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 int len, int lnum, int offs);
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(!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 * Obsolete znodes will be freed by tnc_destroy_cnext()
284 * or free_obsolete_znodes(), copied up znodes should
285 * be added back to tnc and freed by
286 * ubifs_destroy_tnc_subtree().
289 err = add_idx_dirt(c, zbr->lnum, zbr->len);
305 * lnc_add - add a leaf node to the leaf node cache.
306 * @c: UBIFS file-system description object
307 * @zbr: zbranch of leaf node
310 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
311 * purpose of the leaf node cache is to save re-reading the same leaf node over
312 * and over again. Most things are cached by VFS, however the file system must
313 * cache directory entries for readdir and for resolving hash collisions. The
314 * present implementation of the leaf node cache is extremely simple, and
315 * allows for error returns that are not used but that may be needed if a more
316 * complex implementation is created.
318 * Note, this function does not add the @node object to LNC directly, but
319 * allocates a copy of the object and adds the copy to LNC. The reason for this
320 * is that @node has been allocated outside of the TNC subsystem and will be
321 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
322 * may be changed at any time, e.g. freed by the shrinker.
324 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
329 const struct ubifs_dent_node *dent = node;
331 ubifs_assert(!zbr->leaf);
332 ubifs_assert(zbr->len != 0);
333 ubifs_assert(is_hash_key(c, &zbr->key));
335 err = ubifs_validate_entry(c, dent);
338 ubifs_dump_node(c, dent);
342 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
344 /* We don't have to have the cache, so no error */
347 zbr->leaf = lnc_node;
352 * lnc_add_directly - add a leaf node to the leaf-node-cache.
353 * @c: UBIFS file-system description object
354 * @zbr: zbranch of leaf node
357 * This function is similar to 'lnc_add()', but it does not create a copy of
358 * @node but inserts @node to TNC directly.
360 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
365 ubifs_assert(!zbr->leaf);
366 ubifs_assert(zbr->len != 0);
368 err = ubifs_validate_entry(c, node);
371 ubifs_dump_node(c, node);
380 * lnc_free - remove a leaf node from the leaf node cache.
381 * @zbr: zbranch of leaf node
384 static void lnc_free(struct ubifs_zbranch *zbr)
393 * tnc_read_hashed_node - read a "hashed" leaf node.
394 * @c: UBIFS file-system description object
395 * @zbr: key and position of the node
396 * @node: node is returned here
398 * This function reads a "hashed" node defined by @zbr from the leaf node cache
399 * (in it is there) or from the hash media, in which case the node is also
400 * added to LNC. Returns zero in case of success or a negative negative error
401 * code in case of failure.
403 static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr,
408 ubifs_assert(is_hash_key(c, &zbr->key));
411 /* Read from the leaf node cache */
412 ubifs_assert(zbr->len != 0);
413 memcpy(node, zbr->leaf, zbr->len);
418 err = fallible_read_node(c, &zbr->key, zbr, node);
420 * When the node was not found, return -ENOENT, 0 otherwise.
421 * Negative return codes stay as-is.
428 err = ubifs_tnc_read_node(c, zbr, node);
433 /* Add the node to the leaf node cache */
434 err = lnc_add(c, zbr, node);
439 * try_read_node - read a node if it is a node.
440 * @c: UBIFS file-system description object
441 * @buf: buffer to read to
443 * @len: node length (not aligned)
444 * @lnum: LEB number of node to read
445 * @offs: offset of node to read
447 * This function tries to read a node of known type and length, checks it and
448 * stores it in @buf. This function returns %1 if a node is present and %0 if
449 * a node is not present. A negative error code is returned for I/O errors.
450 * This function performs that same function as ubifs_read_node except that
451 * it does not require that there is actually a node present and instead
452 * the return code indicates if a node was read.
454 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
455 * is true (it is controlled by corresponding mount option). However, if
456 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
457 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
458 * because during mounting or re-mounting from R/O mode to R/W mode we may read
459 * journal nodes (when replying the journal or doing the recovery) and the
460 * journal nodes may potentially be corrupted, so checking is required.
462 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
463 int len, int lnum, int offs)
466 struct ubifs_ch *ch = buf;
467 uint32_t crc, node_crc;
469 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
471 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
473 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
474 type, lnum, offs, err);
478 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
481 if (ch->node_type != type)
484 node_len = le32_to_cpu(ch->len);
488 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
492 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
493 node_crc = le32_to_cpu(ch->crc);
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->len, zbr->lnum,
520 union ubifs_key node_key;
521 struct ubifs_dent_node *dent = node;
523 /* All nodes have key in the same place */
524 key_read(c, &dent->key, &node_key);
525 if (keys_cmp(c, key, &node_key) != 0)
528 if (ret == 0 && c->replaying)
529 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
530 zbr->lnum, zbr->offs, zbr->len);
535 * matches_name - determine if a direntry or xattr entry matches a given name.
536 * @c: UBIFS file-system description object
537 * @zbr: zbranch of dent
540 * This function checks if xentry/direntry referred by zbranch @zbr matches name
541 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
542 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
543 * of failure, a negative error code is returned.
545 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
546 const struct fscrypt_name *nm)
548 struct ubifs_dent_node *dent;
551 /* If possible, match against the dent in the leaf node cache */
553 dent = kmalloc(zbr->len, GFP_NOFS);
557 err = ubifs_tnc_read_node(c, zbr, dent);
561 /* Add the node to the leaf node cache */
562 err = lnc_add_directly(c, zbr, dent);
568 nlen = le16_to_cpu(dent->nlen);
569 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
571 if (nlen == fname_len(nm))
573 else if (nlen < fname_len(nm))
588 * get_znode - get a TNC znode that may not be loaded yet.
589 * @c: UBIFS file-system description object
590 * @znode: parent znode
591 * @n: znode branch slot number
593 * This function returns the znode or a negative error code.
595 static struct ubifs_znode *get_znode(struct ubifs_info *c,
596 struct ubifs_znode *znode, int n)
598 struct ubifs_zbranch *zbr;
600 zbr = &znode->zbranch[n];
604 znode = ubifs_load_znode(c, zbr, znode, n);
609 * tnc_next - find next TNC entry.
610 * @c: UBIFS file-system description object
611 * @zn: znode is passed and returned here
612 * @n: znode branch slot number is passed and returned here
614 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
615 * no next entry, or a negative error code otherwise.
617 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
619 struct ubifs_znode *znode = *zn;
623 if (nn < znode->child_cnt) {
628 struct ubifs_znode *zp;
635 if (nn < znode->child_cnt) {
636 znode = get_znode(c, znode, nn);
638 return PTR_ERR(znode);
639 while (znode->level != 0) {
640 znode = get_znode(c, znode, 0);
642 return PTR_ERR(znode);
654 * tnc_prev - find previous TNC entry.
