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>
37 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
38 struct ubifs_zbranch *zbr);
39 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
40 struct ubifs_zbranch *zbr, void *node);
43 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
44 * @NAME_LESS: name corresponding to the first argument is less than second
45 * @NAME_MATCHES: names match
46 * @NAME_GREATER: name corresponding to the second argument is greater than
48 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
50 * These constants were introduce to improve readability.
60 * insert_old_idx - record an index node obsoleted since the last commit start.
61 * @c: UBIFS file-system description object
62 * @lnum: LEB number of obsoleted index node
63 * @offs: offset of obsoleted index node
65 * Returns %0 on success, and a negative error code on failure.
67 * For recovery, there must always be a complete intact version of the index on
68 * flash at all times. That is called the "old index". It is the index as at the
69 * time of the last successful commit. Many of the index nodes in the old index
70 * may be dirty, but they must not be erased until the next successful commit
71 * (at which point that index becomes the old index).
73 * That means that the garbage collection and the in-the-gaps method of
74 * committing must be able to determine if an index node is in the old index.
75 * Most of the old index nodes can be found by looking up the TNC using the
76 * 'lookup_znode()' function. However, some of the old index nodes may have
77 * been deleted from the current index or may have been changed so much that
78 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
79 * That is what this function does. The RB-tree is ordered by LEB number and
80 * offset because they uniquely identify the old index node.
82 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
84 struct ubifs_old_idx *old_idx, *o;
85 struct rb_node **p, *parent = NULL;
87 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
88 if (unlikely(!old_idx))
93 p = &c->old_idx.rb_node;
96 o = rb_entry(parent, struct ubifs_old_idx, rb);
99 else if (lnum > o->lnum)
101 else if (offs < o->offs)
103 else if (offs > o->offs)
106 ubifs_err(c, "old idx added twice!");
111 rb_link_node(&old_idx->rb, parent, p);
112 rb_insert_color(&old_idx->rb, &c->old_idx);
117 * insert_old_idx_znode - record a znode obsoleted since last commit start.
118 * @c: UBIFS file-system description object
119 * @znode: znode of obsoleted index node
121 * Returns %0 on success, and a negative error code on failure.
123 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
126 struct ubifs_zbranch *zbr;
128 zbr = &znode->parent->zbranch[znode->iip];
130 return insert_old_idx(c, zbr->lnum, zbr->offs);
133 return insert_old_idx(c, c->zroot.lnum,
139 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
140 * @c: UBIFS file-system description object
141 * @znode: znode of obsoleted index node
143 * Returns %0 on success, and a negative error code on failure.
145 static int ins_clr_old_idx_znode(struct ubifs_info *c,
146 struct ubifs_znode *znode)
151 struct ubifs_zbranch *zbr;
153 zbr = &znode->parent->zbranch[znode->iip];
155 err = insert_old_idx(c, zbr->lnum, zbr->offs);
164 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
175 * destroy_old_idx - destroy the old_idx RB-tree.
176 * @c: UBIFS file-system description object
178 * During start commit, the old_idx RB-tree is used to avoid overwriting index
179 * nodes that were in the index last commit but have since been deleted. This
180 * is necessary for recovery i.e. the old index must be kept intact until the
181 * new index is successfully written. The old-idx RB-tree is used for the
182 * in-the-gaps method of writing index nodes and is destroyed every commit.
184 void destroy_old_idx(struct ubifs_info *c)
186 struct ubifs_old_idx *old_idx, *n;
188 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
191 c->old_idx = RB_ROOT;
195 * copy_znode - copy a dirty znode.
196 * @c: UBIFS file-system description object
197 * @znode: znode to copy
199 * A dirty znode being committed may not be changed, so it is copied.
201 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
202 struct ubifs_znode *znode)
204 struct ubifs_znode *zn;
206 zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS);
208 return ERR_PTR(-ENOMEM);
211 __set_bit(DIRTY_ZNODE, &zn->flags);
212 __clear_bit(COW_ZNODE, &zn->flags);
214 ubifs_assert(c, !ubifs_zn_obsolete(znode));
215 __set_bit(OBSOLETE_ZNODE, &znode->flags);
217 if (znode->level != 0) {
219 const int n = zn->child_cnt;
221 /* The children now have new parent */
222 for (i = 0; i < n; i++) {
223 struct ubifs_zbranch *zbr = &zn->zbranch[i];
226 zbr->znode->parent = zn;
230 atomic_long_inc(&c->dirty_zn_cnt);
235 * add_idx_dirt - add dirt due to a dirty znode.
236 * @c: UBIFS file-system description object
237 * @lnum: LEB number of index node
238 * @dirt: size of index node
240 * This function updates lprops dirty space and the new size of the index.
242 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
244 c->calc_idx_sz -= ALIGN(dirt, 8);
245 return ubifs_add_dirt(c, lnum, dirt);
249 * dirty_cow_znode - ensure a znode is not being committed.
250 * @c: UBIFS file-system description object
251 * @zbr: branch of znode to check
253 * Returns dirtied znode on success or negative error code on failure.
255 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
256 struct ubifs_zbranch *zbr)
258 struct ubifs_znode *znode = zbr->znode;
259 struct ubifs_znode *zn;
262 if (!ubifs_zn_cow(znode)) {
263 /* znode is not being committed */
264 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
265 atomic_long_inc(&c->dirty_zn_cnt);
266 atomic_long_dec(&c->clean_zn_cnt);
267 atomic_long_dec(&ubifs_clean_zn_cnt);
268 err = add_idx_dirt(c, zbr->lnum, zbr->len);
275 zn = copy_znode(c, znode);
280 err = insert_old_idx(c, zbr->lnum, zbr->offs);
283 err = add_idx_dirt(c, zbr->lnum, zbr->len);
298 * lnc_add - add a leaf node to the leaf node cache.
299 * @c: UBIFS file-system description object
300 * @zbr: zbranch of leaf node
303 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
304 * purpose of the leaf node cache is to save re-reading the same leaf node over
305 * and over again. Most things are cached by VFS, however the file system must
306 * cache directory entries for readdir and for resolving hash collisions. The
307 * present implementation of the leaf node cache is extremely simple, and
308 * allows for error returns that are not used but that may be needed if a more
309 * complex implementation is created.
311 * Note, this function does not add the @node object to LNC directly, but
312 * allocates a copy of the object and adds the copy to LNC. The reason for this
313 * is that @node has been allocated outside of the TNC subsystem and will be
314 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
315 * may be changed at any time, e.g. freed by the shrinker.
317 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
322 const struct ubifs_dent_node *dent = node;
324 ubifs_assert(c, !zbr->leaf);
325 ubifs_assert(c, zbr->len != 0);
326 ubifs_assert(c, is_hash_key(c, &zbr->key));
328 err = ubifs_validate_entry(c, dent);
331 ubifs_dump_node(c, dent);
335 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
337 /* We don't have to have the cache, so no error */
340 zbr->leaf = lnc_node;
345 * lnc_add_directly - add a leaf node to the leaf-node-cache.
346 * @c: UBIFS file-system description object
347 * @zbr: zbranch of leaf node
350 * This function is similar to 'lnc_add()', but it does not create a copy of
351 * @node but inserts @node to TNC directly.
353 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
358 ubifs_assert(c, !zbr->leaf);
359 ubifs_assert(c, zbr->len != 0);
361 err = ubifs_validate_entry(c, node);
364 ubifs_dump_node(c, node);
373 * lnc_free - remove a leaf node from the leaf node cache.
374 * @zbr: zbranch of leaf node
377 static void lnc_free(struct ubifs_zbranch *zbr)
386 * tnc_read_hashed_node - read a "hashed" leaf node.
387 * @c: UBIFS file-system description object
388 * @zbr: key and position of the node
389 * @node: node is returned here
391 * This function reads a "hashed" node defined by @zbr from the leaf node cache
392 * (in it is there) or from the hash media, in which case the node is also
393 * added to LNC. Returns zero in case of success or a negative negative error
394 * code in case of failure.
396 static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr,
401 ubifs_assert(c, is_hash_key(c, &zbr->key));
404 /* Read from the leaf node cache */
405 ubifs_assert(c, zbr->len != 0);
406 memcpy(node, zbr->leaf, zbr->len);
411 err = fallible_read_node(c, &zbr->key, zbr, node);
413 * When the node was not found, return -ENOENT, 0 otherwise.
414 * Negative return codes stay as-is.
421 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 * @zbr: the zbranch describing the node to read
438 * This function tries to read a node of known type and length, checks it and
439 * stores it in @buf. This function returns %1 if a node is present and %0 if
440 * a node is not present. A negative error code is returned for I/O errors.
441 * This function performs that same function as ubifs_read_node except that
442 * it does not require that there is actually a node present and instead
443 * the return code indicates if a node was read.
445 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
446 * is true (it is controlled by corresponding mount option). However, if
447 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
448 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
449 * because during mounting or re-mounting from R/O mode to R/W mode we may read
450 * journal nodes (when replying the journal or doing the recovery) and the
451 * journal nodes may potentially be corrupted, so checking is required.
