1248 * children of current znode. For example, this happens if we split a
1249 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1253 * | 3 | 5 | | 5 | 6 | 7 |
1255 * And this becomes what is at the first "picture" after key "5" marked
1256 * with "^" is removed. What could be done is we could prohibit
1257 * splitting in the middle of the colliding sequence. Also, when
1258 * removing the leftmost key, we would have to correct the key of the
1259 * parent node, which would introduce additional complications. Namely,
1260 * if we changed the leftmost key of the parent znode, the garbage
1261 * collector would be unable to find it (GC is doing this when GC'ing
1262 * indexing LEBs). Although we already have an additional RB-tree where
1263 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1264 * after the commit. But anyway, this does not look easy to implement
1265 * so we did not try this.
1267 err = tnc_prev(c, &znode, n);
1268 if (err == -ENOENT) {
1269 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1273 if (unlikely(err < 0))
1275 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1276 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1281 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1287 * lookup_level0_dirty - search for zero-level znode dirtying.
1288 * @c: UBIFS file-system description object
1289 * @key: key to lookup
1290 * @zn: znode is returned here
1291 * @n: znode branch slot number is returned here
1293 * This function looks up the TNC tree and search for zero-level znode which
1294 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1296 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1297 * is returned and slot number of the matched branch is stored in @n;
1298 * o not exact match, which means that zero-level znode does not contain @key
1299 * then %0 is returned and slot number of the closed branch is stored in
1301 * o @key is so small that it is even less than the lowest key of the
1302 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1304 * Additionally all znodes in the path from the root to the located zero-level
1305 * znode are marked as dirty.
1307 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1308 * function reads corresponding indexing nodes and inserts them to TNC. In
1309 * case of failure, a negative error code is returned.
1311 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1312 struct ubifs_znode **zn, int *n)
1315 struct ubifs_znode *znode;
1316 unsigned long time = get_seconds();
1318 dbg_tnck(key, "search and dirty key ");
1320 znode = c->zroot.znode;
1321 if (unlikely(!znode)) {
1322 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1324 return PTR_ERR(znode);
1327 znode = dirty_cow_znode(c, &c->zroot);
1329 return PTR_ERR(znode);
1334 struct ubifs_zbranch *zbr;
1336 exact = ubifs_search_zbranch(c, znode, key, n);
1338 if (znode->level == 0)
1343 zbr = &znode->zbranch[*n];
1347 znode = dirty_cow_znode(c, zbr);
1349 return PTR_ERR(znode);
1353 /* znode is not in TNC cache, load it from the media */
1354 znode = ubifs_load_znode(c, zbr, znode, *n);
1356 return PTR_ERR(znode);
1357 znode = dirty_cow_znode(c, zbr);
1359 return PTR_ERR(znode);
1363 if (exact || !is_hash_key(c, key) || *n != -1) {
1364 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1369 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1372 err = tnc_prev(c, &znode, n);
1373 if (err == -ENOENT) {
1375 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1378 if (unlikely(err < 0))
1380 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1382 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1386 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1387 znode = dirty_cow_bottom_up(c, znode);
1389 return PTR_ERR(znode);
1392 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1398 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1399 * @c: UBIFS file-system description object
1401 * @gc_seq1: garbage collection sequence number
1403 * This function determines if @lnum may have been garbage collected since
1404 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1407 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1409 int gc_seq2, gced_lnum;
1411 gced_lnum = c->gced_lnum;
1413 gc_seq2 = c->gc_seq;
1414 /* Same seq means no GC */
1415 if (gc_seq1 == gc_seq2)
1417 /* Different by more than 1 means we don't know */
1418 if (gc_seq1 + 1 != gc_seq2)
1421 * We have seen the sequence number has increased by 1. Now we need to
1422 * be sure we read the right LEB number, so read it again.
1425 if (gced_lnum != c->gced_lnum)
1427 /* Finally we can check lnum */
1428 if (gced_lnum == lnum)
1434 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1435 * @c: UBIFS file-system description object
1436 * @key: node key to lookup
1437 * @node: the node is returned here
1438 * @lnum: LEB number is returned here
1439 * @offs: offset is returned here
1441 * This function looks up and reads node with key @key. The caller has to make
1442 * sure the @node buffer is large enough to fit the node. Returns zero in case
1443 * of success, %-ENOENT if the node was not found, and a negative error code in
1444 * case of failure. The node location can be returned in @lnum and @offs.
1446 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1447 void *node, int *lnum, int *offs)
1449 int found, n, err, safely = 0, gc_seq1;
1450 struct ubifs_znode *znode;
1451 struct ubifs_zbranch zbr, *zt;
1454 mutex_lock(&c->tnc_mutex);
1455 found = ubifs_lookup_level0(c, key, &znode, &n);
1459 } else if (found < 0) {
1463 zt = &znode->zbranch[n];
1468 if (is_hash_key(c, key)) {
1470 * In this case the leaf node cache gets used, so we pass the
1471 * address of the zbranch and keep the mutex locked
1473 err = tnc_read_node_nm(c, zt, node);
1477 err = ubifs_tnc_read_node(c, zt, node);
1480 /* Drop the TNC mutex prematurely and race with garbage collection */
1481 zbr = znode->zbranch[n];
1482 gc_seq1 = c->gc_seq;
1483 mutex_unlock(&c->tnc_mutex);
1485 if (ubifs_get_wbuf(c, zbr.lnum)) {
1486 /* We do not GC journal heads */
1487 err = ubifs_tnc_read_node(c, &zbr, node);
1491 err = fallible_read_node(c, key, &zbr, node);
1492 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1494 * The node may have been GC'ed out from under us so try again
1495 * while keeping the TNC mutex locked.