655 * @c: UBIFS file-system description object
656 * @zn: znode is returned here
657 * @n: znode branch slot number is passed and returned here
659 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
660 * there is no next entry, or a negative error code otherwise.
662 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
664 struct ubifs_znode *znode = *zn;
672 struct ubifs_znode *zp;
680 znode = get_znode(c, znode, nn);
682 return PTR_ERR(znode);
683 while (znode->level != 0) {
684 nn = znode->child_cnt - 1;
685 znode = get_znode(c, znode, nn);
687 return PTR_ERR(znode);
689 nn = znode->child_cnt - 1;
699 * resolve_collision - resolve a collision.
700 * @c: UBIFS file-system description object
701 * @key: key of a directory or extended attribute entry
702 * @zn: znode is returned here
703 * @n: zbranch number is passed and returned here
704 * @nm: name of the entry
706 * This function is called for "hashed" keys to make sure that the found key
707 * really corresponds to the looked up node (directory or extended attribute
708 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
709 * %0 is returned if @nm is not found and @zn and @n are set to the previous
710 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
711 * This means that @n may be set to %-1 if the leftmost key in @zn is the
712 * previous one. A negative error code is returned on failures.
714 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
715 struct ubifs_znode **zn, int *n,
716 const struct fscrypt_name *nm)
720 err = matches_name(c, &(*zn)->zbranch[*n], nm);
721 if (unlikely(err < 0))
723 if (err == NAME_MATCHES)
726 if (err == NAME_GREATER) {
729 err = tnc_prev(c, zn, n);
730 if (err == -ENOENT) {
731 ubifs_assert(*n == 0);
737 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
739 * We have found the branch after which we would
740 * like to insert, but inserting in this znode
741 * may still be wrong. Consider the following 3
742 * znodes, in the case where we are resolving a
743 * collision with Key2.
746 * ----------------------
747 * level 1 | Key0 | Key1 |
748 * -----------------------
750 * znode za | | znode zb
751 * ------------ ------------
752 * level 0 | Key0 | | Key2 |
753 * ------------ ------------
755 * The lookup finds Key2 in znode zb. Lets say
756 * there is no match and the name is greater so
757 * we look left. When we find Key0, we end up
758 * here. If we return now, we will insert into
759 * znode za at slot n = 1. But that is invalid
760 * according to the parent's keys. Key2 must
761 * be inserted into znode zb.
763 * Note, this problem is not relevant for the
764 * case when we go right, because
765 * 'tnc_insert()' would correct the parent key.
767 if (*n == (*zn)->child_cnt - 1) {
768 err = tnc_next(c, zn, n);
770 /* Should be impossible */
776 ubifs_assert(*n == 0);
781 err = matches_name(c, &(*zn)->zbranch[*n], nm);
784 if (err == NAME_LESS)
786 if (err == NAME_MATCHES)
788 ubifs_assert(err == NAME_GREATER);
792 struct ubifs_znode *znode = *zn;
796 err = tnc_next(c, &znode, &nn);
801 if (keys_cmp(c, &znode->zbranch[nn].key, key))
803 err = matches_name(c, &znode->zbranch[nn], nm);
806 if (err == NAME_GREATER)
810 if (err == NAME_MATCHES)
812 ubifs_assert(err == NAME_LESS);
818 * fallible_matches_name - determine if a dent matches a given name.
819 * @c: UBIFS file-system description object
820 * @zbr: zbranch of dent
823 * This is a "fallible" version of 'matches_name()' function which does not
824 * panic if the direntry/xentry referred by @zbr does not exist on the media.
826 * This function checks if xentry/direntry referred by zbranch @zbr matches name
827 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
828 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
829 * if xentry/direntry referred by @zbr does not exist on the media. A negative
830 * error code is returned in case of failure.
832 static int fallible_matches_name(struct ubifs_info *c,
833 struct ubifs_zbranch *zbr,
834 const struct fscrypt_name *nm)
836 struct ubifs_dent_node *dent;
839 /* If possible, match against the dent in the leaf node cache */
841 dent = kmalloc(zbr->len, GFP_NOFS);
845 err = fallible_read_node(c, &zbr->key, zbr, dent);
849 /* The node was not present */
853 ubifs_assert(err == 1);
855 err = lnc_add_directly(c, zbr, dent);
861 nlen = le16_to_cpu(dent->nlen);
862 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
864 if (nlen == fname_len(nm))
866 else if (nlen < fname_len(nm))
881 * fallible_resolve_collision - resolve a collision even if nodes are missing.
882 * @c: UBIFS file-system description object
884 * @zn: znode is returned here
885 * @n: branch number is passed and returned here
886 * @nm: name of directory entry
887 * @adding: indicates caller is adding a key to the TNC
889 * This is a "fallible" version of the 'resolve_collision()' function which
890 * does not panic if one of the nodes referred to by TNC does not exist on the
891 * media. This may happen when replaying the journal if a deleted node was
892 * Garbage-collected and the commit was not done. A branch that refers to a node
893 * that is not present is called a dangling branch. The following are the return
894 * codes for this function:
895 * o if @nm was found, %1 is returned and @zn and @n are set to the found
897 * o if we are @adding and @nm was not found, %0 is returned;
898 * o if we are not @adding and @nm was not found, but a dangling branch was
899 * found, then %1 is returned and @zn and @n are set to the dangling branch;
900 * o a negative error code is returned in case of failure.
902 static int fallible_resolve_collision(struct ubifs_info *c,
903 const union ubifs_key *key,
904 struct ubifs_znode **zn, int *n,
905 const struct fscrypt_name *nm,
908 struct ubifs_znode *o_znode = NULL, *znode = *zn;
909 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
911 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
912 if (unlikely(cmp < 0))
914 if (cmp == NAME_MATCHES)
916 if (cmp == NOT_ON_MEDIA) {
920 * We are unlucky and hit a dangling branch straight away.
921 * Now we do not really know where to go to find the needed
922 * branch - to the left or to the right. Well, let's try left.