453 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
454 struct ubifs_zbranch *zbr)
457 int lnum = zbr->lnum;
458 int offs = zbr->offs;
460 struct ubifs_ch *ch = buf;
461 uint32_t crc, node_crc;
463 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
465 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
467 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
468 type, lnum, offs, err);
472 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
475 if (ch->node_type != type)
478 node_len = le32_to_cpu(ch->len);
482 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
486 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
487 node_crc = le32_to_cpu(ch->crc);
491 err = ubifs_node_check_hash(c, buf, zbr->hash);
493 ubifs_bad_hash(c, buf, zbr->hash, lnum, offs);
501 * fallible_read_node - try to read a leaf node.
502 * @c: UBIFS file-system description object
503 * @key: key of node to read
504 * @zbr: position of node
505 * @node: node returned
507 * This function tries to read a node and returns %1 if the node is read, %0
508 * if the node is not present, and a negative error code in the case of error.
510 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
511 struct ubifs_zbranch *zbr, void *node)
515 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
517 ret = try_read_node(c, node, key_type(c, key), zbr);
519 union ubifs_key node_key;
520 struct ubifs_dent_node *dent = node;
522 /* All nodes have key in the same place */
523 key_read(c, &dent->key, &node_key);
524 if (keys_cmp(c, key, &node_key) != 0)
527 if (ret == 0 && c->replaying)
528 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
529 zbr->lnum, zbr->offs, zbr->len);
534 * matches_name - determine if a direntry or xattr entry matches a given name.
535 * @c: UBIFS file-system description object
536 * @zbr: zbranch of dent
539 * This function checks if xentry/direntry referred by zbranch @zbr matches name
540 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
541 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
542 * of failure, a negative error code is returned.
544 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
545 const struct fscrypt_name *nm)
547 struct ubifs_dent_node *dent;
550 /* If possible, match against the dent in the leaf node cache */
552 dent = kmalloc(zbr->len, GFP_NOFS);
556 err = ubifs_tnc_read_node(c, zbr, dent);
560 /* Add the node to the leaf node cache */
561 err = lnc_add_directly(c, zbr, dent);
567 nlen = le16_to_cpu(dent->nlen);
568 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
570 if (nlen == fname_len(nm))
572 else if (nlen < fname_len(nm))
587 * get_znode - get a TNC znode that may not be loaded yet.
588 * @c: UBIFS file-system description object
589 * @znode: parent znode
590 * @n: znode branch slot number
592 * This function returns the znode or a negative error code.
594 static struct ubifs_znode *get_znode(struct ubifs_info *c,
595 struct ubifs_znode *znode, int n)
597 struct ubifs_zbranch *zbr;
599 zbr = &znode->zbranch[n];
603 znode = ubifs_load_znode(c, zbr, znode, n);
608 * tnc_next - find next TNC entry.
609 * @c: UBIFS file-system description object
610 * @zn: znode is passed and returned here
611 * @n: znode branch slot number is passed and returned here
613 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
614 * no next entry, or a negative error code otherwise.
616 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
618 struct ubifs_znode *znode = *zn;
622 if (nn < znode->child_cnt) {
627 struct ubifs_znode *zp;
634 if (nn < znode->child_cnt) {
635 znode = get_znode(c, znode, nn);
637 return PTR_ERR(znode);
638 while (znode->level != 0) {
639 znode = get_znode(c, znode, 0);
641 return PTR_ERR(znode);
653 * tnc_prev - find previous TNC entry.
654 * @c: UBIFS file-system description object
655 * @zn: znode is returned here
656 * @n: znode branch slot number is passed and returned here
658 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
659 * there is no next entry, or a negative error code otherwise.
661 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
663 struct ubifs_znode *znode = *zn;
671 struct ubifs_znode *zp;
679 znode = get_znode(c, znode, nn);
681 return PTR_ERR(znode);
682 while (znode->level != 0) {
683 nn = znode->child_cnt - 1;
684 znode = get_znode(c, znode, nn);
686 return PTR_ERR(znode);
688 nn = znode->child_cnt - 1;
698 * resolve_collision - resolve a collision.
699 * @c: UBIFS file-system description object
700 * @key: key of a directory or extended attribute entry
701 * @zn: znode is returned here
702 * @n: zbranch number is passed and returned here
703 * @nm: name of the entry
705 * This function is called for "hashed" keys to make sure that the found key
706 * really corresponds to the looked up node (directory or extended attribute
707 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
708 * %0 is returned if @nm is not found and @zn and @n are set to the previous
709 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
710 * This means that @n may be set to %-1 if the leftmost key in @zn is the
711 * previous one. A negative error code is returned on failures.
713 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
714 struct ubifs_znode **zn, int *n,
715 const struct fscrypt_name *nm)
719 err = matches_name(c, &(*zn)->zbranch[*n], nm);
720 if (unlikely(err < 0))
722 if (err == NAME_MATCHES)
725 if (err == NAME_GREATER) {
728 err = tnc_prev(c, zn, n);
729 if (err == -ENOENT) {
730 ubifs_assert(c, *n == 0);
736 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
738 * We have found the branch after which we would
739 * like to insert, but inserting in this znode
740 * may still be wrong. Consider the following 3
741 * znodes, in the case where we are resolving a
742 * collision with Key2.
745 * ----------------------
746 * level 1 | Key0 | Key1 |
747 * -----------------------
749 * znode za | | znode zb
750 * ------------ ------------
751 * level 0 | Key0 | | Key2 |
752 * ------------ ------------
754 * The lookup finds Key2 in znode zb. Lets say
755 * there is no match and the name is greater so
756 * we look left. When we find Key0, we end up
757 * here. If we return now, we will insert into
758 * znode za at slot n = 1. But that is invalid
759 * according to the parent's keys. Key2 must
760 * be inserted into znode zb.
762 * Note, this problem is not relevant for the
763 * case when we go right, because
764 * 'tnc_insert()' would correct the parent key.
766 if (*n == (*zn)->child_cnt - 1) {
767 err = tnc_next(c, zn, n);
769 /* Should be impossible */
775 ubifs_assert(c, *n == 0);
780 err = matches_name(c, &(*zn)->zbranch[*n], nm);
783 if (err == NAME_LESS)
785 if (err == NAME_MATCHES)
787 ubifs_assert(c, err == NAME_GREATER);
791 struct ubifs_znode *znode = *zn;
795 err = tnc_next(c, &znode, &nn);
800 if (keys_cmp(c, &znode->zbranch[nn].key, key))
802 err = matches_name(c, &znode->zbranch[nn], nm);
805 if (err == NAME_GREATER)
809 if (err == NAME_MATCHES)
811 ubifs_assert(c, err == NAME_LESS);
817 * fallible_matches_name - determine if a dent matches a given name.
818 * @c: UBIFS file-system description object
819 * @zbr: zbranch of dent
822 * This is a "fallible" version of 'matches_name()' function which does not
823 * panic if the direntry/xentry referred by @zbr does not exist on the media.
825 * This function checks if xentry/direntry referred by zbranch @zbr matches name
826 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
827 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
828 * if xentry/direntry referred by @zbr does not exist on the media. A negative
829 * error code is returned in case of failure.
831 static int fallible_matches_name(struct ubifs_info *c,
832 struct ubifs_zbranch *zbr,
833 const struct fscrypt_name *nm)
835 struct ubifs_dent_node *dent;
838 /* If possible, match against the dent in the leaf node cache */
840 dent = kmalloc(zbr->len, GFP_NOFS);
844 err = fallible_read_node(c, &zbr->key, zbr, dent);
848 /* The node was not present */
852 ubifs_assert(c, err == 1);
854 err = lnc_add_directly(c, zbr, dent);
860 nlen = le16_to_cpu(dent->nlen);
861 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
863 if (nlen == fname_len(nm))
865 else if (nlen < fname_len(nm))
880 * fallible_resolve_collision - resolve a collision even if nodes are missing.
881 * @c: UBIFS file-system description object
883 * @zn: znode is returned here
884 * @n: branch number is passed and returned here
885 * @nm: name of directory entry
886 * @adding: indicates caller is adding a key to the TNC
888 * This is a "fallible" version of the 'resolve_collision()' function which
889 * does not panic if one of the nodes referred to by TNC does not exist on the
890 * media. This may happen when replaying the journal if a deleted node was
891 * Garbage-collected and the commit was not done. A branch that refers to a node
892 * that is not present is called a dangling branch. The following are the return
893 * codes for this function:
894 * o if @nm was found, %1 is returned and @zn and @n are set to the found
896 * o if we are @adding and @nm was not found, %0 is returned;
897 * o if we are not @adding and @nm was not found, but a dangling branch was
898 * found, then %1 is returned and @zn and @n are set to the dangling branch;
899 * o a negative error code is returned in case of failure.
901 static int fallible_resolve_collision(struct ubifs_info *c,
902 const union ubifs_key *key,
903 struct ubifs_znode **zn, int *n,
904 const struct fscrypt_name *nm,
907 struct ubifs_znode *o_znode = NULL, *znode = *zn;
908 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
910 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
911 if (unlikely(cmp < 0))
913 if (cmp == NAME_MATCHES)
915 if (cmp == NOT_ON_MEDIA) {
919 * We are unlucky and hit a dangling branch straight away.
920 * Now we do not really know where to go to find the needed
921 * branch - to the left or to the right. Well, let's try left.