1503 mutex_unlock(&c->tnc_mutex);
1508 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1509 * @c: UBIFS file-system description object
1510 * @bu: bulk-read parameters and results
1512 * Lookup consecutive data node keys for the same inode that reside
1513 * consecutively in the same LEB. This function returns zero in case of success
1514 * and a negative error code in case of failure.
1516 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1517 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1518 * maximum possible amount of nodes for bulk-read.
1520 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1522 int n, err = 0, lnum = -1, uninitialized_var(offs);
1523 int uninitialized_var(len);
1524 unsigned int block = key_block(c, &bu->key);
1525 struct ubifs_znode *znode;
1531 mutex_lock(&c->tnc_mutex);
1532 /* Find first key */
1533 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1538 len = znode->zbranch[n].len;
1539 /* The buffer must be big enough for at least 1 node */
1540 if (len > bu->buf_len) {
1545 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1547 lnum = znode->zbranch[n].lnum;
1548 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1551 struct ubifs_zbranch *zbr;
1552 union ubifs_key *key;
1553 unsigned int next_block;
1556 err = tnc_next(c, &znode, &n);
1559 zbr = &znode->zbranch[n];
1561 /* See if there is another data key for this file */
1562 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1563 key_type(c, key) != UBIFS_DATA_KEY) {
1568 /* First key found */
1570 offs = ALIGN(zbr->offs + zbr->len, 8);
1572 if (len > bu->buf_len) {
1578 * The data nodes must be in consecutive positions in
1581 if (zbr->lnum != lnum || zbr->offs != offs)
1583 offs += ALIGN(zbr->len, 8);
1584 len = ALIGN(len, 8) + zbr->len;
1585 /* Must not exceed buffer length */
1586 if (len > bu->buf_len)
1589 /* Allow for holes */
1590 next_block = key_block(c, key);
1591 bu->blk_cnt += (next_block - block - 1);
1592 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1596 bu->zbranch[bu->cnt++] = *zbr;
1598 /* See if we have room for more */
1599 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1601 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1605 if (err == -ENOENT) {
1609 bu->gc_seq = c->gc_seq;
1610 mutex_unlock(&c->tnc_mutex);
1614 * An enormous hole could cause bulk-read to encompass too many
1615 * page cache pages, so limit the number here.
1617 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1618 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1620 * Ensure that bulk-read covers a whole number of page cache
1623 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1624 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1627 /* At the end of file we can round up */
1628 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1631 /* Exclude data nodes that do not make up a whole page cache page */
1632 block = key_block(c, &bu->key) + bu->blk_cnt;
1633 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1635 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1643 * read_wbuf - bulk-read from a LEB with a wbuf.
1644 * @wbuf: wbuf that may overlap the read
1645 * @buf: buffer into which to read
1647 * @lnum: LEB number from which to read
1648 * @offs: offset from which to read
1650 * This functions returns %0 on success or a negative error code on failure.
1652 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1655 const struct ubifs_info *c = wbuf->c;
1658 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1659 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1660 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1661 ubifs_assert(offs + len <= c->leb_size);
1663 spin_lock(&wbuf->lock);
1664 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1666 /* We may safely unlock the write-buffer and read the data */
1667 spin_unlock(&wbuf->lock);
1668 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1671 /* Don't read under wbuf */
1672 rlen = wbuf->offs - offs;
1676 /* Copy the rest from the write-buffer */
1677 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1678 spin_unlock(&wbuf->lock);
1681 /* Read everything that goes before write-buffer */
1682 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1688 * validate_data_node - validate data nodes for bulk-read.
1689 * @c: UBIFS file-system description object
1690 * @buf: buffer containing data node to validate
1691 * @zbr: zbranch of data node to validate
1693 * This functions returns %0 on success or a negative error code on failure.
1695 static int validate_data_node(struct ubifs_info *c, void *buf,
1696 struct ubifs_zbranch *zbr)
1698 union ubifs_key key1;
1699 struct ubifs_ch *ch = buf;
1702 if (ch->node_type != UBIFS_DATA_NODE) {
1703 ubifs_err("bad node type (%d but expected %d)",
1704 ch->node_type, UBIFS_DATA_NODE);
1708 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1710 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1714 len = le32_to_cpu(ch->len);
1715 if (len != zbr->len) {
1716 ubifs_err("bad node length %d, expected %d", len, zbr->len);
1720 /* Make sure the key of the read node is correct */
1721 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1722 if (!keys_eq(c, &zbr->key, &key1)) {
1723 ubifs_err("bad key in node at LEB %d:%d",
1724 zbr->lnum, zbr->offs);
1725 dbg_tnck(&zbr->key, "looked for key ");
1726 dbg_tnck(&key1, "found node's key ");
1735 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1736 ubifs_dump_node(c, buf);
1742 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1743 * @c: UBIFS file-system description object
1744 * @bu: bulk-read parameters and results
1746 * This functions reads and validates the data nodes that were identified by the
1747 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1748 * -EAGAIN to indicate a race with GC, or another negative error code on
1751 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1753 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1754 struct ubifs_wbuf *wbuf;
1757 len = bu->zbranch[bu->cnt - 1].offs;
1758 len += bu->zbranch[bu->cnt - 1].len - offs;
1759 if (len > bu->buf_len) {
1760 ubifs_err("buffer too small %d vs %d", bu->buf_len, len);
1765 wbuf = ubifs_get_wbuf(c, lnum);
1767 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1769 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1771 /* Check for a race with GC */
1772 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1775 if (err && err != -EBADMSG) {
1776 ubifs_err("failed to read from LEB %d:%d, error %d",
1779 dbg_tnck(&bu->key, "key ");
1783 /* Validate the nodes read */
1785 for (i = 0; i < bu->cnt; i++) {
1786 err = validate_data_node(c, buf, &bu->zbranch[i]);
1789 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1796 * do_lookup_nm- look up a "hashed" node.