926 unsure = 1; /* Remove a dangling branch wherever it is */
928 if (cmp == NAME_GREATER || unsure) {
931 err = tnc_prev(c, zn, n);
932 if (err == -ENOENT) {
933 ubifs_assert(*n == 0);
939 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
940 /* See comments in 'resolve_collision()' */
941 if (*n == (*zn)->child_cnt - 1) {
942 err = tnc_next(c, zn, n);
944 /* Should be impossible */
950 ubifs_assert(*n == 0);
955 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
958 if (err == NAME_MATCHES)
960 if (err == NOT_ON_MEDIA) {
967 if (err == NAME_LESS)
974 if (cmp == NAME_LESS || unsure) {
979 err = tnc_next(c, &znode, &nn);
984 if (keys_cmp(c, &znode->zbranch[nn].key, key))
986 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
989 if (err == NAME_GREATER)
993 if (err == NAME_MATCHES)
995 if (err == NOT_ON_MEDIA) {
1002 /* Never match a dangling branch when adding */
1003 if (adding || !o_znode)
1006 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
1007 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1008 o_znode->zbranch[o_n].len);
1015 * matches_position - determine if a zbranch matches a given position.
1016 * @zbr: zbranch of dent
1017 * @lnum: LEB number of dent to match
1018 * @offs: offset of dent to match
1020 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1022 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1024 if (zbr->lnum == lnum && zbr->offs == offs)
1031 * resolve_collision_directly - resolve a collision directly.
1032 * @c: UBIFS file-system description object
1033 * @key: key of directory entry
1034 * @zn: znode is passed and returned here
1035 * @n: zbranch number is passed and returned here
1036 * @lnum: LEB number of dent node to match
1037 * @offs: offset of dent node to match
1039 * This function is used for "hashed" keys to make sure the found directory or
1040 * extended attribute entry node is what was looked for. It is used when the
1041 * flash address of the right node is known (@lnum:@offs) which makes it much
1042 * easier to resolve collisions (no need to read entries and match full
1043 * names). This function returns %1 and sets @zn and @n if the collision is
1044 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1045 * previous directory entry. Otherwise a negative error code is returned.
1047 static int resolve_collision_directly(struct ubifs_info *c,
1048 const union ubifs_key *key,
1049 struct ubifs_znode **zn, int *n,
1052 struct ubifs_znode *znode;
1057 if (matches_position(&znode->zbranch[nn], lnum, offs))
1062 err = tnc_prev(c, &znode, &nn);
1067 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1069 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1080 err = tnc_next(c, &znode, &nn);
1085 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1089 if (matches_position(&znode->zbranch[nn], lnum, offs))
1095 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1096 * @c: UBIFS file-system description object
1097 * @znode: znode to dirty
1099 * If we do not have a unique key that resides in a znode, then we cannot
1100 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1101 * This function records the path back to the last dirty ancestor, and then
1102 * dirties the znodes on that path.
1104 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1105 struct ubifs_znode *znode)
1107 struct ubifs_znode *zp;
1108 int *path = c->bottom_up_buf, p = 0;
1110 ubifs_assert(c->zroot.znode);
1111 ubifs_assert(znode);
1112 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1113 kfree(c->bottom_up_buf);
1114 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
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(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(path[p - 1] >= 0);
1144 ubifs_assert(path[p - 1] < zp->child_cnt);
1145 zbr = &zp->zbranch[path[--p]];
1146 znode = dirty_cow_znode(c, zbr);
1148 ubifs_assert(znode == c->zroot.znode);
1149 znode = dirty_cow_znode(c, &c->zroot);
1151 if (IS_ERR(znode) || !p)
1153 ubifs_assert(path[p - 1] >= 0);
1154 ubifs_assert(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 or %-1
1175 * is stored in @n; In this case calling tnc_next() is mandatory.
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 unsigned long time = get_seconds();
1190 dbg_tnck(key, "search key ");
1191 ubifs_assert(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 unsigned long time = 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(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1668 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1669 ubifs_assert(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 len = le32_to_cpu(ch->len);
1723 if (len != zbr->len) {
1724 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1728 /* Make sure the key of the read node is correct */
1729 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1730 if (!keys_eq(c, &zbr->key, &key1)) {
1731 ubifs_err(c, "bad key in node at LEB %d:%d",
1732 zbr->lnum, zbr->offs);
1733 dbg_tnck(&zbr->key, "looked for key ");
1734 dbg_tnck(&key1, "found node's key ");
1743 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1744 ubifs_dump_node(c, buf);
1750 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1751 * @c: UBIFS file-system description object
1752 * @bu: bulk-read parameters and results
1754 * This functions reads and validates the data nodes that were identified by the
1755 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1756 * -EAGAIN to indicate a race with GC, or another negative error code on
1759 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1761 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1762 struct ubifs_wbuf *wbuf;
1765 len = bu->zbranch[bu->cnt - 1].offs;
1766 len += bu->zbranch[bu->cnt - 1].len - offs;
1767 if (len > bu->buf_len) {
1768 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1773 wbuf = ubifs_get_wbuf(c, lnum);
1775 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1777 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1779 /* Check for a race with GC */
1780 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1783 if (err && err != -EBADMSG) {
1784 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1787 dbg_tnck(&bu->key, "key ");
1791 /* Validate the nodes read */
1793 for (i = 0; i < bu->cnt; i++) {
1794 err = validate_data_node(c, buf, &bu->zbranch[i]);
1797 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1804 * do_lookup_nm- look up a "hashed" node.
1805 * @c: UBIFS file-system description object
1806 * @key: node key to lookup
1807 * @node: the node is returned here
1810 * This function looks up and reads a node which contains name hash in the key.
1811 * Since the hash may have collisions, there may be many nodes with the same
1812 * key, so we have to sequentially look to all of them until the needed one is
1813 * found. This function returns zero in case of success, %-ENOENT if the node
1814 * was not found, and a negative error code in case of failure.
1816 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1817 void *node, const struct fscrypt_name *nm)
1820 struct ubifs_znode *znode;
1822 dbg_tnck(key, "key ");
1823 mutex_lock(&c->tnc_mutex);
1824 found = ubifs_lookup_level0(c, key, &znode, &n);
1828 } else if (found < 0) {
1833 ubifs_assert(n >= 0);
1835 err = resolve_collision(c, key, &znode, &n, nm);
1836 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1837 if (unlikely(err < 0))
1844 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1847 mutex_unlock(&c->tnc_mutex);
1852 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1853 * @c: UBIFS file-system description object
1854 * @key: node key to lookup
1855 * @node: the node is returned here
1858 * This function looks up and reads a node which contains name hash in the key.
1859 * Since the hash may have collisions, there may be many nodes with the same
1860 * key, so we have to sequentially look to all of them until the needed one is
1861 * found. This function returns zero in case of success, %-ENOENT if the node
1862 * was not found, and a negative error code in case of failure.
1864 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1865 void *node, const struct fscrypt_name *nm)
1868 const struct ubifs_dent_node *dent = node;
1871 * We assume that in most of the cases there are no name collisions and
1872 * 'ubifs_tnc_lookup()' returns us the right direntry.