925 unsure = 1; /* Remove a dangling branch wherever it is */
927 if (cmp == NAME_GREATER || unsure) {
930 err = tnc_prev(c, zn, n);
931 if (err == -ENOENT) {
932 ubifs_assert(c, *n == 0);
938 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
939 /* See comments in 'resolve_collision()' */
940 if (*n == (*zn)->child_cnt - 1) {
941 err = tnc_next(c, zn, n);
943 /* Should be impossible */
949 ubifs_assert(c, *n == 0);
954 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
957 if (err == NAME_MATCHES)
959 if (err == NOT_ON_MEDIA) {
966 if (err == NAME_LESS)
973 if (cmp == NAME_LESS || unsure) {
978 err = tnc_next(c, &znode, &nn);
983 if (keys_cmp(c, &znode->zbranch[nn].key, key))
985 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
988 if (err == NAME_GREATER)
992 if (err == NAME_MATCHES)
994 if (err == NOT_ON_MEDIA) {
1001 /* Never match a dangling branch when adding */
1002 if (adding || !o_znode)
1005 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
1006 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1007 o_znode->zbranch[o_n].len);
1014 * matches_position - determine if a zbranch matches a given position.
1015 * @zbr: zbranch of dent
1016 * @lnum: LEB number of dent to match
1017 * @offs: offset of dent to match
1019 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1021 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1023 if (zbr->lnum == lnum && zbr->offs == offs)
1030 * resolve_collision_directly - resolve a collision directly.
1031 * @c: UBIFS file-system description object
1032 * @key: key of directory entry
1033 * @zn: znode is passed and returned here
1034 * @n: zbranch number is passed and returned here
1035 * @lnum: LEB number of dent node to match
1036 * @offs: offset of dent node to match
1038 * This function is used for "hashed" keys to make sure the found directory or
1039 * extended attribute entry node is what was looked for. It is used when the
1040 * flash address of the right node is known (@lnum:@offs) which makes it much
1041 * easier to resolve collisions (no need to read entries and match full
1042 * names). This function returns %1 and sets @zn and @n if the collision is
1043 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1044 * previous directory entry. Otherwise a negative error code is returned.
1046 static int resolve_collision_directly(struct ubifs_info *c,
1047 const union ubifs_key *key,
1048 struct ubifs_znode **zn, int *n,
1051 struct ubifs_znode *znode;
1056 if (matches_position(&znode->zbranch[nn], lnum, offs))
1061 err = tnc_prev(c, &znode, &nn);
1066 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1068 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1079 err = tnc_next(c, &znode, &nn);
1084 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1088 if (matches_position(&znode->zbranch[nn], lnum, offs))
1094 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1095 * @c: UBIFS file-system description object
1096 * @znode: znode to dirty
1098 * If we do not have a unique key that resides in a znode, then we cannot
1099 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1100 * This function records the path back to the last dirty ancestor, and then
1101 * dirties the znodes on that path.
1103 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1104 struct ubifs_znode *znode)
1106 struct ubifs_znode *zp;
1107 int *path = c->bottom_up_buf, p = 0;
1109 ubifs_assert(c, c->zroot.znode);
1110 ubifs_assert(c, znode);
1111 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1112 kfree(c->bottom_up_buf);
1113 c->bottom_up_buf = kmalloc_array(c->zroot.znode->level,
1116 if (!c->bottom_up_buf)
1117 return ERR_PTR(-ENOMEM);
1118 path = c->bottom_up_buf;
1120 if (c->zroot.znode->level) {
1121 /* Go up until parent is dirty */
1129 ubifs_assert(c, p < c->zroot.znode->level);
1131 if (!zp->cnext && ubifs_zn_dirty(znode))
1137 /* Come back down, dirtying as we go */
1139 struct ubifs_zbranch *zbr;
1143 ubifs_assert(c, path[p - 1] >= 0);
1144 ubifs_assert(c, path[p - 1] < zp->child_cnt);
1145 zbr = &zp->zbranch[path[--p]];
1146 znode = dirty_cow_znode(c, zbr);
1148 ubifs_assert(c, znode == c->zroot.znode);
1149 znode = dirty_cow_znode(c, &c->zroot);
1151 if (IS_ERR(znode) || !p)
1153 ubifs_assert(c, path[p - 1] >= 0);
1154 ubifs_assert(c, path[p - 1] < znode->child_cnt);
1155 znode = znode->zbranch[path[p - 1]].znode;
1162 * ubifs_lookup_level0 - search for zero-level znode.
1163 * @c: UBIFS file-system description object
1164 * @key: key to lookup
1165 * @zn: znode is returned here
1166 * @n: znode branch slot number is returned here
1168 * This function looks up the TNC tree and search for zero-level znode which
1169 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1171 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1172 * is returned and slot number of the matched branch is stored in @n;
1173 * o not exact match, which means that zero-level znode does not contain
1174 * @key, then %0 is returned and slot number of the closest branch is stored
1176 * o @key is so small that it is even less than the lowest key of the
1177 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1179 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1180 * function reads corresponding indexing nodes and inserts them to TNC. In
1181 * case of failure, a negative error code is returned.
1183 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1184 struct ubifs_znode **zn, int *n)
1187 struct ubifs_znode *znode;
1188 time64_t time = ktime_get_seconds();
1190 dbg_tnck(key, "search key ");
1191 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
1193 znode = c->zroot.znode;
1194 if (unlikely(!znode)) {
1195 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1197 return PTR_ERR(znode);
1203 struct ubifs_zbranch *zbr;
1205 exact = ubifs_search_zbranch(c, znode, key, n);
1207 if (znode->level == 0)
1212 zbr = &znode->zbranch[*n];
1220 /* znode is not in TNC cache, load it from the media */
1221 znode = ubifs_load_znode(c, zbr, znode, *n);
1223 return PTR_ERR(znode);
1227 if (exact || !is_hash_key(c, key) || *n != -1) {
1228 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1233 * Here is a tricky place. We have not found the key and this is a
1234 * "hashed" key, which may collide. The rest of the code deals with
1235 * situations like this:
1239 * | 3 | 5 | | 6 | 7 | (x)
1241 * Or more a complex example:
1245 * | 1 | 3 | | 5 | 8 |
1247 * | 5 | 5 | | 6 | 7 | (x)
1249 * In the examples, if we are looking for key "5", we may reach nodes
1250 * marked with "(x)". In this case what we have do is to look at the
1251 * left and see if there is "5" key there. If there is, we have to
1254 * Note, this whole situation is possible because we allow to have
1255 * elements which are equivalent to the next key in the parent in the
1256 * children of current znode. For example, this happens if we split a
1257 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1261 * | 3 | 5 | | 5 | 6 | 7 |
1263 * And this becomes what is at the first "picture" after key "5" marked
1264 * with "^" is removed. What could be done is we could prohibit
1265 * splitting in the middle of the colliding sequence. Also, when
1266 * removing the leftmost key, we would have to correct the key of the
1267 * parent node, which would introduce additional complications. Namely,
1268 * if we changed the leftmost key of the parent znode, the garbage
1269 * collector would be unable to find it (GC is doing this when GC'ing
1270 * indexing LEBs). Although we already have an additional RB-tree where
1271 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1272 * after the commit. But anyway, this does not look easy to implement
1273 * so we did not try this.
1275 err = tnc_prev(c, &znode, n);
1276 if (err == -ENOENT) {
1277 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1281 if (unlikely(err < 0))
1283 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1284 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1289 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1295 * lookup_level0_dirty - search for zero-level znode dirtying.
1296 * @c: UBIFS file-system description object
1297 * @key: key to lookup
1298 * @zn: znode is returned here
1299 * @n: znode branch slot number is returned here
1301 * This function looks up the TNC tree and search for zero-level znode which
1302 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1304 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1305 * is returned and slot number of the matched branch is stored in @n;
1306 * o not exact match, which means that zero-level znode does not contain @key
1307 * then %0 is returned and slot number of the closed branch is stored in
1309 * o @key is so small that it is even less than the lowest key of the
1310 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1312 * Additionally all znodes in the path from the root to the located zero-level
1313 * znode are marked as dirty.
1315 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1316 * function reads corresponding indexing nodes and inserts them to TNC. In
1317 * case of failure, a negative error code is returned.
1319 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1320 struct ubifs_znode **zn, int *n)
1323 struct ubifs_znode *znode;
1324 time64_t time = ktime_get_seconds();
1326 dbg_tnck(key, "search and dirty key ");
1328 znode = c->zroot.znode;
1329 if (unlikely(!znode)) {
1330 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1332 return PTR_ERR(znode);
1335 znode = dirty_cow_znode(c, &c->zroot);
1337 return PTR_ERR(znode);
1342 struct ubifs_zbranch *zbr;
1344 exact = ubifs_search_zbranch(c, znode, key, n);
1346 if (znode->level == 0)
1351 zbr = &znode->zbranch[*n];
1355 znode = dirty_cow_znode(c, zbr);
1357 return PTR_ERR(znode);
1361 /* znode is not in TNC cache, load it from the media */
1362 znode = ubifs_load_znode(c, zbr, znode, *n);
1364 return PTR_ERR(znode);
1365 znode = dirty_cow_znode(c, zbr);
1367 return PTR_ERR(znode);
1371 if (exact || !is_hash_key(c, key) || *n != -1) {
1372 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1377 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1380 err = tnc_prev(c, &znode, n);
1381 if (err == -ENOENT) {
1383 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1386 if (unlikely(err < 0))
1388 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1390 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1394 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1395 znode = dirty_cow_bottom_up(c, znode);
1397 return PTR_ERR(znode);
1400 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1406 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1407 * @c: UBIFS file-system description object
1409 * @gc_seq1: garbage collection sequence number
1411 * This function determines if @lnum may have been garbage collected since
1412 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1415 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1417 int gc_seq2, gced_lnum;
1419 gced_lnum = c->gced_lnum;
1421 gc_seq2 = c->gc_seq;
1422 /* Same seq means no GC */
1423 if (gc_seq1 == gc_seq2)
1425 /* Different by more than 1 means we don't know */
1426 if (gc_seq1 + 1 != gc_seq2)
1429 * We have seen the sequence number has increased by 1. Now we need to
1430 * be sure we read the right LEB number, so read it again.