1797 * @c: UBIFS file-system description object
1798 * @key: node key to lookup
1799 * @node: the node is returned here
1802 * This function look up and reads a node which contains name hash in the key.
1803 * Since the hash may have collisions, there may be many nodes with the same
1804 * key, so we have to sequentially look to all of them until the needed one is
1805 * found. This function returns zero in case of success, %-ENOENT if the node
1806 * was not found, and a negative error code in case of failure.
1808 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1809 void *node, const struct qstr *nm)
1812 struct ubifs_znode *znode;
1814 dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name);
1815 mutex_lock(&c->tnc_mutex);
1816 found = ubifs_lookup_level0(c, key, &znode, &n);
1820 } else if (found < 0) {
1825 ubifs_assert(n >= 0);
1827 err = resolve_collision(c, key, &znode, &n, nm);
1828 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1829 if (unlikely(err < 0))
1836 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1839 mutex_unlock(&c->tnc_mutex);
1844 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1845 * @c: UBIFS file-system description object
1846 * @key: node key to lookup
1847 * @node: the node is returned here
1850 * This function look up and reads a node which contains name hash in the key.
1851 * Since the hash may have collisions, there may be many nodes with the same
1852 * key, so we have to sequentially look to all of them until the needed one is
1853 * found. This function returns zero in case of success, %-ENOENT if the node
1854 * was not found, and a negative error code in case of failure.
1856 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1857 void *node, const struct qstr *nm)
1860 const struct ubifs_dent_node *dent = node;
1863 * We assume that in most of the cases there are no name collisions and
1864 * 'ubifs_tnc_lookup()' returns us the right direntry.
1866 err = ubifs_tnc_lookup(c, key, node);
1870 len = le16_to_cpu(dent->nlen);
1871 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1875 * Unluckily, there are hash collisions and we have to iterate over
1876 * them look at each direntry with colliding name hash sequentially.
1878 return do_lookup_nm(c, key, node, nm);
1882 * correct_parent_keys - correct parent znodes' keys.
1883 * @c: UBIFS file-system description object
1884 * @znode: znode to correct parent znodes for
1886 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1887 * zbranch changes, keys of parent znodes have to be corrected. This helper
1888 * function is called in such situations and corrects the keys if needed.
1890 static void correct_parent_keys(const struct ubifs_info *c,
1891 struct ubifs_znode *znode)
1893 union ubifs_key *key, *key1;
1895 ubifs_assert(znode->parent);
1896 ubifs_assert(znode->iip == 0);
1898 key = &znode->zbranch[0].key;
1899 key1 = &znode->parent->zbranch[0].key;
1901 while (keys_cmp(c, key, key1) < 0) {
1902 key_copy(c, key, key1);
1903 znode = znode->parent;
1905 if (!znode->parent || znode->iip)
1907 key1 = &znode->parent->zbranch[0].key;
1912 * insert_zbranch - insert a zbranch into a znode.
1913 * @znode: znode into which to insert
1914 * @zbr: zbranch to insert
1915 * @n: slot number to insert to
1917 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1918 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1919 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1920 * slot, zbranches starting from @n have to be moved right.
1922 static void insert_zbranch(struct ubifs_znode *znode,
1923 const struct ubifs_zbranch *zbr, int n)
1927 ubifs_assert(ubifs_zn_dirty(znode));
1930 for (i = znode->child_cnt; i > n; i--) {
1931 znode->zbranch[i] = znode->zbranch[i - 1];
1932 if (znode->zbranch[i].znode)
1933 znode->zbranch[i].znode->iip = i;
1936 zbr->znode->iip = n;
1938 for (i = znode->child_cnt; i > n; i--)
1939 znode->zbranch[i] = znode->zbranch[i - 1];
1941 znode->zbranch[n] = *zbr;
1942 znode->child_cnt += 1;
1945 * After inserting at slot zero, the lower bound of the key range of
1946 * this znode may have changed. If this znode is subsequently split
1947 * then the upper bound of the key range may change, and furthermore
1948 * it could change to be lower than the original lower bound. If that
1949 * happens, then it will no longer be possible to find this znode in the
1950 * TNC using the key from the index node on flash. That is bad because
1951 * if it is not found, we will assume it is obsolete and may overwrite
1952 * it. Then if there is an unclean unmount, we will start using the
1953 * old index which will be broken.
1955 * So we first mark znodes that have insertions at slot zero, and then
1956 * if they are split we add their lnum/offs to the old_idx tree.
1963 * tnc_insert - insert a node into TNC.
1964 * @c: UBIFS file-system description object
1965 * @znode: znode to insert into
1966 * @zbr: branch to insert
1967 * @n: slot number to insert new zbranch to
1969 * This function inserts a new node described by @zbr into znode @znode. If
1970 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1971 * are splat as well if needed. Returns zero in case of success or a negative
1972 * error code in case of failure.
1974 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1975 struct ubifs_zbranch *zbr, int n)
1977 struct ubifs_znode *zn, *zi, *zp;
1978 int i, keep, move, appending = 0;
1979 union ubifs_key *key = &zbr->key, *key1;
1981 ubifs_assert(n >= 0 && n <= c->fanout);
1983 /* Implement naive insert for now */
1986 if (znode->child_cnt < c->fanout) {
1987 ubifs_assert(n != c->fanout);
1988 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
1990 insert_zbranch(znode, zbr, n);
1992 /* Ensure parent's key is correct */
1993 if (n == 0 && zp && znode->iip == 0)
1994 correct_parent_keys(c, znode);
2000 * Unfortunately, @znode does not have more empty slots and we have to
2003 dbg_tnck(key, "splitting level %d, key ", znode->level);
2007 * We can no longer be sure of finding this znode by key, so we
2008 * record it in the old_idx tree.