1874 err = ubifs_tnc_lookup(c, key, node);
1878 len = le16_to_cpu(dent->nlen);
1879 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1883 * Unluckily, there are hash collisions and we have to iterate over
1884 * them look at each direntry with colliding name hash sequentially.
1887 return do_lookup_nm(c, key, node, nm);
1890 static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1891 struct ubifs_dent_node *dent, uint32_t cookie,
1892 struct ubifs_znode **zn, int *n, int exact)
1895 struct ubifs_znode *znode = *zn;
1896 struct ubifs_zbranch *zbr;
1897 union ubifs_key *dkey;
1900 err = tnc_next(c, &znode, n);
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(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, err);
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(znode->parent);
2009 ubifs_assert(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 * @znode: znode into which to insert
2027 * @zbr: zbranch to insert
2028 * @n: slot number to insert to
2030 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2031 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2032 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2033 * slot, zbranches starting from @n have to be moved right.
2035 static void insert_zbranch(struct ubifs_znode *znode,
2036 const struct ubifs_zbranch *zbr, int n)
2040 ubifs_assert(ubifs_zn_dirty(znode));
2043 for (i = znode->child_cnt; i > n; i--) {
2044 znode->zbranch[i] = znode->zbranch[i - 1];
2045 if (znode->zbranch[i].znode)
2046 znode->zbranch[i].znode->iip = i;
2049 zbr->znode->iip = n;
2051 for (i = znode->child_cnt; i > n; i--)
2052 znode->zbranch[i] = znode->zbranch[i - 1];
2054 znode->zbranch[n] = *zbr;
2055 znode->child_cnt += 1;
2058 * After inserting at slot zero, the lower bound of the key range of
2059 * this znode may have changed. If this znode is subsequently split
2060 * then the upper bound of the key range may change, and furthermore
2061 * it could change to be lower than the original lower bound. If that
2062 * happens, then it will no longer be possible to find this znode in the
2063 * TNC using the key from the index node on flash. That is bad because
2064 * if it is not found, we will assume it is obsolete and may overwrite
2065 * it. Then if there is an unclean unmount, we will start using the
2066 * old index which will be broken.
2068 * So we first mark znodes that have insertions at slot zero, and then
2069 * if they are split we add their lnum/offs to the old_idx tree.
2076 * tnc_insert - insert a node into TNC.
2077 * @c: UBIFS file-system description object
2078 * @znode: znode to insert into
2079 * @zbr: branch to insert
2080 * @n: slot number to insert new zbranch to
2082 * This function inserts a new node described by @zbr into znode @znode. If
2083 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2084 * are splat as well if needed. Returns zero in case of success or a negative
2085 * error code in case of failure.
2087 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2088 struct ubifs_zbranch *zbr, int n)
2090 struct ubifs_znode *zn, *zi, *zp;
2091 int i, keep, move, appending = 0;
2092 union ubifs_key *key = &zbr->key, *key1;
2094 ubifs_assert(n >= 0 && n <= c->fanout);
2096 /* Implement naive insert for now */
2099 if (znode->child_cnt < c->fanout) {
2100 ubifs_assert(n != c->fanout);
2101 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2103 insert_zbranch(znode, zbr, n);
2105 /* Ensure parent's key is correct */
2106 if (n == 0 && zp && znode->iip == 0)
2107 correct_parent_keys(c, znode);
2113 * Unfortunately, @znode does not have more empty slots and we have to
2116 dbg_tnck(key, "splitting level %d, key ", znode->level);
2120 * We can no longer be sure of finding this znode by key, so we
2121 * record it in the old_idx tree.
2123 ins_clr_old_idx_znode(c, znode);
2125 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2129 zn->level = znode->level;
2131 /* Decide where to split */
2132 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2133 /* Try not to split consecutive data keys */
2134 if (n == c->fanout) {
2135 key1 = &znode->zbranch[n - 1].key;
2136 if (key_inum(c, key1) == key_inum(c, key) &&
2137 key_type(c, key1) == UBIFS_DATA_KEY)
2141 } else if (appending && n != c->fanout) {
2142 /* Try not to split consecutive data keys */
2145 if (n >= (c->fanout + 1) / 2) {
2146 key1 = &znode->zbranch[0].key;
2147 if (key_inum(c, key1) == key_inum(c, key) &&
2148 key_type(c, key1) == UBIFS_DATA_KEY) {
2149 key1 = &znode->zbranch[n].key;
2150 if (key_inum(c, key1) != key_inum(c, key) ||
2151 key_type(c, key1) != UBIFS_DATA_KEY) {
2153 move = c->fanout - keep;
2165 keep = (c->fanout + 1) / 2;
2166 move = c->fanout - keep;
2170 * Although we don't at present, we could look at the neighbors and see
2171 * if we can move some zbranches there.
2175 /* Insert into existing znode */
2180 /* Insert into new znode */
2185 zbr->znode->parent = zn;
2190 __set_bit(DIRTY_ZNODE, &zn->flags);
2191 atomic_long_inc(&c->dirty_zn_cnt);
2193 zn->child_cnt = move;
2194 znode->child_cnt = keep;
2196 dbg_tnc("moving %d, keeping %d", move, keep);
2199 for (i = 0; i < move; i++) {
2200 zn->zbranch[i] = znode->zbranch[keep + i];
2203 if (zn->zbranch[i].znode) {
2204 zn->zbranch[i].znode->parent = zn;
2205 zn->zbranch[i].znode->iip = i;
2209 /* Insert new key and branch */
2210 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2212 insert_zbranch(zi, zbr, n);
2214 /* Insert new znode (produced by spitting) into the parent */
2216 if (n == 0 && zi == znode && znode->iip == 0)
2217 correct_parent_keys(c, znode);
2219 /* Locate insertion point */
2222 /* Tail recursion */
2223 zbr->key = zn->zbranch[0].key;
2233 /* We have to split root znode */
2234 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2236 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2241 zi->level = znode->level + 1;
2243 __set_bit(DIRTY_ZNODE, &zi->flags);
2244 atomic_long_inc(&c->dirty_zn_cnt);
2246 zi->zbranch[0].key = znode->zbranch[0].key;
2247 zi->zbranch[0].znode = znode;
2248 zi->zbranch[0].lnum = c->zroot.lnum;
2249 zi->zbranch[0].offs = c->zroot.offs;
2250 zi->zbranch[0].len = c->zroot.len;
2251 zi->zbranch[1].key = zn->zbranch[0].key;
2252 zi->zbranch[1].znode = zn;
2257 c->zroot.znode = zi;
2268 * ubifs_tnc_add - add a node to TNC.