1433 if (gced_lnum != c->gced_lnum)
1435 /* Finally we can check lnum */
1436 if (gced_lnum == lnum)
1442 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1443 * @c: UBIFS file-system description object
1444 * @key: node key to lookup
1445 * @node: the node is returned here
1446 * @lnum: LEB number is returned here
1447 * @offs: offset is returned here
1449 * This function looks up and reads node with key @key. The caller has to make
1450 * sure the @node buffer is large enough to fit the node. Returns zero in case
1451 * of success, %-ENOENT if the node was not found, and a negative error code in
1452 * case of failure. The node location can be returned in @lnum and @offs.
1454 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1455 void *node, int *lnum, int *offs)
1457 int found, n, err, safely = 0, gc_seq1;
1458 struct ubifs_znode *znode;
1459 struct ubifs_zbranch zbr, *zt;
1462 mutex_lock(&c->tnc_mutex);
1463 found = ubifs_lookup_level0(c, key, &znode, &n);
1467 } else if (found < 0) {
1471 zt = &znode->zbranch[n];
1476 if (is_hash_key(c, key)) {
1478 * In this case the leaf node cache gets used, so we pass the
1479 * address of the zbranch and keep the mutex locked
1481 err = tnc_read_hashed_node(c, zt, node);
1485 err = ubifs_tnc_read_node(c, zt, node);
1488 /* Drop the TNC mutex prematurely and race with garbage collection */
1489 zbr = znode->zbranch[n];
1490 gc_seq1 = c->gc_seq;
1491 mutex_unlock(&c->tnc_mutex);
1493 if (ubifs_get_wbuf(c, zbr.lnum)) {
1494 /* We do not GC journal heads */
1495 err = ubifs_tnc_read_node(c, &zbr, node);
1499 err = fallible_read_node(c, key, &zbr, node);
1500 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1502 * The node may have been GC'ed out from under us so try again
1503 * while keeping the TNC mutex locked.
1511 mutex_unlock(&c->tnc_mutex);
1516 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1517 * @c: UBIFS file-system description object
1518 * @bu: bulk-read parameters and results
1520 * Lookup consecutive data node keys for the same inode that reside
1521 * consecutively in the same LEB. This function returns zero in case of success
1522 * and a negative error code in case of failure.
1524 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1525 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1526 * maximum possible amount of nodes for bulk-read.
1528 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1530 int n, err = 0, lnum = -1, uninitialized_var(offs);
1531 int uninitialized_var(len);
1532 unsigned int block = key_block(c, &bu->key);
1533 struct ubifs_znode *znode;
1539 mutex_lock(&c->tnc_mutex);
1540 /* Find first key */
1541 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1546 len = znode->zbranch[n].len;
1547 /* The buffer must be big enough for at least 1 node */
1548 if (len > bu->buf_len) {
1553 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1555 lnum = znode->zbranch[n].lnum;
1556 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1559 struct ubifs_zbranch *zbr;
1560 union ubifs_key *key;
1561 unsigned int next_block;
1564 err = tnc_next(c, &znode, &n);
1567 zbr = &znode->zbranch[n];
1569 /* See if there is another data key for this file */
1570 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1571 key_type(c, key) != UBIFS_DATA_KEY) {
1576 /* First key found */
1578 offs = ALIGN(zbr->offs + zbr->len, 8);
1580 if (len > bu->buf_len) {
1586 * The data nodes must be in consecutive positions in
1589 if (zbr->lnum != lnum || zbr->offs != offs)
1591 offs += ALIGN(zbr->len, 8);
1592 len = ALIGN(len, 8) + zbr->len;
1593 /* Must not exceed buffer length */
1594 if (len > bu->buf_len)
1597 /* Allow for holes */
1598 next_block = key_block(c, key);
1599 bu->blk_cnt += (next_block - block - 1);
1600 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1604 bu->zbranch[bu->cnt++] = *zbr;
1606 /* See if we have room for more */
1607 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1609 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1613 if (err == -ENOENT) {
1617 bu->gc_seq = c->gc_seq;
1618 mutex_unlock(&c->tnc_mutex);
1622 * An enormous hole could cause bulk-read to encompass too many
1623 * page cache pages, so limit the number here.
1625 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1626 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1628 * Ensure that bulk-read covers a whole number of page cache
1631 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1632 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1635 /* At the end of file we can round up */
1636 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1639 /* Exclude data nodes that do not make up a whole page cache page */
1640 block = key_block(c, &bu->key) + bu->blk_cnt;
1641 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1643 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1651 * read_wbuf - bulk-read from a LEB with a wbuf.
1652 * @wbuf: wbuf that may overlap the read
1653 * @buf: buffer into which to read
1655 * @lnum: LEB number from which to read
1656 * @offs: offset from which to read
1658 * This functions returns %0 on success or a negative error code on failure.
1660 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1663 const struct ubifs_info *c = wbuf->c;
1666 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1667 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1668 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1669 ubifs_assert(c, offs + len <= c->leb_size);
1671 spin_lock(&wbuf->lock);
1672 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1674 /* We may safely unlock the write-buffer and read the data */
1675 spin_unlock(&wbuf->lock);
1676 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1679 /* Don't read under wbuf */
1680 rlen = wbuf->offs - offs;
1684 /* Copy the rest from the write-buffer */
1685 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1686 spin_unlock(&wbuf->lock);
1689 /* Read everything that goes before write-buffer */
1690 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1696 * validate_data_node - validate data nodes for bulk-read.
1697 * @c: UBIFS file-system description object
1698 * @buf: buffer containing data node to validate
1699 * @zbr: zbranch of data node to validate
1701 * This functions returns %0 on success or a negative error code on failure.
1703 static int validate_data_node(struct ubifs_info *c, void *buf,
1704 struct ubifs_zbranch *zbr)
1706 union ubifs_key key1;
1707 struct ubifs_ch *ch = buf;
1710 if (ch->node_type != UBIFS_DATA_NODE) {
1711 ubifs_err(c, "bad node type (%d but expected %d)",
1712 ch->node_type, UBIFS_DATA_NODE);
1716 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1718 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1722 err = ubifs_node_check_hash(c, buf, zbr->hash);
1724 ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs);
1728 len = le32_to_cpu(ch->len);
1729 if (len != zbr->len) {
1730 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1734 /* Make sure the key of the read node is correct */
1735 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1736 if (!keys_eq(c, &zbr->key, &key1)) {
1737 ubifs_err(c, "bad key in node at LEB %d:%d",
1738 zbr->lnum, zbr->offs);
1739 dbg_tnck(&zbr->key, "looked for key ");
1740 dbg_tnck(&key1, "found node's key ");
1749 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1750 ubifs_dump_node(c, buf);
1756 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1757 * @c: UBIFS file-system description object
1758 * @bu: bulk-read parameters and results
1760 * This functions reads and validates the data nodes that were identified by the
1761 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1762 * -EAGAIN to indicate a race with GC, or another negative error code on
1765 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1767 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1768 struct ubifs_wbuf *wbuf;
1771 len = bu->zbranch[bu->cnt - 1].offs;
1772 len += bu->zbranch[bu->cnt - 1].len - offs;
1773 if (len > bu->buf_len) {
1774 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1779 wbuf = ubifs_get_wbuf(c, lnum);
1781 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1783 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1785 /* Check for a race with GC */
1786 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1789 if (err && err != -EBADMSG) {
1790 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1793 dbg_tnck(&bu->key, "key ");
1797 /* Validate the nodes read */
1799 for (i = 0; i < bu->cnt; i++) {
1800 err = validate_data_node(c, buf, &bu->zbranch[i]);
1803 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1810 * do_lookup_nm- look up a "hashed" node.
1811 * @c: UBIFS file-system description object
1812 * @key: node key to lookup
1813 * @node: the node is returned here
1816 * This function looks up and reads a node which contains name hash in the key.
1817 * Since the hash may have collisions, there may be many nodes with the same
1818 * key, so we have to sequentially look to all of them until the needed one is
1819 * found. This function returns zero in case of success, %-ENOENT if the node
1820 * was not found, and a negative error code in case of failure.