2010 ins_clr_old_idx_znode(c, znode);
2012 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2016 zn->level = znode->level;
2018 /* Decide where to split */
2019 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2020 /* Try not to split consecutive data keys */
2021 if (n == c->fanout) {
2022 key1 = &znode->zbranch[n - 1].key;
2023 if (key_inum(c, key1) == key_inum(c, key) &&
2024 key_type(c, key1) == UBIFS_DATA_KEY)
2028 } else if (appending && n != c->fanout) {
2029 /* Try not to split consecutive data keys */
2032 if (n >= (c->fanout + 1) / 2) {
2033 key1 = &znode->zbranch[0].key;
2034 if (key_inum(c, key1) == key_inum(c, key) &&
2035 key_type(c, key1) == UBIFS_DATA_KEY) {
2036 key1 = &znode->zbranch[n].key;
2037 if (key_inum(c, key1) != key_inum(c, key) ||
2038 key_type(c, key1) != UBIFS_DATA_KEY) {
2040 move = c->fanout - keep;
2052 keep = (c->fanout + 1) / 2;
2053 move = c->fanout - keep;
2057 * Although we don't at present, we could look at the neighbors and see
2058 * if we can move some zbranches there.
2062 /* Insert into existing znode */
2067 /* Insert into new znode */
2072 zbr->znode->parent = zn;
2077 __set_bit(DIRTY_ZNODE, &zn->flags);
2078 atomic_long_inc(&c->dirty_zn_cnt);
2080 zn->child_cnt = move;
2081 znode->child_cnt = keep;
2083 dbg_tnc("moving %d, keeping %d", move, keep);
2086 for (i = 0; i < move; i++) {
2087 zn->zbranch[i] = znode->zbranch[keep + i];
2090 if (zn->zbranch[i].znode) {
2091 zn->zbranch[i].znode->parent = zn;
2092 zn->zbranch[i].znode->iip = i;
2096 /* Insert new key and branch */
2097 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2099 insert_zbranch(zi, zbr, n);
2101 /* Insert new znode (produced by spitting) into the parent */
2103 if (n == 0 && zi == znode && znode->iip == 0)
2104 correct_parent_keys(c, znode);
2106 /* Locate insertion point */
2109 /* Tail recursion */
2110 zbr->key = zn->zbranch[0].key;
2120 /* We have to split root znode */
2121 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2123 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2128 zi->level = znode->level + 1;
2130 __set_bit(DIRTY_ZNODE, &zi->flags);
2131 atomic_long_inc(&c->dirty_zn_cnt);
2133 zi->zbranch[0].key = znode->zbranch[0].key;
2134 zi->zbranch[0].znode = znode;
2135 zi->zbranch[0].lnum = c->zroot.lnum;
2136 zi->zbranch[0].offs = c->zroot.offs;
2137 zi->zbranch[0].len = c->zroot.len;
2138 zi->zbranch[1].key = zn->zbranch[0].key;
2139 zi->zbranch[1].znode = zn;
2144 c->zroot.znode = zi;
2155 * ubifs_tnc_add - add a node to TNC.
2156 * @c: UBIFS file-system description object
2158 * @lnum: LEB number of node
2159 * @offs: node offset
2162 * This function adds a node with key @key to TNC. The node may be new or it may
2163 * obsolete some existing one. Returns %0 on success or negative error code on
2166 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2169 int found, n, err = 0;
2170 struct ubifs_znode *znode;
2172 mutex_lock(&c->tnc_mutex);
2173 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2174 found = lookup_level0_dirty(c, key, &znode, &n);
2176 struct ubifs_zbranch zbr;
2182 key_copy(c, key, &zbr.key);
2183 err = tnc_insert(c, znode, &zbr, n + 1);
2184 } else if (found == 1) {
2185 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2188 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2195 err = dbg_check_tnc(c, 0);
2196 mutex_unlock(&c->tnc_mutex);
2202 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2203 * @c: UBIFS file-system description object
2205 * @old_lnum: LEB number of old node
2206 * @old_offs: old node offset
2207 * @lnum: LEB number of node
2208 * @offs: node offset
2211 * This function replaces a node with key @key in the TNC only if the old node
2212 * is found. This function is called by garbage collection when node are moved.
2213 * Returns %0 on success or negative error code on failure.
2215 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2216 int old_lnum, int old_offs, int lnum, int offs, int len)
2218 int found, n, err = 0;
2219 struct ubifs_znode *znode;
2221 mutex_lock(&c->tnc_mutex);
2222 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2223 old_offs, lnum, offs, len);
2224 found = lookup_level0_dirty(c, key, &znode, &n);
2231 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2234 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2236 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2243 } else if (is_hash_key(c, key)) {
2244 found = resolve_collision_directly(c, key, &znode, &n,
2245 old_lnum, old_offs);
2246 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2247 found, znode, n, old_lnum, old_offs);
2254 /* Ensure the znode is dirtied */
2255 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2256 znode = dirty_cow_bottom_up(c, znode);
2257 if (IS_ERR(znode)) {
2258 err = PTR_ERR(znode);
2262 zbr = &znode->zbranch[n];
2264 err = ubifs_add_dirt(c, zbr->lnum,
2276 err = ubifs_add_dirt(c, lnum, len);
2279 err = dbg_check_tnc(c, 0);
2282 mutex_unlock(&c->tnc_mutex);
2287 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2288 * @c: UBIFS file-system description object
2290 * @lnum: LEB number of node
2291 * @offs: node offset
2295 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2296 * may have collisions, like directory entry keys.