2269 * @c: UBIFS file-system description object
2271 * @lnum: LEB number of node
2272 * @offs: node offset
2275 * This function adds a node with key @key to TNC. The node may be new or it may
2276 * obsolete some existing one. Returns %0 on success or negative error code on
2279 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2282 int found, n, err = 0;
2283 struct ubifs_znode *znode;
2285 mutex_lock(&c->tnc_mutex);
2286 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2287 found = lookup_level0_dirty(c, key, &znode, &n);
2289 struct ubifs_zbranch zbr;
2295 key_copy(c, key, &zbr.key);
2296 err = tnc_insert(c, znode, &zbr, n + 1);
2297 } else if (found == 1) {
2298 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2301 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2308 err = dbg_check_tnc(c, 0);
2309 mutex_unlock(&c->tnc_mutex);
2315 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2316 * @c: UBIFS file-system description object
2318 * @old_lnum: LEB number of old node
2319 * @old_offs: old node offset
2320 * @lnum: LEB number of node
2321 * @offs: node offset
2324 * This function replaces a node with key @key in the TNC only if the old node
2325 * is found. This function is called by garbage collection when node are moved.
2326 * Returns %0 on success or negative error code on failure.
2328 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2329 int old_lnum, int old_offs, int lnum, int offs, int len)
2331 int found, n, err = 0;
2332 struct ubifs_znode *znode;
2334 mutex_lock(&c->tnc_mutex);
2335 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2336 old_offs, lnum, offs, len);
2337 found = lookup_level0_dirty(c, key, &znode, &n);
2344 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2347 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2349 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2356 } else if (is_hash_key(c, key)) {
2357 found = resolve_collision_directly(c, key, &znode, &n,
2358 old_lnum, old_offs);
2359 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2360 found, znode, n, old_lnum, old_offs);
2367 /* Ensure the znode is dirtied */
2368 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2369 znode = dirty_cow_bottom_up(c, znode);
2370 if (IS_ERR(znode)) {
2371 err = PTR_ERR(znode);
2375 zbr = &znode->zbranch[n];
2377 err = ubifs_add_dirt(c, zbr->lnum,
2389 err = ubifs_add_dirt(c, lnum, len);
2392 err = dbg_check_tnc(c, 0);
2395 mutex_unlock(&c->tnc_mutex);
2400 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2401 * @c: UBIFS file-system description object
2403 * @lnum: LEB number of node
2404 * @offs: node offset
2408 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2409 * may have collisions, like directory entry keys.
2411 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2412 int lnum, int offs, int len,
2413 const struct fscrypt_name *nm)
2415 int found, n, err = 0;
2416 struct ubifs_znode *znode;
2418 mutex_lock(&c->tnc_mutex);
2419 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2420 found = lookup_level0_dirty(c, key, &znode, &n);
2428 found = fallible_resolve_collision(c, key, &znode, &n,
2431 found = resolve_collision(c, key, &znode, &n, nm);
2432 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2438 /* Ensure the znode is dirtied */
2439 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2440 znode = dirty_cow_bottom_up(c, znode);
2441 if (IS_ERR(znode)) {
2442 err = PTR_ERR(znode);
2448 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2451 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2460 struct ubifs_zbranch zbr;
2466 key_copy(c, key, &zbr.key);
2467 err = tnc_insert(c, znode, &zbr, n + 1);
2472 * We did not find it in the index so there may be a
2473 * dangling branch still in the index. So we remove it
2474 * by passing 'ubifs_tnc_remove_nm()' the same key but
2475 * an unmatchable name.
2477 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2479 err = dbg_check_tnc(c, 0);
2480 mutex_unlock(&c->tnc_mutex);
2483 return ubifs_tnc_remove_nm(c, key, &noname);
2489 err = dbg_check_tnc(c, 0);
2490 mutex_unlock(&c->tnc_mutex);
2495 * tnc_delete - delete a znode form TNC.
2496 * @c: UBIFS file-system description object
2497 * @znode: znode to delete from
2498 * @n: zbranch slot number to delete
2500 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2501 * case of success and a negative error code in case of failure.
2503 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2505 struct ubifs_zbranch *zbr;
2506 struct ubifs_znode *zp;
2509 /* Delete without merge for now */
2510 ubifs_assert(znode->level == 0);
2511 ubifs_assert(n >= 0 && n < c->fanout);
2512 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2514 zbr = &znode->zbranch[n];
2517 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2519 ubifs_dump_znode(c, znode);
2523 /* We do not "gap" zbranch slots */
2524 for (i = n; i < znode->child_cnt - 1; i++)
2525 znode->zbranch[i] = znode->zbranch[i + 1];
2526 znode->child_cnt -= 1;
2528 if (znode->child_cnt > 0)
2532 * This was the last zbranch, we have to delete this znode from the
2537 ubifs_assert(!ubifs_zn_obsolete(znode));
2538 ubifs_assert(ubifs_zn_dirty(znode));
2543 atomic_long_dec(&c->dirty_zn_cnt);
2545 err = insert_old_idx_znode(c, znode);
2550 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2551 atomic_long_inc(&c->clean_zn_cnt);
2552 atomic_long_inc(&ubifs_clean_zn_cnt);
2556 } while (znode->child_cnt == 1); /* while removing last child */
2558 /* Remove from znode, entry n - 1 */
2559 znode->child_cnt -= 1;
2560 ubifs_assert(znode->level != 0);
2561 for (i = n; i < znode->child_cnt; i++) {
2562 znode->zbranch[i] = znode->zbranch[i + 1];
2563 if (znode->zbranch[i].znode)
2564 znode->zbranch[i].znode->iip = i;
2568 * If this is the root and it has only 1 child then
2569 * collapse the tree.
2571 if (!znode->parent) {
2572 while (znode->child_cnt == 1 && znode->level != 0) {
2574 zbr = &znode->zbranch[0];
2575 znode = get_znode(c, znode, 0);
2577 return PTR_ERR(znode);
2578 znode = dirty_cow_znode(c, zbr);
2580 return PTR_ERR(znode);
2581 znode->parent = NULL;
2584 err = insert_old_idx(c, c->zroot.lnum,
2589 c->zroot.lnum = zbr->lnum;
2590 c->zroot.offs = zbr->offs;
2591 c->zroot.len = zbr->len;
2592 c->zroot.znode = znode;
2593 ubifs_assert(!ubifs_zn_obsolete(zp));
2594 ubifs_assert(ubifs_zn_dirty(zp));
2595 atomic_long_dec(&c->dirty_zn_cnt);
2598 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2599 atomic_long_inc(&c->clean_zn_cnt);
2600 atomic_long_inc(&ubifs_clean_zn_cnt);
2610 * ubifs_tnc_remove - remove an index entry of a node.
2611 * @c: UBIFS file-system description object
2614 * Returns %0 on success or negative error code on failure.