1822 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1823 void *node, const struct fscrypt_name *nm)
1826 struct ubifs_znode *znode;
1828 dbg_tnck(key, "key ");
1829 mutex_lock(&c->tnc_mutex);
1830 found = ubifs_lookup_level0(c, key, &znode, &n);
1834 } else if (found < 0) {
1839 ubifs_assert(c, n >= 0);
1841 err = resolve_collision(c, key, &znode, &n, nm);
1842 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1843 if (unlikely(err < 0))
1850 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1853 mutex_unlock(&c->tnc_mutex);
1858 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1859 * @c: UBIFS file-system description object
1860 * @key: node key to lookup
1861 * @node: the node is returned here
1864 * This function looks up and reads a node which contains name hash in the key.
1865 * Since the hash may have collisions, there may be many nodes with the same
1866 * key, so we have to sequentially look to all of them until the needed one is
1867 * found. This function returns zero in case of success, %-ENOENT if the node
1868 * was not found, and a negative error code in case of failure.
1870 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1871 void *node, const struct fscrypt_name *nm)
1874 const struct ubifs_dent_node *dent = node;
1877 * We assume that in most of the cases there are no name collisions and
1878 * 'ubifs_tnc_lookup()' returns us the right direntry.
1880 err = ubifs_tnc_lookup(c, key, node);
1884 len = le16_to_cpu(dent->nlen);
1885 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1889 * Unluckily, there are hash collisions and we have to iterate over
1890 * them look at each direntry with colliding name hash sequentially.
1893 return do_lookup_nm(c, key, node, nm);
1896 static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1897 struct ubifs_dent_node *dent, uint32_t cookie,
1898 struct ubifs_znode **zn, int *n)
1901 struct ubifs_znode *znode = *zn;
1902 struct ubifs_zbranch *zbr;
1903 union ubifs_key *dkey;
1906 zbr = &znode->zbranch[*n];
1909 if (key_inum(c, dkey) != key_inum(c, key) ||
1910 key_type(c, dkey) != key_type(c, key)) {
1914 err = tnc_read_hashed_node(c, zbr, dent);
1918 if (key_hash(c, key) == key_hash(c, dkey) &&
1919 le32_to_cpu(dent->cookie) == cookie) {
1924 err = tnc_next(c, &znode, n);
1930 static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1931 struct ubifs_dent_node *dent, uint32_t cookie)
1934 struct ubifs_znode *znode;
1935 union ubifs_key start_key;
1937 ubifs_assert(c, is_hash_key(c, key));
1939 lowest_dent_key(c, &start_key, key_inum(c, key));
1941 mutex_lock(&c->tnc_mutex);
1942 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1943 if (unlikely(err < 0))
1946 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
1949 mutex_unlock(&c->tnc_mutex);
1954 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1955 * @c: UBIFS file-system description object
1956 * @key: node key to lookup
1957 * @node: the node is returned here
1958 * @cookie: node cookie for collision resolution
1960 * This function looks up and reads a node which contains name hash in the key.
1961 * Since the hash may have collisions, there may be many nodes with the same
1962 * key, so we have to sequentially look to all of them until the needed one
1963 * with the same cookie value is found.
1964 * This function returns zero in case of success, %-ENOENT if the node
1965 * was not found, and a negative error code in case of failure.
1967 int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1968 void *node, uint32_t cookie)
1971 const struct ubifs_dent_node *dent = node;
1973 if (!c->double_hash)
1977 * We assume that in most of the cases there are no name collisions and
1978 * 'ubifs_tnc_lookup()' returns us the right direntry.
1980 err = ubifs_tnc_lookup(c, key, node);
1984 if (le32_to_cpu(dent->cookie) == cookie)
1988 * Unluckily, there are hash collisions and we have to iterate over
1989 * them look at each direntry with colliding name hash sequentially.
1991 return do_lookup_dh(c, key, node, cookie);
1995 * correct_parent_keys - correct parent znodes' keys.
1996 * @c: UBIFS file-system description object
1997 * @znode: znode to correct parent znodes for
1999 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
2000 * zbranch changes, keys of parent znodes have to be corrected. This helper
2001 * function is called in such situations and corrects the keys if needed.
2003 static void correct_parent_keys(const struct ubifs_info *c,
2004 struct ubifs_znode *znode)
2006 union ubifs_key *key, *key1;
2008 ubifs_assert(c, znode->parent);
2009 ubifs_assert(c, znode->iip == 0);
2011 key = &znode->zbranch[0].key;
2012 key1 = &znode->parent->zbranch[0].key;
2014 while (keys_cmp(c, key, key1) < 0) {
2015 key_copy(c, key, key1);
2016 znode = znode->parent;
2018 if (!znode->parent || znode->iip)
2020 key1 = &znode->parent->zbranch[0].key;
2025 * insert_zbranch - insert a zbranch into a znode.
2026 * @c: UBIFS file-system description object
2027 * @znode: znode into which to insert
2028 * @zbr: zbranch to insert
2029 * @n: slot number to insert to
2031 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2032 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2033 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2034 * slot, zbranches starting from @n have to be moved right.
2036 static void insert_zbranch(struct ubifs_info *c, struct ubifs_znode *znode,
2037 const struct ubifs_zbranch *zbr, int n)
2041 ubifs_assert(c, ubifs_zn_dirty(znode));
2044 for (i = znode->child_cnt; i > n; i--) {
2045 znode->zbranch[i] = znode->zbranch[i - 1];
2046 if (znode->zbranch[i].znode)
2047 znode->zbranch[i].znode->iip = i;
2050 zbr->znode->iip = n;
2052 for (i = znode->child_cnt; i > n; i--)
2053 znode->zbranch[i] = znode->zbranch[i - 1];
2055 znode->zbranch[n] = *zbr;
2056 znode->child_cnt += 1;
2059 * After inserting at slot zero, the lower bound of the key range of
2060 * this znode may have changed. If this znode is subsequently split
2061 * then the upper bound of the key range may change, and furthermore
2062 * it could change to be lower than the original lower bound. If that
2063 * happens, then it will no longer be possible to find this znode in the
2064 * TNC using the key from the index node on flash. That is bad because
2065 * if it is not found, we will assume it is obsolete and may overwrite
2066 * it. Then if there is an unclean unmount, we will start using the
2067 * old index which will be broken.
2069 * So we first mark znodes that have insertions at slot zero, and then
2070 * if they are split we add their lnum/offs to the old_idx tree.
2077 * tnc_insert - insert a node into TNC.
2078 * @c: UBIFS file-system description object
2079 * @znode: znode to insert into
2080 * @zbr: branch to insert
2081 * @n: slot number to insert new zbranch to
2083 * This function inserts a new node described by @zbr into znode @znode. If
2084 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2085 * are splat as well if needed. Returns zero in case of success or a negative
2086 * error code in case of failure.
2088 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2089 struct ubifs_zbranch *zbr, int n)
2091 struct ubifs_znode *zn, *zi, *zp;
2092 int i, keep, move, appending = 0;
2093 union ubifs_key *key = &zbr->key, *key1;
2095 ubifs_assert(c, n >= 0 && n <= c->fanout);
2097 /* Implement naive insert for now */
2100 if (znode->child_cnt < c->fanout) {
2101 ubifs_assert(c, n != c->fanout);
2102 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2104 insert_zbranch(c, znode, zbr, n);
2106 /* Ensure parent's key is correct */
2107 if (n == 0 && zp && znode->iip == 0)
2108 correct_parent_keys(c, znode);
2114 * Unfortunately, @znode does not have more empty slots and we have to
2117 dbg_tnck(key, "splitting level %d, key ", znode->level);
2121 * We can no longer be sure of finding this znode by key, so we
2122 * record it in the old_idx tree.
2124 ins_clr_old_idx_znode(c, znode);
2126 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2130 zn->level = znode->level;
2132 /* Decide where to split */
2133 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2134 /* Try not to split consecutive data keys */
2135 if (n == c->fanout) {
2136 key1 = &znode->zbranch[n - 1].key;
2137 if (key_inum(c, key1) == key_inum(c, key) &&
2138 key_type(c, key1) == UBIFS_DATA_KEY)
2142 } else if (appending && n != c->fanout) {
2143 /* Try not to split consecutive data keys */
2146 if (n >= (c->fanout + 1) / 2) {
2147 key1 = &znode->zbranch[0].key;
2148 if (key_inum(c, key1) == key_inum(c, key) &&
2149 key_type(c, key1) == UBIFS_DATA_KEY) {
2150 key1 = &znode->zbranch[n].key;
2151 if (key_inum(c, key1) != key_inum(c, key) ||
2152 key_type(c, key1) != UBIFS_DATA_KEY) {
2154 move = c->fanout - keep;
2166 keep = (c->fanout + 1) / 2;
2167 move = c->fanout - keep;
2171 * Although we don't at present, we could look at the neighbors and see
2172 * if we can move some zbranches there.