2298 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2299 int lnum, int offs, int len, const struct qstr *nm)
2301 int found, n, err = 0;
2302 struct ubifs_znode *znode;
2304 mutex_lock(&c->tnc_mutex);
2305 dbg_tnck(key, "LEB %d:%d, name '%.*s', key ",
2306 lnum, offs, nm->len, nm->name);
2307 found = lookup_level0_dirty(c, key, &znode, &n);
2315 found = fallible_resolve_collision(c, key, &znode, &n,
2318 found = resolve_collision(c, key, &znode, &n, nm);
2319 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2325 /* Ensure the znode is dirtied */
2326 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2327 znode = dirty_cow_bottom_up(c, znode);
2328 if (IS_ERR(znode)) {
2329 err = PTR_ERR(znode);
2335 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2338 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2347 struct ubifs_zbranch zbr;
2353 key_copy(c, key, &zbr.key);
2354 err = tnc_insert(c, znode, &zbr, n + 1);
2359 * We did not find it in the index so there may be a
2360 * dangling branch still in the index. So we remove it
2361 * by passing 'ubifs_tnc_remove_nm()' the same key but
2362 * an unmatchable name.
2364 struct qstr noname = { .name = "" };
2366 err = dbg_check_tnc(c, 0);
2367 mutex_unlock(&c->tnc_mutex);
2370 return ubifs_tnc_remove_nm(c, key, &noname);
2376 err = dbg_check_tnc(c, 0);
2377 mutex_unlock(&c->tnc_mutex);
2382 * tnc_delete - delete a znode form TNC.
2383 * @c: UBIFS file-system description object
2384 * @znode: znode to delete from
2385 * @n: zbranch slot number to delete
2387 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2388 * case of success and a negative error code in case of failure.
2390 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2392 struct ubifs_zbranch *zbr;
2393 struct ubifs_znode *zp;
2396 /* Delete without merge for now */
2397 ubifs_assert(znode->level == 0);
2398 ubifs_assert(n >= 0 && n < c->fanout);
2399 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2401 zbr = &znode->zbranch[n];
2404 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2406 ubifs_dump_znode(c, znode);
2410 /* We do not "gap" zbranch slots */
2411 for (i = n; i < znode->child_cnt - 1; i++)
2412 znode->zbranch[i] = znode->zbranch[i + 1];
2413 znode->child_cnt -= 1;
2415 if (znode->child_cnt > 0)
2419 * This was the last zbranch, we have to delete this znode from the
2424 ubifs_assert(!ubifs_zn_obsolete(znode));
2425 ubifs_assert(ubifs_zn_dirty(znode));
2430 atomic_long_dec(&c->dirty_zn_cnt);
2432 err = insert_old_idx_znode(c, znode);
2437 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2438 atomic_long_inc(&c->clean_zn_cnt);
2439 atomic_long_inc(&ubifs_clean_zn_cnt);
2443 } while (znode->child_cnt == 1); /* while removing last child */
2445 /* Remove from znode, entry n - 1 */
2446 znode->child_cnt -= 1;
2447 ubifs_assert(znode->level != 0);
2448 for (i = n; i < znode->child_cnt; i++) {
2449 znode->zbranch[i] = znode->zbranch[i + 1];
2450 if (znode->zbranch[i].znode)
2451 znode->zbranch[i].znode->iip = i;
2455 * If this is the root and it has only 1 child then
2456 * collapse the tree.
2458 if (!znode->parent) {
2459 while (znode->child_cnt == 1 && znode->level != 0) {
2461 zbr = &znode->zbranch[0];
2462 znode = get_znode(c, znode, 0);
2464 return PTR_ERR(znode);
2465 znode = dirty_cow_znode(c, zbr);
2467 return PTR_ERR(znode);
2468 znode->parent = NULL;
2471 err = insert_old_idx(c, c->zroot.lnum,
2476 c->zroot.lnum = zbr->lnum;
2477 c->zroot.offs = zbr->offs;
2478 c->zroot.len = zbr->len;
2479 c->zroot.znode = znode;
2480 ubifs_assert(!ubifs_zn_obsolete(zp));
2481 ubifs_assert(ubifs_zn_dirty(zp));
2482 atomic_long_dec(&c->dirty_zn_cnt);
2485 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2486 atomic_long_inc(&c->clean_zn_cnt);
2487 atomic_long_inc(&ubifs_clean_zn_cnt);
2497 * ubifs_tnc_remove - remove an index entry of a node.
2498 * @c: UBIFS file-system description object
2501 * Returns %0 on success or negative error code on failure.
2503 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2505 int found, n, err = 0;
2506 struct ubifs_znode *znode;
2508 mutex_lock(&c->tnc_mutex);
2509 dbg_tnck(key, "key ");
2510 found = lookup_level0_dirty(c, key, &znode, &n);
2516 err = tnc_delete(c, znode, n);
2518 err = dbg_check_tnc(c, 0);
2521 mutex_unlock(&c->tnc_mutex);
2526 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2527 * @c: UBIFS file-system description object
2529 * @nm: directory entry name
2531 * Returns %0 on success or negative error code on failure.