2616 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2618 int found, n, err = 0;
2619 struct ubifs_znode *znode;
2621 mutex_lock(&c->tnc_mutex);
2622 dbg_tnck(key, "key ");
2623 found = lookup_level0_dirty(c, key, &znode, &n);
2629 err = tnc_delete(c, znode, n);
2631 err = dbg_check_tnc(c, 0);
2634 mutex_unlock(&c->tnc_mutex);
2639 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2640 * @c: UBIFS file-system description object
2642 * @nm: directory entry name
2644 * Returns %0 on success or negative error code on failure.
2646 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2647 const struct fscrypt_name *nm)
2650 struct ubifs_znode *znode;
2652 mutex_lock(&c->tnc_mutex);
2653 dbg_tnck(key, "key ");
2654 err = lookup_level0_dirty(c, key, &znode, &n);
2660 err = fallible_resolve_collision(c, key, &znode, &n,
2663 err = resolve_collision(c, key, &znode, &n, nm);
2664 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2668 /* Ensure the znode is dirtied */
2669 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2670 znode = dirty_cow_bottom_up(c, znode);
2671 if (IS_ERR(znode)) {
2672 err = PTR_ERR(znode);
2676 err = tnc_delete(c, znode, n);
2682 err = dbg_check_tnc(c, 0);
2683 mutex_unlock(&c->tnc_mutex);
2688 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2689 * @c: UBIFS file-system description object
2691 * @cookie: node cookie for collision resolution
2693 * Returns %0 on success or negative error code on failure.
2695 int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2699 struct ubifs_znode *znode;
2700 struct ubifs_dent_node *dent;
2701 struct ubifs_zbranch *zbr;
2703 if (!c->double_hash)
2706 mutex_lock(&c->tnc_mutex);
2707 err = lookup_level0_dirty(c, key, &znode, &n);
2711 zbr = &znode->zbranch[n];
2712 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2718 err = tnc_read_hashed_node(c, zbr, dent);
2722 /* If the cookie does not match, we're facing a hash collision. */
2723 if (le32_to_cpu(dent->cookie) != cookie) {
2724 union ubifs_key start_key;
2726 lowest_dent_key(c, &start_key, key_inum(c, key));
2728 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2729 if (unlikely(err < 0))
2732 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
2737 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2738 znode = dirty_cow_bottom_up(c, znode);
2739 if (IS_ERR(znode)) {
2740 err = PTR_ERR(znode);
2744 err = tnc_delete(c, znode, n);
2750 err = dbg_check_tnc(c, 0);
2751 mutex_unlock(&c->tnc_mutex);
2756 * key_in_range - determine if a key falls within a range of keys.
2757 * @c: UBIFS file-system description object
2758 * @key: key to check
2759 * @from_key: lowest key in range
2760 * @to_key: highest key in range
2762 * This function returns %1 if the key is in range and %0 otherwise.
2764 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2765 union ubifs_key *from_key, union ubifs_key *to_key)
2767 if (keys_cmp(c, key, from_key) < 0)
2769 if (keys_cmp(c, key, to_key) > 0)
2775 * ubifs_tnc_remove_range - remove index entries in range.
2776 * @c: UBIFS file-system description object
2777 * @from_key: lowest key to remove
2778 * @to_key: highest key to remove
2780 * This function removes index entries starting at @from_key and ending at
2781 * @to_key. This function returns zero in case of success and a negative error
2782 * code in case of failure.
2784 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2785 union ubifs_key *to_key)
2787 int i, n, k, err = 0;
2788 struct ubifs_znode *znode;
2789 union ubifs_key *key;
2791 mutex_lock(&c->tnc_mutex);
2793 /* Find first level 0 znode that contains keys to remove */
2794 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2801 err = tnc_next(c, &znode, &n);
2802 if (err == -ENOENT) {
2808 key = &znode->zbranch[n].key;
2809 if (!key_in_range(c, key, from_key, to_key)) {
2815 /* Ensure the znode is dirtied */
2816 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2817 znode = dirty_cow_bottom_up(c, znode);
2818 if (IS_ERR(znode)) {
2819 err = PTR_ERR(znode);
2824 /* Remove all keys in range except the first */
2825 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2826 key = &znode->zbranch[i].key;
2827 if (!key_in_range(c, key, from_key, to_key))
2829 lnc_free(&znode->zbranch[i]);
2830 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2831 znode->zbranch[i].len);
2833 ubifs_dump_znode(c, znode);
2836 dbg_tnck(key, "removing key ");
2839 for (i = n + 1 + k; i < znode->child_cnt; i++)
2840 znode->zbranch[i - k] = znode->zbranch[i];
2841 znode->child_cnt -= k;
2844 /* Now delete the first */
2845 err = tnc_delete(c, znode, n);
2852 err = dbg_check_tnc(c, 0);
2853 mutex_unlock(&c->tnc_mutex);
2858 * ubifs_tnc_remove_ino - remove an inode from TNC.
2859 * @c: UBIFS file-system description object
2860 * @inum: inode number to remove
2862 * This function remove inode @inum and all the extended attributes associated
2863 * with the anode from TNC and returns zero in case of success or a negative
2864 * error code in case of failure.
2866 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2868 union ubifs_key key1, key2;
2869 struct ubifs_dent_node *xent, *pxent = NULL;
2870 struct fscrypt_name nm = {0};
2872 dbg_tnc("ino %lu", (unsigned long)inum);
2875 * Walk all extended attribute entries and remove them together with
2876 * corresponding extended attribute inodes.
2878 lowest_xent_key(c, &key1, inum);
2883 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2885 err = PTR_ERR(xent);
2891 xattr_inum = le64_to_cpu(xent->inum);
2892 dbg_tnc("xent '%s', ino %lu", xent->name,
2893 (unsigned long)xattr_inum);
2895 ubifs_evict_xattr_inode(c, xattr_inum);
2897 fname_name(&nm) = xent->name;
2898 fname_len(&nm) = le16_to_cpu(xent->nlen);
2899 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2905 lowest_ino_key(c, &key1, xattr_inum);
2906 highest_ino_key(c, &key2, xattr_inum);
2907 err = ubifs_tnc_remove_range(c, &key1, &key2);
2915 key_read(c, &xent->key, &key1);
2919 lowest_ino_key(c, &key1, inum);
2920 highest_ino_key(c, &key2, inum);
2922 return ubifs_tnc_remove_range(c, &key1, &key2);
2926 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2927 * @c: UBIFS file-system description object
2928 * @key: key of last entry
2929 * @nm: name of last entry found or %NULL
2931 * This function finds and reads the next directory or extended attribute entry
2932 * after the given key (@key) if there is one. @nm is used to resolve
2935 * If the name of the current entry is not known and only the key is known,
2936 * @nm->name has to be %NULL. In this case the semantics of this function is a
2937 * little bit different and it returns the entry corresponding to this key, not
2938 * the next one. If the key was not found, the closest "right" entry is
2941 * If the fist entry has to be found, @key has to contain the lowest possible
2942 * key value for this inode and @name has to be %NULL.