2176 /* Insert into existing znode */
2181 /* Insert into new znode */
2186 zbr->znode->parent = zn;
2191 __set_bit(DIRTY_ZNODE, &zn->flags);
2192 atomic_long_inc(&c->dirty_zn_cnt);
2194 zn->child_cnt = move;
2195 znode->child_cnt = keep;
2197 dbg_tnc("moving %d, keeping %d", move, keep);
2200 for (i = 0; i < move; i++) {
2201 zn->zbranch[i] = znode->zbranch[keep + i];
2204 if (zn->zbranch[i].znode) {
2205 zn->zbranch[i].znode->parent = zn;
2206 zn->zbranch[i].znode->iip = i;
2210 /* Insert new key and branch */
2211 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2213 insert_zbranch(c, zi, zbr, n);
2215 /* Insert new znode (produced by spitting) into the parent */
2217 if (n == 0 && zi == znode && znode->iip == 0)
2218 correct_parent_keys(c, znode);
2220 /* Locate insertion point */
2223 /* Tail recursion */
2224 zbr->key = zn->zbranch[0].key;
2234 /* We have to split root znode */
2235 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2237 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2242 zi->level = znode->level + 1;
2244 __set_bit(DIRTY_ZNODE, &zi->flags);
2245 atomic_long_inc(&c->dirty_zn_cnt);
2247 zi->zbranch[0].key = znode->zbranch[0].key;
2248 zi->zbranch[0].znode = znode;
2249 zi->zbranch[0].lnum = c->zroot.lnum;
2250 zi->zbranch[0].offs = c->zroot.offs;
2251 zi->zbranch[0].len = c->zroot.len;
2252 zi->zbranch[1].key = zn->zbranch[0].key;
2253 zi->zbranch[1].znode = zn;
2258 c->zroot.znode = zi;
2269 * ubifs_tnc_add - add a node to TNC.
2270 * @c: UBIFS file-system description object
2272 * @lnum: LEB number of node
2273 * @offs: node offset
2275 * @hash: The hash over the node
2277 * This function adds a node with key @key to TNC. The node may be new or it may
2278 * obsolete some existing one. Returns %0 on success or negative error code on
2281 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2282 int offs, int len, const u8 *hash)
2284 int found, n, err = 0;
2285 struct ubifs_znode *znode;
2287 mutex_lock(&c->tnc_mutex);
2288 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2289 found = lookup_level0_dirty(c, key, &znode, &n);
2291 struct ubifs_zbranch zbr;
2297 ubifs_copy_hash(c, hash, zbr.hash);
2298 key_copy(c, key, &zbr.key);
2299 err = tnc_insert(c, znode, &zbr, n + 1);
2300 } else if (found == 1) {
2301 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2304 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2308 ubifs_copy_hash(c, hash, zbr->hash);
2312 err = dbg_check_tnc(c, 0);
2313 mutex_unlock(&c->tnc_mutex);
2319 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2320 * @c: UBIFS file-system description object
2322 * @old_lnum: LEB number of old node
2323 * @old_offs: old node offset
2324 * @lnum: LEB number of node
2325 * @offs: node offset
2328 * This function replaces a node with key @key in the TNC only if the old node
2329 * is found. This function is called by garbage collection when node are moved.
2330 * Returns %0 on success or negative error code on failure.
2332 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2333 int old_lnum, int old_offs, int lnum, int offs, int len)
2335 int found, n, err = 0;
2336 struct ubifs_znode *znode;
2338 mutex_lock(&c->tnc_mutex);
2339 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2340 old_offs, lnum, offs, len);
2341 found = lookup_level0_dirty(c, key, &znode, &n);
2348 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2351 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2353 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2360 } else if (is_hash_key(c, key)) {
2361 found = resolve_collision_directly(c, key, &znode, &n,
2362 old_lnum, old_offs);
2363 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2364 found, znode, n, old_lnum, old_offs);
2371 /* Ensure the znode is dirtied */
2372 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2373 znode = dirty_cow_bottom_up(c, znode);
2374 if (IS_ERR(znode)) {
2375 err = PTR_ERR(znode);
2379 zbr = &znode->zbranch[n];
2381 err = ubifs_add_dirt(c, zbr->lnum,
2393 err = ubifs_add_dirt(c, lnum, len);
2396 err = dbg_check_tnc(c, 0);
2399 mutex_unlock(&c->tnc_mutex);
2404 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2405 * @c: UBIFS file-system description object
2407 * @lnum: LEB number of node
2408 * @offs: node offset
2410 * @hash: The hash over the node
2413 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2414 * may have collisions, like directory entry keys.
2416 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2417 int lnum, int offs, int len, const u8 *hash,
2418 const struct fscrypt_name *nm)
2420 int found, n, err = 0;
2421 struct ubifs_znode *znode;
2423 mutex_lock(&c->tnc_mutex);
2424 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2425 found = lookup_level0_dirty(c, key, &znode, &n);
2433 found = fallible_resolve_collision(c, key, &znode, &n,
2436 found = resolve_collision(c, key, &znode, &n, nm);
2437 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2443 /* Ensure the znode is dirtied */
2444 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2445 znode = dirty_cow_bottom_up(c, znode);
2446 if (IS_ERR(znode)) {
2447 err = PTR_ERR(znode);
2453 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2456 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2460 ubifs_copy_hash(c, hash, zbr->hash);
2466 struct ubifs_zbranch zbr;
2472 ubifs_copy_hash(c, hash, zbr.hash);
2473 key_copy(c, key, &zbr.key);
2474 err = tnc_insert(c, znode, &zbr, n + 1);
2479 * We did not find it in the index so there may be a
2480 * dangling branch still in the index. So we remove it
2481 * by passing 'ubifs_tnc_remove_nm()' the same key but
2482 * an unmatchable name.
2484 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2486 err = dbg_check_tnc(c, 0);
2487 mutex_unlock(&c->tnc_mutex);
2490 return ubifs_tnc_remove_nm(c, key, &noname);
2496 err = dbg_check_tnc(c, 0);
2497 mutex_unlock(&c->tnc_mutex);
2502 * tnc_delete - delete a znode form TNC.
2503 * @c: UBIFS file-system description object
2504 * @znode: znode to delete from
2505 * @n: zbranch slot number to delete
2507 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2508 * case of success and a negative error code in case of failure.
2510 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2512 struct ubifs_zbranch *zbr;
2513 struct ubifs_znode *zp;
2516 /* Delete without merge for now */
2517 ubifs_assert(c, znode->level == 0);
2518 ubifs_assert(c, n >= 0 && n < c->fanout);
2519 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2521 zbr = &znode->zbranch[n];
2524 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2526 ubifs_dump_znode(c, znode);
2530 /* We do not "gap" zbranch slots */
2531 for (i = n; i < znode->child_cnt - 1; i++)
2532 znode->zbranch[i] = znode->zbranch[i + 1];
2533 znode->child_cnt -= 1;
2535 if (znode->child_cnt > 0)
2539 * This was the last zbranch, we have to delete this znode from the
2544 ubifs_assert(c, !ubifs_zn_obsolete(znode));
2545 ubifs_assert(c, ubifs_zn_dirty(znode));
2550 atomic_long_dec(&c->dirty_zn_cnt);
2552 err = insert_old_idx_znode(c, znode);
2557 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2558 atomic_long_inc(&c->clean_zn_cnt);
2559 atomic_long_inc(&ubifs_clean_zn_cnt);
2563 } while (znode->child_cnt == 1); /* while removing last child */
2565 /* Remove from znode, entry n - 1 */
2566 znode->child_cnt -= 1;
2567 ubifs_assert(c, znode->level != 0);
2568 for (i = n; i < znode->child_cnt; i++) {
2569 znode->zbranch[i] = znode->zbranch[i + 1];
2570 if (znode->zbranch[i].znode)
2571 znode->zbranch[i].znode->iip = i;
2575 * If this is the root and it has only 1 child then
2576 * collapse the tree.
2578 if (!znode->parent) {
2579 while (znode->child_cnt == 1 && znode->level != 0) {
2581 zbr = &znode->zbranch[0];
2582 znode = get_znode(c, znode, 0);
2584 return PTR_ERR(znode);
2585 znode = dirty_cow_znode(c, zbr);
2587 return PTR_ERR(znode);
2588 znode->parent = NULL;
2591 err = insert_old_idx(c, c->zroot.lnum,
2596 c->zroot.lnum = zbr->lnum;
2597 c->zroot.offs = zbr->offs;
2598 c->zroot.len = zbr->len;
2599 c->zroot.znode = znode;
2600 ubifs_assert(c, !ubifs_zn_obsolete(zp));
2601 ubifs_assert(c, ubifs_zn_dirty(zp));
2602 atomic_long_dec(&c->dirty_zn_cnt);
2605 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2606 atomic_long_inc(&c->clean_zn_cnt);
2607 atomic_long_inc(&ubifs_clean_zn_cnt);
2617 * ubifs_tnc_remove - remove an index entry of a node.
2618 * @c: UBIFS file-system description object
2621 * Returns %0 on success or negative error code on failure.
2623 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2625 int found, n, err = 0;
2626 struct ubifs_znode *znode;
2628 mutex_lock(&c->tnc_mutex);
2629 dbg_tnck(key, "key ");
2630 found = lookup_level0_dirty(c, key, &znode, &n);
2636 err = tnc_delete(c, znode, n);
2638 err = dbg_check_tnc(c, 0);
2641 mutex_unlock(&c->tnc_mutex);
2646 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2647 * @c: UBIFS file-system description object
2649 * @nm: directory entry name
2651 * Returns %0 on success or negative error code on failure.