2533 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2534 const struct qstr *nm)
2537 struct ubifs_znode *znode;
2539 mutex_lock(&c->tnc_mutex);
2540 dbg_tnck(key, "%.*s, key ", nm->len, nm->name);
2541 err = lookup_level0_dirty(c, key, &znode, &n);
2547 err = fallible_resolve_collision(c, key, &znode, &n,
2550 err = resolve_collision(c, key, &znode, &n, nm);
2551 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2555 /* Ensure the znode is dirtied */
2556 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2557 znode = dirty_cow_bottom_up(c, znode);
2558 if (IS_ERR(znode)) {
2559 err = PTR_ERR(znode);
2563 err = tnc_delete(c, znode, n);
2569 err = dbg_check_tnc(c, 0);
2570 mutex_unlock(&c->tnc_mutex);
2575 * key_in_range - determine if a key falls within a range of keys.
2576 * @c: UBIFS file-system description object
2577 * @key: key to check
2578 * @from_key: lowest key in range
2579 * @to_key: highest key in range
2581 * This function returns %1 if the key is in range and %0 otherwise.
2583 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2584 union ubifs_key *from_key, union ubifs_key *to_key)
2586 if (keys_cmp(c, key, from_key) < 0)
2588 if (keys_cmp(c, key, to_key) > 0)
2594 * ubifs_tnc_remove_range - remove index entries in range.
2595 * @c: UBIFS file-system description object
2596 * @from_key: lowest key to remove
2597 * @to_key: highest key to remove
2599 * This function removes index entries starting at @from_key and ending at
2600 * @to_key. This function returns zero in case of success and a negative error
2601 * code in case of failure.
2603 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2604 union ubifs_key *to_key)
2606 int i, n, k, err = 0;
2607 struct ubifs_znode *znode;
2608 union ubifs_key *key;
2610 mutex_lock(&c->tnc_mutex);
2612 /* Find first level 0 znode that contains keys to remove */
2613 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2620 err = tnc_next(c, &znode, &n);
2621 if (err == -ENOENT) {
2627 key = &znode->zbranch[n].key;
2628 if (!key_in_range(c, key, from_key, to_key)) {
2634 /* Ensure the znode is dirtied */
2635 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2636 znode = dirty_cow_bottom_up(c, znode);
2637 if (IS_ERR(znode)) {
2638 err = PTR_ERR(znode);
2643 /* Remove all keys in range except the first */
2644 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2645 key = &znode->zbranch[i].key;
2646 if (!key_in_range(c, key, from_key, to_key))
2648 lnc_free(&znode->zbranch[i]);
2649 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2650 znode->zbranch[i].len);
2652 ubifs_dump_znode(c, znode);
2655 dbg_tnck(key, "removing key ");
2658 for (i = n + 1 + k; i < znode->child_cnt; i++)
2659 znode->zbranch[i - k] = znode->zbranch[i];
2660 znode->child_cnt -= k;
2663 /* Now delete the first */
2664 err = tnc_delete(c, znode, n);
2671 err = dbg_check_tnc(c, 0);
2672 mutex_unlock(&c->tnc_mutex);
2677 * ubifs_tnc_remove_ino - remove an inode from TNC.
2678 * @c: UBIFS file-system description object
2679 * @inum: inode number to remove
2681 * This function remove inode @inum and all the extended attributes associated
2682 * with the anode from TNC and returns zero in case of success or a negative
2683 * error code in case of failure.
2685 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2687 union ubifs_key key1, key2;
2688 struct ubifs_dent_node *xent, *pxent = NULL;
2689 struct qstr nm = { .name = NULL };
2691 dbg_tnc("ino %lu", (unsigned long)inum);
2694 * Walk all extended attribute entries and remove them together with
2695 * corresponding extended attribute inodes.
2697 lowest_xent_key(c, &key1, inum);
2702 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2704 err = PTR_ERR(xent);
2710 xattr_inum = le64_to_cpu(xent->inum);
2711 dbg_tnc("xent '%s', ino %lu", xent->name,
2712 (unsigned long)xattr_inum);
2714 nm.name = xent->name;
2715 nm.len = le16_to_cpu(xent->nlen);
2716 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2722 lowest_ino_key(c, &key1, xattr_inum);
2723 highest_ino_key(c, &key2, xattr_inum);
2724 err = ubifs_tnc_remove_range(c, &key1, &key2);
2732 key_read(c, &xent->key, &key1);
2736 lowest_ino_key(c, &key1, inum);
2737 highest_ino_key(c, &key2, inum);
2739 return ubifs_tnc_remove_range(c, &key1, &key2);
2743 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2744 * @c: UBIFS file-system description object
2745 * @key: key of last entry
2746 * @nm: name of last entry found or %NULL
2748 * This function finds and reads the next directory or extended attribute entry
2749 * after the given key (@key) if there is one. @nm is used to resolve
2752 * If the name of the current entry is not known and only the key is known,
2753 * @nm->name has to be %NULL. In this case the semantics of this function is a
2754 * little bit different and it returns the entry corresponding to this key, not
2755 * the next one. If the key was not found, the closest "right" entry is
2758 * If the fist entry has to be found, @key has to contain the lowest possible
2759 * key value for this inode and @name has to be %NULL.
2761 * This function returns the found directory or extended attribute entry node
2762 * in case of success, %-ENOENT is returned if no entry was found, and a
2763 * negative error code is returned in case of failure.
2765 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2766 union ubifs_key *key,
2767 const struct qstr *nm)
2769 int n, err, type = key_type(c, key);
2770 struct ubifs_znode *znode;
2771 struct ubifs_dent_node *dent;
2772 struct ubifs_zbranch *zbr;
2773 union ubifs_key *dkey;
2775 dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)");
2776 ubifs_assert(is_hash_key(c, key));
2778 mutex_lock(&c->tnc_mutex);
2779 err = ubifs_lookup_level0(c, key, &znode, &n);
2780 if (unlikely(err < 0))
2785 /* Handle collisions */
2786 err = resolve_collision(c, key, &znode, &n, nm);
2787 dbg_tnc("rc returned %d, znode %p, n %d",
2789 if (unlikely(err < 0))
2793 /* Now find next entry */
2794 err = tnc_next(c, &znode, &n);
2799 * The full name of the entry was not given, in which case the
2800 * behavior of this function is a little different and it
2801 * returns current entry, not the next one.