2944 * This function returns the found directory or extended attribute entry node
2945 * in case of success, %-ENOENT is returned if no entry was found, and a
2946 * negative error code is returned in case of failure.
2948 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2949 union ubifs_key *key,
2950 const struct fscrypt_name *nm)
2952 int n, err, type = key_type(c, key);
2953 struct ubifs_znode *znode;
2954 struct ubifs_dent_node *dent;
2955 struct ubifs_zbranch *zbr;
2956 union ubifs_key *dkey;
2958 dbg_tnck(key, "key ");
2959 ubifs_assert(is_hash_key(c, key));
2961 mutex_lock(&c->tnc_mutex);
2962 err = ubifs_lookup_level0(c, key, &znode, &n);
2963 if (unlikely(err < 0))
2966 if (fname_len(nm) > 0) {
2968 /* Handle collisions */
2970 err = fallible_resolve_collision(c, key, &znode, &n,
2973 err = resolve_collision(c, key, &znode, &n, nm);
2974 dbg_tnc("rc returned %d, znode %p, n %d",
2976 if (unlikely(err < 0))
2980 /* Now find next entry */
2981 err = tnc_next(c, &znode, &n);
2986 * The full name of the entry was not given, in which case the
2987 * behavior of this function is a little different and it
2988 * returns current entry, not the next one.
2992 * However, the given key does not exist in the TNC
2993 * tree and @znode/@n variables contain the closest
2994 * "preceding" element. Switch to the next one.
2996 err = tnc_next(c, &znode, &n);
3002 zbr = &znode->zbranch[n];
3003 dent = kmalloc(zbr->len, GFP_NOFS);
3004 if (unlikely(!dent)) {
3010 * The above 'tnc_next()' call could lead us to the next inode, check
3014 if (key_inum(c, dkey) != key_inum(c, key) ||
3015 key_type(c, dkey) != type) {
3020 err = tnc_read_hashed_node(c, zbr, dent);
3024 mutex_unlock(&c->tnc_mutex);
3030 mutex_unlock(&c->tnc_mutex);
3031 return ERR_PTR(err);
3035 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3036 * @c: UBIFS file-system description object
3038 * Destroy left-over obsolete znodes from a failed commit.
3040 static void tnc_destroy_cnext(struct ubifs_info *c)
3042 struct ubifs_znode *cnext;
3046 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
3049 struct ubifs_znode *znode = cnext;
3051 cnext = cnext->cnext;
3052 if (ubifs_zn_obsolete(znode))
3054 else if (!ubifs_zn_cow(znode)) {
3056 * Don't forget to update clean znode count after
3057 * committing failed, because ubifs will check this
3058 * count while closing tnc. Non-obsolete znode could
3059 * be re-dirtied during committing process, so dirty
3060 * flag is untrustable. The flag 'COW_ZNODE' is set
3061 * for each dirty znode before committing, and it is
3062 * cleared as long as the znode become clean, so we
3063 * can statistic clean znode count according to this
3066 atomic_long_inc(&c->clean_zn_cnt);
3067 atomic_long_inc(&ubifs_clean_zn_cnt);
3069 } while (cnext && cnext != c->cnext);
3073 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3074 * @c: UBIFS file-system description object
3076 void ubifs_tnc_close(struct ubifs_info *c)
3078 tnc_destroy_cnext(c);
3079 if (c->zroot.znode) {
3082 n = atomic_long_read(&c->clean_zn_cnt);
3083 freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
3084 ubifs_assert(freed == n);
3085 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3093 * left_znode - get the znode to the left.
3094 * @c: UBIFS file-system description object
3097 * This function returns a pointer to the znode to the left of @znode or NULL if
3098 * there is not one. A negative error code is returned on failure.
3100 static struct ubifs_znode *left_znode(struct ubifs_info *c,
3101 struct ubifs_znode *znode)
3103 int level = znode->level;
3106 int n = znode->iip - 1;
3108 /* Go up until we can go left */
3109 znode = znode->parent;
3113 /* Now go down the rightmost branch to 'level' */
3114 znode = get_znode(c, znode, n);
3117 while (znode->level != level) {
3118 n = znode->child_cnt - 1;
3119 znode = get_znode(c, znode, n);
3130 * right_znode - get the znode to the right.
3131 * @c: UBIFS file-system description object
3134 * This function returns a pointer to the znode to the right of @znode or NULL
3135 * if there is not one. A negative error code is returned on failure.
3137 static struct ubifs_znode *right_znode(struct ubifs_info *c,
3138 struct ubifs_znode *znode)
3140 int level = znode->level;
3143 int n = znode->iip + 1;
3145 /* Go up until we can go right */
3146 znode = znode->parent;
3149 if (n < znode->child_cnt) {
3150 /* Now go down the leftmost branch to 'level' */
3151 znode = get_znode(c, znode, n);
3154 while (znode->level != level) {
3155 znode = get_znode(c, znode, 0);
3166 * lookup_znode - find a particular indexing node from TNC.
3167 * @c: UBIFS file-system description object
3168 * @key: index node key to lookup
3169 * @level: index node level
3170 * @lnum: index node LEB number
3171 * @offs: index node offset
3173 * This function searches an indexing node by its first key @key and its
3174 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3175 * nodes it traverses to TNC. This function is called for indexing nodes which
3176 * were found on the media by scanning, for example when garbage-collecting or
3177 * when doing in-the-gaps commit. This means that the indexing node which is
3178 * looked for does not have to have exactly the same leftmost key @key, because
3179 * the leftmost key may have been changed, in which case TNC will contain a
3180 * dirty znode which still refers the same @lnum:@offs. This function is clever
3181 * enough to recognize such indexing nodes.
3183 * Note, if a znode was deleted or changed too much, then this function will
3184 * not find it. For situations like this UBIFS has the old index RB-tree
3185 * (indexed by @lnum:@offs).
3187 * This function returns a pointer to the znode found or %NULL if it is not
3188 * found. A negative error code is returned on failure.