2653 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2654 const struct fscrypt_name *nm)
2657 struct ubifs_znode *znode;
2659 mutex_lock(&c->tnc_mutex);
2660 dbg_tnck(key, "key ");
2661 err = lookup_level0_dirty(c, key, &znode, &n);
2667 err = fallible_resolve_collision(c, key, &znode, &n,
2670 err = resolve_collision(c, key, &znode, &n, nm);
2671 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2675 /* Ensure the znode is dirtied */
2676 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2677 znode = dirty_cow_bottom_up(c, znode);
2678 if (IS_ERR(znode)) {
2679 err = PTR_ERR(znode);
2683 err = tnc_delete(c, znode, n);
2689 err = dbg_check_tnc(c, 0);
2690 mutex_unlock(&c->tnc_mutex);
2695 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2696 * @c: UBIFS file-system description object
2698 * @cookie: node cookie for collision resolution
2700 * Returns %0 on success or negative error code on failure.
2702 int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2706 struct ubifs_znode *znode;
2707 struct ubifs_dent_node *dent;
2708 struct ubifs_zbranch *zbr;
2710 if (!c->double_hash)
2713 mutex_lock(&c->tnc_mutex);
2714 err = lookup_level0_dirty(c, key, &znode, &n);
2718 zbr = &znode->zbranch[n];
2719 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2725 err = tnc_read_hashed_node(c, zbr, dent);
2729 /* If the cookie does not match, we're facing a hash collision. */
2730 if (le32_to_cpu(dent->cookie) != cookie) {
2731 union ubifs_key start_key;
2733 lowest_dent_key(c, &start_key, key_inum(c, key));
2735 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2736 if (unlikely(err < 0))
2739 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
2744 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2745 znode = dirty_cow_bottom_up(c, znode);
2746 if (IS_ERR(znode)) {
2747 err = PTR_ERR(znode);
2751 err = tnc_delete(c, znode, n);
2757 err = dbg_check_tnc(c, 0);
2758 mutex_unlock(&c->tnc_mutex);
2763 * key_in_range - determine if a key falls within a range of keys.
2764 * @c: UBIFS file-system description object
2765 * @key: key to check
2766 * @from_key: lowest key in range
2767 * @to_key: highest key in range
2769 * This function returns %1 if the key is in range and %0 otherwise.
2771 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2772 union ubifs_key *from_key, union ubifs_key *to_key)
2774 if (keys_cmp(c, key, from_key) < 0)
2776 if (keys_cmp(c, key, to_key) > 0)
2782 * ubifs_tnc_remove_range - remove index entries in range.
2783 * @c: UBIFS file-system description object
2784 * @from_key: lowest key to remove
2785 * @to_key: highest key to remove
2787 * This function removes index entries starting at @from_key and ending at
2788 * @to_key. This function returns zero in case of success and a negative error
2789 * code in case of failure.
2791 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2792 union ubifs_key *to_key)
2794 int i, n, k, err = 0;
2795 struct ubifs_znode *znode;
2796 union ubifs_key *key;
2798 mutex_lock(&c->tnc_mutex);
2800 /* Find first level 0 znode that contains keys to remove */
2801 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2808 err = tnc_next(c, &znode, &n);
2809 if (err == -ENOENT) {
2815 key = &znode->zbranch[n].key;
2816 if (!key_in_range(c, key, from_key, to_key)) {
2822 /* Ensure the znode is dirtied */
2823 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2824 znode = dirty_cow_bottom_up(c, znode);
2825 if (IS_ERR(znode)) {
2826 err = PTR_ERR(znode);
2831 /* Remove all keys in range except the first */
2832 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2833 key = &znode->zbranch[i].key;
2834 if (!key_in_range(c, key, from_key, to_key))
2836 lnc_free(&znode->zbranch[i]);
2837 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2838 znode->zbranch[i].len);
2840 ubifs_dump_znode(c, znode);
2843 dbg_tnck(key, "removing key ");
2846 for (i = n + 1 + k; i < znode->child_cnt; i++)
2847 znode->zbranch[i - k] = znode->zbranch[i];
2848 znode->child_cnt -= k;
2851 /* Now delete the first */
2852 err = tnc_delete(c, znode, n);
2859 err = dbg_check_tnc(c, 0);
2860 mutex_unlock(&c->tnc_mutex);
2865 * ubifs_tnc_remove_ino - remove an inode from TNC.
2866 * @c: UBIFS file-system description object
2867 * @inum: inode number to remove
2869 * This function remove inode @inum and all the extended attributes associated
2870 * with the anode from TNC and returns zero in case of success or a negative
2871 * error code in case of failure.
2873 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2875 union ubifs_key key1, key2;
2876 struct ubifs_dent_node *xent, *pxent = NULL;
2877 struct fscrypt_name nm = {0};
2879 dbg_tnc("ino %lu", (unsigned long)inum);
2882 * Walk all extended attribute entries and remove them together with
2883 * corresponding extended attribute inodes.
2885 lowest_xent_key(c, &key1, inum);
2890 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2892 err = PTR_ERR(xent);
2898 xattr_inum = le64_to_cpu(xent->inum);
2899 dbg_tnc("xent '%s', ino %lu", xent->name,
2900 (unsigned long)xattr_inum);
2902 ubifs_evict_xattr_inode(c, xattr_inum);
2904 fname_name(&nm) = xent->name;
2905 fname_len(&nm) = le16_to_cpu(xent->nlen);
2906 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2912 lowest_ino_key(c, &key1, xattr_inum);
2913 highest_ino_key(c, &key2, xattr_inum);
2914 err = ubifs_tnc_remove_range(c, &key1, &key2);
2922 key_read(c, &xent->key, &key1);
2926 lowest_ino_key(c, &key1, inum);
2927 highest_ino_key(c, &key2, inum);
2929 return ubifs_tnc_remove_range(c, &key1, &key2);
2933 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2934 * @c: UBIFS file-system description object
2935 * @key: key of last entry
2936 * @nm: name of last entry found or %NULL
2938 * This function finds and reads the next directory or extended attribute entry
2939 * after the given key (@key) if there is one. @nm is used to resolve
2942 * If the name of the current entry is not known and only the key is known,
2943 * @nm->name has to be %NULL. In this case the semantics of this function is a
2944 * little bit different and it returns the entry corresponding to this key, not
2945 * the next one. If the key was not found, the closest "right" entry is
2948 * If the fist entry has to be found, @key has to contain the lowest possible
2949 * key value for this inode and @name has to be %NULL.
2951 * This function returns the found directory or extended attribute entry node
2952 * in case of success, %-ENOENT is returned if no entry was found, and a
2953 * negative error code is returned in case of failure.
2955 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2956 union ubifs_key *key,
2957 const struct fscrypt_name *nm)
2959 int n, err, type = key_type(c, key);
2960 struct ubifs_znode *znode;
2961 struct ubifs_dent_node *dent;
2962 struct ubifs_zbranch *zbr;
2963 union ubifs_key *dkey;
2965 dbg_tnck(key, "key ");
2966 ubifs_assert(c, is_hash_key(c, key));
2968 mutex_lock(&c->tnc_mutex);
2969 err = ubifs_lookup_level0(c, key, &znode, &n);
2970 if (unlikely(err < 0))
2973 if (fname_len(nm) > 0) {
2975 /* Handle collisions */
2977 err = fallible_resolve_collision(c, key, &znode, &n,
2980 err = resolve_collision(c, key, &znode, &n, nm);
2981 dbg_tnc("rc returned %d, znode %p, n %d",
2983 if (unlikely(err < 0))
2987 /* Now find next entry */
2988 err = tnc_next(c, &znode, &n);
2993 * The full name of the entry was not given, in which case the
2994 * behavior of this function is a little different and it
2995 * returns current entry, not the next one.
2999 * However, the given key does not exist in the TNC
3000 * tree and @znode/@n variables contain the closest
3001 * "preceding" element. Switch to the next one.
3003 err = tnc_next(c, &znode, &n);
3009 zbr = &znode->zbranch[n];
3010 dent = kmalloc(zbr->len, GFP_NOFS);
3011 if (unlikely(!dent)) {
3017 * The above 'tnc_next()' call could lead us to the next inode, check
3021 if (key_inum(c, dkey) != key_inum(c, key) ||
3022 key_type(c, dkey) != type) {
3027 err = tnc_read_hashed_node(c, zbr, dent);
3031 mutex_unlock(&c->tnc_mutex);
3037 mutex_unlock(&c->tnc_mutex);
3038 return ERR_PTR(err);
3042 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3043 * @c: UBIFS file-system description object
3045 * Destroy left-over obsolete znodes from a failed commit.
3047 static void tnc_destroy_cnext(struct ubifs_info *c)
3049 struct ubifs_znode *cnext;
3053 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3056 struct ubifs_znode *znode = cnext;
3058 cnext = cnext->cnext;
3059 if (ubifs_zn_obsolete(znode))
3061 } while (cnext && cnext != c->cnext);
3065 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3066 * @c: UBIFS file-system description object
3068 void ubifs_tnc_close(struct ubifs_info *c)
3070 tnc_destroy_cnext(c);
3071 if (c->zroot.znode) {
3074 n = atomic_long_read(&c->clean_zn_cnt);
3075 freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode);
3076 ubifs_assert(c, freed == n);
3077 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3085 * left_znode - get the znode to the left.
3086 * @c: UBIFS file-system description object
3089 * This function returns a pointer to the znode to the left of @znode or NULL if
3090 * there is not one. A negative error code is returned on failure.