2805 * However, the given key does not exist in the TNC
2806 * tree and @znode/@n variables contain the closest
2807 * "preceding" element. Switch to the next one.
2809 err = tnc_next(c, &znode, &n);
2815 zbr = &znode->zbranch[n];
2816 dent = kmalloc(zbr->len, GFP_NOFS);
2817 if (unlikely(!dent)) {
2823 * The above 'tnc_next()' call could lead us to the next inode, check
2827 if (key_inum(c, dkey) != key_inum(c, key) ||
2828 key_type(c, dkey) != type) {
2833 err = tnc_read_node_nm(c, zbr, dent);
2837 mutex_unlock(&c->tnc_mutex);
2843 mutex_unlock(&c->tnc_mutex);
2844 return ERR_PTR(err);
2848 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2849 * @c: UBIFS file-system description object
2851 * Destroy left-over obsolete znodes from a failed commit.
2853 static void tnc_destroy_cnext(struct ubifs_info *c)
2855 struct ubifs_znode *cnext;
2859 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2862 struct ubifs_znode *znode = cnext;
2864 cnext = cnext->cnext;
2865 if (ubifs_zn_obsolete(znode))
2867 } while (cnext && cnext != c->cnext);
2871 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2872 * @c: UBIFS file-system description object
2874 void ubifs_tnc_close(struct ubifs_info *c)
2876 tnc_destroy_cnext(c);
2877 if (c->zroot.znode) {
2880 n = atomic_long_read(&c->clean_zn_cnt);
2881 freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
2882 ubifs_assert(freed == n);
2883 atomic_long_sub(n, &ubifs_clean_zn_cnt);
2891 * left_znode - get the znode to the left.
2892 * @c: UBIFS file-system description object
2895 * This function returns a pointer to the znode to the left of @znode or NULL if
2896 * there is not one. A negative error code is returned on failure.
2898 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2899 struct ubifs_znode *znode)
2901 int level = znode->level;
2904 int n = znode->iip - 1;
2906 /* Go up until we can go left */
2907 znode = znode->parent;
2911 /* Now go down the rightmost branch to 'level' */
2912 znode = get_znode(c, znode, n);
2915 while (znode->level != level) {
2916 n = znode->child_cnt - 1;
2917 znode = get_znode(c, znode, n);
2928 * right_znode - get the znode to the right.
2929 * @c: UBIFS file-system description object
2932 * This function returns a pointer to the znode to the right of @znode or NULL
2933 * if there is not one. A negative error code is returned on failure.
2935 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2936 struct ubifs_znode *znode)
2938 int level = znode->level;
2941 int n = znode->iip + 1;
2943 /* Go up until we can go right */
2944 znode = znode->parent;
2947 if (n < znode->child_cnt) {
2948 /* Now go down the leftmost branch to 'level' */
2949 znode = get_znode(c, znode, n);
2952 while (znode->level != level) {
2953 znode = get_znode(c, znode, 0);
2964 * lookup_znode - find a particular indexing node from TNC.
2965 * @c: UBIFS file-system description object
2966 * @key: index node key to lookup
2967 * @level: index node level
2968 * @lnum: index node LEB number
2969 * @offs: index node offset
2971 * This function searches an indexing node by its first key @key and its
2972 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2973 * nodes it traverses to TNC. This function is called for indexing nodes which
2974 * were found on the media by scanning, for example when garbage-collecting or
2975 * when doing in-the-gaps commit. This means that the indexing node which is
2976 * looked for does not have to have exactly the same leftmost key @key, because
2977 * the leftmost key may have been changed, in which case TNC will contain a
2978 * dirty znode which still refers the same @lnum:@offs. This function is clever
2979 * enough to recognize such indexing nodes.
2981 * Note, if a znode was deleted or changed too much, then this function will
2982 * not find it. For situations like this UBIFS has the old index RB-tree
2983 * (indexed by @lnum:@offs).
2985 * This function returns a pointer to the znode found or %NULL if it is not
2986 * found. A negative error code is returned on failure.
2988 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2989 union ubifs_key *key, int level,
2992 struct ubifs_znode *znode, *zn;
2995 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
2998 * The arguments have probably been read off flash, so don't assume
3002 return ERR_PTR(-EINVAL);
3004 /* Get the root znode */
3005 znode = c->zroot.znode;
3007 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3011 /* Check if it is the one we are looking for */
3012 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3014 /* Descend to the parent level i.e. (level + 1) */
3015 if (level >= znode->level)
3018 ubifs_search_zbranch(c, znode, key, &n);
3021 * We reached a znode where the leftmost key is greater
3022 * than the key we are searching for. This is the same
3023 * situation as the one described in a huge comment at
3024 * the end of the 'ubifs_lookup_level0()' function. And
3025 * for exactly the same reasons we have to try to look
3026 * left before giving up.