3190 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3191 union ubifs_key *key, int level,
3194 struct ubifs_znode *znode, *zn;
3197 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
3200 * The arguments have probably been read off flash, so don't assume
3204 return ERR_PTR(-EINVAL);
3206 /* Get the root znode */
3207 znode = c->zroot.znode;
3209 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3213 /* Check if it is the one we are looking for */
3214 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3216 /* Descend to the parent level i.e. (level + 1) */
3217 if (level >= znode->level)
3220 ubifs_search_zbranch(c, znode, key, &n);
3223 * We reached a znode where the leftmost key is greater
3224 * than the key we are searching for. This is the same
3225 * situation as the one described in a huge comment at
3226 * the end of the 'ubifs_lookup_level0()' function. And
3227 * for exactly the same reasons we have to try to look
3228 * left before giving up.
3230 znode = left_znode(c, znode);
3235 ubifs_search_zbranch(c, znode, key, &n);
3236 ubifs_assert(n >= 0);
3238 if (znode->level == level + 1)
3240 znode = get_znode(c, znode, n);
3244 /* Check if the child is the one we are looking for */
3245 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3246 return get_znode(c, znode, n);
3247 /* If the key is unique, there is nowhere else to look */
3248 if (!is_hash_key(c, key))
3251 * The key is not unique and so may be also in the znodes to either
3258 /* Move one branch to the left */
3262 znode = left_znode(c, znode);
3267 n = znode->child_cnt - 1;
3270 if (znode->zbranch[n].lnum == lnum &&
3271 znode->zbranch[n].offs == offs)
3272 return get_znode(c, znode, n);
3273 /* Stop if the key is less than the one we are looking for */
3274 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3277 /* Back to the middle */
3282 /* Move one branch to the right */
3283 if (++n >= znode->child_cnt) {
3284 znode = right_znode(c, znode);
3292 if (znode->zbranch[n].lnum == lnum &&
3293 znode->zbranch[n].offs == offs)
3294 return get_znode(c, znode, n);
3295 /* Stop if the key is greater than the one we are looking for */
3296 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3303 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3304 * @c: UBIFS file-system description object
3305 * @key: key of index node
3306 * @level: index node level
3307 * @lnum: LEB number of index node
3308 * @offs: offset of index node
3310 * This function returns %0 if the index node is not referred to in the TNC, %1
3311 * if the index node is referred to in the TNC and the corresponding znode is
3312 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3313 * znode is clean, and a negative error code in case of failure.
3315 * Note, the @key argument has to be the key of the first child. Also note,
3316 * this function relies on the fact that 0:0 is never a valid LEB number and
3317 * offset for a main-area node.
3319 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3322 struct ubifs_znode *znode;
3324 znode = lookup_znode(c, key, level, lnum, offs);
3328 return PTR_ERR(znode);
3330 return ubifs_zn_dirty(znode) ? 1 : 2;
3334 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3335 * @c: UBIFS file-system description object
3337 * @lnum: node LEB number
3338 * @offs: node offset
3340 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3341 * not, and a negative error code in case of failure.
3343 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3344 * and offset for a main-area node.
3346 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3349 struct ubifs_zbranch *zbr;
3350 struct ubifs_znode *znode, *zn;
3351 int n, found, err, nn;
3352 const int unique = !is_hash_key(c, key);
3354 found = ubifs_lookup_level0(c, key, &znode, &n);
3356 return found; /* Error code */
3359 zbr = &znode->zbranch[n];
3360 if (lnum == zbr->lnum && offs == zbr->offs)
3361 return 1; /* Found it */
3365 * Because the key is not unique, we have to look left
3372 err = tnc_prev(c, &znode, &n);
3377 if (keys_cmp(c, key, &znode->zbranch[n].key))
3379 zbr = &znode->zbranch[n];
3380 if (lnum == zbr->lnum && offs == zbr->offs)
3381 return 1; /* Found it */
3387 err = tnc_next(c, &znode, &n);
3393 if (keys_cmp(c, key, &znode->zbranch[n].key))
3395 zbr = &znode->zbranch[n];
3396 if (lnum == zbr->lnum && offs == zbr->offs)
3397 return 1; /* Found it */
3403 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3404 * @c: UBIFS file-system description object
3406 * @level: index node level (if it is an index node)
3407 * @lnum: node LEB number
3408 * @offs: node offset
3409 * @is_idx: non-zero if the node is an index node
3411 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3412 * negative error code in case of failure. For index nodes, @key has to be the
3413 * key of the first child. An index node is considered to be in the TNC only if
3414 * the corresponding znode is clean or has not been loaded.
3416 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3417 int lnum, int offs, int is_idx)
3421 mutex_lock(&c->tnc_mutex);
3423 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3427 /* The index node was found but it was dirty */
3430 /* The index node was found and it was clean */
3435 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3438 mutex_unlock(&c->tnc_mutex);
3443 * ubifs_dirty_idx_node - dirty an index node.
3444 * @c: UBIFS file-system description object
3445 * @key: index node key
3446 * @level: index node level
3447 * @lnum: index node LEB number
3448 * @offs: index node offset
3450 * This function loads and dirties an index node so that it can be garbage
3451 * collected. The @key argument has to be the key of the first child. This
3452 * function relies on the fact that 0:0 is never a valid LEB number and offset
3453 * for a main-area node. Returns %0 on success and a negative error code on
3456 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3459 struct ubifs_znode *znode;
3462 mutex_lock(&c->tnc_mutex);
3463 znode = lookup_znode(c, key, level, lnum, offs);
3466 if (IS_ERR(znode)) {
3467 err = PTR_ERR(znode);
3470 znode = dirty_cow_bottom_up(c, znode);
3471 if (IS_ERR(znode)) {
3472 err = PTR_ERR(znode);
3477 mutex_unlock(&c->tnc_mutex);
3482 * dbg_check_inode_size - check if inode size is correct.
3483 * @c: UBIFS file-system description object
3484 * @inum: inode number
3487 * This function makes sure that the inode size (@size) is correct and it does
3488 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3489 * if it has a data page beyond @size, and other negative error code in case of
3492 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3496 union ubifs_key from_key, to_key, *key;
3497 struct ubifs_znode *znode;
3500 if (!S_ISREG(inode->i_mode))
3502 if (!dbg_is_chk_gen(c))
3505 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3506 data_key_init(c, &from_key, inode->i_ino, block);
3507 highest_data_key(c, &to_key, inode->i_ino);
3509 mutex_lock(&c->tnc_mutex);
3510 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3519 err = tnc_next(c, &znode, &n);
3520 if (err == -ENOENT) {
3527 ubifs_assert(err == 0);
3528 key = &znode->zbranch[n].key;
3529 if (!key_in_range(c, key, &from_key, &to_key))
3533 block = key_block(c, key);
3534 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3535 (unsigned long)inode->i_ino, size,
3536 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3537 mutex_unlock(&c->tnc_mutex);
3538 ubifs_dump_inode(c, inode);
3543 mutex_unlock(&c->tnc_mutex);