3092 static struct ubifs_znode *left_znode(struct ubifs_info *c,
3093 struct ubifs_znode *znode)
3095 int level = znode->level;
3098 int n = znode->iip - 1;
3100 /* Go up until we can go left */
3101 znode = znode->parent;
3105 /* Now go down the rightmost branch to 'level' */
3106 znode = get_znode(c, znode, n);
3109 while (znode->level != level) {
3110 n = znode->child_cnt - 1;
3111 znode = get_znode(c, znode, n);
3122 * right_znode - get the znode to the right.
3123 * @c: UBIFS file-system description object
3126 * This function returns a pointer to the znode to the right of @znode or NULL
3127 * if there is not one. A negative error code is returned on failure.
3129 static struct ubifs_znode *right_znode(struct ubifs_info *c,
3130 struct ubifs_znode *znode)
3132 int level = znode->level;
3135 int n = znode->iip + 1;
3137 /* Go up until we can go right */
3138 znode = znode->parent;
3141 if (n < znode->child_cnt) {
3142 /* Now go down the leftmost branch to 'level' */
3143 znode = get_znode(c, znode, n);
3146 while (znode->level != level) {
3147 znode = get_znode(c, znode, 0);
3158 * lookup_znode - find a particular indexing node from TNC.
3159 * @c: UBIFS file-system description object
3160 * @key: index node key to lookup
3161 * @level: index node level
3162 * @lnum: index node LEB number
3163 * @offs: index node offset
3165 * This function searches an indexing node by its first key @key and its
3166 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3167 * nodes it traverses to TNC. This function is called for indexing nodes which
3168 * were found on the media by scanning, for example when garbage-collecting or
3169 * when doing in-the-gaps commit. This means that the indexing node which is
3170 * looked for does not have to have exactly the same leftmost key @key, because
3171 * the leftmost key may have been changed, in which case TNC will contain a
3172 * dirty znode which still refers the same @lnum:@offs. This function is clever
3173 * enough to recognize such indexing nodes.
3175 * Note, if a znode was deleted or changed too much, then this function will
3176 * not find it. For situations like this UBIFS has the old index RB-tree
3177 * (indexed by @lnum:@offs).
3179 * This function returns a pointer to the znode found or %NULL if it is not
3180 * found. A negative error code is returned on failure.
3182 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3183 union ubifs_key *key, int level,
3186 struct ubifs_znode *znode, *zn;
3189 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3192 * The arguments have probably been read off flash, so don't assume
3196 return ERR_PTR(-EINVAL);
3198 /* Get the root znode */
3199 znode = c->zroot.znode;
3201 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3205 /* Check if it is the one we are looking for */
3206 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3208 /* Descend to the parent level i.e. (level + 1) */
3209 if (level >= znode->level)
3212 ubifs_search_zbranch(c, znode, key, &n);
3215 * We reached a znode where the leftmost key is greater
3216 * than the key we are searching for. This is the same
3217 * situation as the one described in a huge comment at
3218 * the end of the 'ubifs_lookup_level0()' function. And
3219 * for exactly the same reasons we have to try to look
3220 * left before giving up.
3222 znode = left_znode(c, znode);
3227 ubifs_search_zbranch(c, znode, key, &n);
3228 ubifs_assert(c, n >= 0);
3230 if (znode->level == level + 1)
3232 znode = get_znode(c, znode, n);
3236 /* Check if the child is the one we are looking for */
3237 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3238 return get_znode(c, znode, n);
3239 /* If the key is unique, there is nowhere else to look */
3240 if (!is_hash_key(c, key))
3243 * The key is not unique and so may be also in the znodes to either
3250 /* Move one branch to the left */
3254 znode = left_znode(c, znode);
3259 n = znode->child_cnt - 1;
3262 if (znode->zbranch[n].lnum == lnum &&
3263 znode->zbranch[n].offs == offs)
3264 return get_znode(c, znode, n);
3265 /* Stop if the key is less than the one we are looking for */
3266 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3269 /* Back to the middle */
3274 /* Move one branch to the right */
3275 if (++n >= znode->child_cnt) {
3276 znode = right_znode(c, znode);
3284 if (znode->zbranch[n].lnum == lnum &&
3285 znode->zbranch[n].offs == offs)
3286 return get_znode(c, znode, n);
3287 /* Stop if the key is greater than the one we are looking for */
3288 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3295 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3296 * @c: UBIFS file-system description object
3297 * @key: key of index node
3298 * @level: index node level
3299 * @lnum: LEB number of index node
3300 * @offs: offset of index node
3302 * This function returns %0 if the index node is not referred to in the TNC, %1
3303 * if the index node is referred to in the TNC and the corresponding znode is
3304 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3305 * znode is clean, and a negative error code in case of failure.
3307 * Note, the @key argument has to be the key of the first child. Also note,
3308 * this function relies on the fact that 0:0 is never a valid LEB number and
3309 * offset for a main-area node.
3311 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3314 struct ubifs_znode *znode;
3316 znode = lookup_znode(c, key, level, lnum, offs);
3320 return PTR_ERR(znode);
3322 return ubifs_zn_dirty(znode) ? 1 : 2;
3326 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3327 * @c: UBIFS file-system description object
3329 * @lnum: node LEB number
3330 * @offs: node offset
3332 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3333 * not, and a negative error code in case of failure.
3335 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3336 * and offset for a main-area node.
3338 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3341 struct ubifs_zbranch *zbr;
3342 struct ubifs_znode *znode, *zn;
3343 int n, found, err, nn;
3344 const int unique = !is_hash_key(c, key);
3346 found = ubifs_lookup_level0(c, key, &znode, &n);
3348 return found; /* Error code */
3351 zbr = &znode->zbranch[n];
3352 if (lnum == zbr->lnum && offs == zbr->offs)
3353 return 1; /* Found it */
3357 * Because the key is not unique, we have to look left
3364 err = tnc_prev(c, &znode, &n);
3369 if (keys_cmp(c, key, &znode->zbranch[n].key))
3371 zbr = &znode->zbranch[n];
3372 if (lnum == zbr->lnum && offs == zbr->offs)
3373 return 1; /* Found it */
3379 err = tnc_next(c, &znode, &n);
3385 if (keys_cmp(c, key, &znode->zbranch[n].key))
3387 zbr = &znode->zbranch[n];
3388 if (lnum == zbr->lnum && offs == zbr->offs)
3389 return 1; /* Found it */
3395 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3396 * @c: UBIFS file-system description object
3398 * @level: index node level (if it is an index node)
3399 * @lnum: node LEB number
3400 * @offs: node offset
3401 * @is_idx: non-zero if the node is an index node
3403 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3404 * negative error code in case of failure. For index nodes, @key has to be the
3405 * key of the first child. An index node is considered to be in the TNC only if
3406 * the corresponding znode is clean or has not been loaded.
3408 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3409 int lnum, int offs, int is_idx)
3413 mutex_lock(&c->tnc_mutex);
3415 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3419 /* The index node was found but it was dirty */
3422 /* The index node was found and it was clean */
3427 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3430 mutex_unlock(&c->tnc_mutex);
3435 * ubifs_dirty_idx_node - dirty an index node.
3436 * @c: UBIFS file-system description object
3437 * @key: index node key
3438 * @level: index node level
3439 * @lnum: index node LEB number
3440 * @offs: index node offset
3442 * This function loads and dirties an index node so that it can be garbage
3443 * collected. The @key argument has to be the key of the first child. This
3444 * function relies on the fact that 0:0 is never a valid LEB number and offset
3445 * for a main-area node. Returns %0 on success and a negative error code on
3448 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3451 struct ubifs_znode *znode;
3454 mutex_lock(&c->tnc_mutex);
3455 znode = lookup_znode(c, key, level, lnum, offs);
3458 if (IS_ERR(znode)) {
3459 err = PTR_ERR(znode);
3462 znode = dirty_cow_bottom_up(c, znode);
3463 if (IS_ERR(znode)) {
3464 err = PTR_ERR(znode);
3469 mutex_unlock(&c->tnc_mutex);
3474 * dbg_check_inode_size - check if inode size is correct.
3475 * @c: UBIFS file-system description object
3476 * @inum: inode number
3479 * This function makes sure that the inode size (@size) is correct and it does
3480 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3481 * if it has a data page beyond @size, and other negative error code in case of
3484 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3488 union ubifs_key from_key, to_key, *key;
3489 struct ubifs_znode *znode;
3492 if (!S_ISREG(inode->i_mode))
3494 if (!dbg_is_chk_gen(c))
3497 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3498 data_key_init(c, &from_key, inode->i_ino, block);
3499 highest_data_key(c, &to_key, inode->i_ino);
3501 mutex_lock(&c->tnc_mutex);
3502 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3511 err = tnc_next(c, &znode, &n);
3512 if (err == -ENOENT) {
3519 ubifs_assert(c, err == 0);
3520 key = &znode->zbranch[n].key;
3521 if (!key_in_range(c, key, &from_key, &to_key))
3525 block = key_block(c, key);
3526 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3527 (unsigned long)inode->i_ino, size,
3528 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3529 mutex_unlock(&c->tnc_mutex);
3530 ubifs_dump_inode(c, inode);
3535 mutex_unlock(&c->tnc_mutex);