3028 znode = left_znode(c, znode);
3033 ubifs_search_zbranch(c, znode, key, &n);
3034 ubifs_assert(n >= 0);
3036 if (znode->level == level + 1)
3038 znode = get_znode(c, znode, n);
3042 /* Check if the child is the one we are looking for */
3043 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3044 return get_znode(c, znode, n);
3045 /* If the key is unique, there is nowhere else to look */
3046 if (!is_hash_key(c, key))
3049 * The key is not unique and so may be also in the znodes to either
3056 /* Move one branch to the left */
3060 znode = left_znode(c, znode);
3065 n = znode->child_cnt - 1;
3068 if (znode->zbranch[n].lnum == lnum &&
3069 znode->zbranch[n].offs == offs)
3070 return get_znode(c, znode, n);
3071 /* Stop if the key is less than the one we are looking for */
3072 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3075 /* Back to the middle */
3080 /* Move one branch to the right */
3081 if (++n >= znode->child_cnt) {
3082 znode = right_znode(c, znode);
3090 if (znode->zbranch[n].lnum == lnum &&
3091 znode->zbranch[n].offs == offs)
3092 return get_znode(c, znode, n);
3093 /* Stop if the key is greater than the one we are looking for */
3094 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3101 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3102 * @c: UBIFS file-system description object
3103 * @key: key of index node
3104 * @level: index node level
3105 * @lnum: LEB number of index node
3106 * @offs: offset of index node
3108 * This function returns %0 if the index node is not referred to in the TNC, %1
3109 * if the index node is referred to in the TNC and the corresponding znode is
3110 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3111 * znode is clean, and a negative error code in case of failure.
3113 * Note, the @key argument has to be the key of the first child. Also note,
3114 * this function relies on the fact that 0:0 is never a valid LEB number and
3115 * offset for a main-area node.
3117 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3120 struct ubifs_znode *znode;
3122 znode = lookup_znode(c, key, level, lnum, offs);
3126 return PTR_ERR(znode);
3128 return ubifs_zn_dirty(znode) ? 1 : 2;
3132 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3133 * @c: UBIFS file-system description object
3135 * @lnum: node LEB number
3136 * @offs: node offset
3138 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3139 * not, and a negative error code in case of failure.
3141 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3142 * and offset for a main-area node.
3144 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3147 struct ubifs_zbranch *zbr;
3148 struct ubifs_znode *znode, *zn;
3149 int n, found, err, nn;
3150 const int unique = !is_hash_key(c, key);
3152 found = ubifs_lookup_level0(c, key, &znode, &n);
3154 return found; /* Error code */
3157 zbr = &znode->zbranch[n];
3158 if (lnum == zbr->lnum && offs == zbr->offs)
3159 return 1; /* Found it */
3163 * Because the key is not unique, we have to look left
3170 err = tnc_prev(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 err = tnc_next(c, &znode, &n);
3191 if (keys_cmp(c, key, &znode->zbranch[n].key))
3193 zbr = &znode->zbranch[n];
3194 if (lnum == zbr->lnum && offs == zbr->offs)
3195 return 1; /* Found it */
3201 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3202 * @c: UBIFS file-system description object
3204 * @level: index node level (if it is an index node)
3205 * @lnum: node LEB number
3206 * @offs: node offset
3207 * @is_idx: non-zero if the node is an index node
3209 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3210 * negative error code in case of failure. For index nodes, @key has to be the
3211 * key of the first child. An index node is considered to be in the TNC only if
3212 * the corresponding znode is clean or has not been loaded.
3214 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3215 int lnum, int offs, int is_idx)
3219 mutex_lock(&c->tnc_mutex);
3221 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3225 /* The index node was found but it was dirty */
3228 /* The index node was found and it was clean */
3233 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3236 mutex_unlock(&c->tnc_mutex);
3241 * ubifs_dirty_idx_node - dirty an index node.
3242 * @c: UBIFS file-system description object
3243 * @key: index node key
3244 * @level: index node level
3245 * @lnum: index node LEB number
3246 * @offs: index node offset
3248 * This function loads and dirties an index node so that it can be garbage
3249 * collected. The @key argument has to be the key of the first child. This
3250 * function relies on the fact that 0:0 is never a valid LEB number and offset
3251 * for a main-area node. Returns %0 on success and a negative error code on
3254 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3257 struct ubifs_znode *znode;
3260 mutex_lock(&c->tnc_mutex);
3261 znode = lookup_znode(c, key, level, lnum, offs);
3264 if (IS_ERR(znode)) {
3265 err = PTR_ERR(znode);
3268 znode = dirty_cow_bottom_up(c, znode);
3269 if (IS_ERR(znode)) {
3270 err = PTR_ERR(znode);
3275 mutex_unlock(&c->tnc_mutex);
3280 * dbg_check_inode_size - check if inode size is correct.
3281 * @c: UBIFS file-system description object
3282 * @inum: inode number
3285 * This function makes sure that the inode size (@size) is correct and it does
3286 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3287 * if it has a data page beyond @size, and other negative error code in case of
3290 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3294 union ubifs_key from_key, to_key, *key;
3295 struct ubifs_znode *znode;
3298 if (!S_ISREG(inode->i_mode))
3300 if (!dbg_is_chk_gen(c))
3303 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3304 data_key_init(c, &from_key, inode->i_ino, block);
3305 highest_data_key(c, &to_key, inode->i_ino);
3307 mutex_lock(&c->tnc_mutex);
3308 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3318 err = tnc_next(c, &znode, &n);
3319 if (err == -ENOENT) {
3326 ubifs_assert(err == 0);
3327 key = &znode->zbranch[n].key;
3328 if (!key_in_range(c, key, &from_key, &to_key))
3332 block = key_block(c, key);
3333 ubifs_err("inode %lu has size %lld, but there are data at offset %lld",
3334 (unsigned long)inode->i_ino, size,
3335 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3336 mutex_unlock(&c->tnc_mutex);
3337 ubifs_dump_inode(c, inode);
3342 mutex_unlock(&c->tnc_mutex);