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 ||
484 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
485 node_crc = le32_to_cpu(ch->crc);
490 err = ubifs_node_check_hash(c, buf, zbr->hash);
492 ubifs_bad_hash(c, buf, zbr->hash, lnum, offs);
500 * fallible_read_node - try to read a leaf node.
501 * @c: UBIFS file-system description object
502 * @key: key of node to read
503 * @zbr: position of node
504 * @node: node returned
506 * This function tries to read a node and returns %1 if the node is read, %0
507 * if the node is not present, and a negative error code in the case of error.
509 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
510 struct ubifs_zbranch *zbr, void *node)
514 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
516 ret = try_read_node(c, node, key_type(c, key), zbr);
518 union ubifs_key node_key;
519 struct ubifs_dent_node *dent = node;
521 /* All nodes have key in the same place */
522 key_read(c, &dent->key, &node_key);
523 if (keys_cmp(c, key, &node_key) != 0)
526 if (ret == 0 && c->replaying)
527 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
528 zbr->lnum, zbr->offs, zbr->len);
533 * matches_name - determine if a direntry or xattr entry matches a given name.
534 * @c: UBIFS file-system description object
535 * @zbr: zbranch of dent
538 * This function checks if xentry/direntry referred by zbranch @zbr matches name
539 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
540 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
541 * of failure, a negative error code is returned.
543 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
544 const struct fscrypt_name *nm)
546 struct ubifs_dent_node *dent;
549 /* If possible, match against the dent in the leaf node cache */
551 dent = kmalloc(zbr->len, GFP_NOFS);
555 err = ubifs_tnc_read_node(c, zbr, dent);
559 /* Add the node to the leaf node cache */
560 err = lnc_add_directly(c, zbr, dent);
566 nlen = le16_to_cpu(dent->nlen);
567 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
569 if (nlen == fname_len(nm))
571 else if (nlen < fname_len(nm))
586 * get_znode - get a TNC znode that may not be loaded yet.
587 * @c: UBIFS file-system description object
588 * @znode: parent znode
589 * @n: znode branch slot number
591 * This function returns the znode or a negative error code.
593 static struct ubifs_znode *get_znode(struct ubifs_info *c,
594 struct ubifs_znode *znode, int n)
596 struct ubifs_zbranch *zbr;
598 zbr = &znode->zbranch[n];
602 znode = ubifs_load_znode(c, zbr, znode, n);
607 * tnc_next - find next TNC entry.
608 * @c: UBIFS file-system description object
609 * @zn: znode is passed and returned here
610 * @n: znode branch slot number is passed and returned here
612 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
613 * no next entry, or a negative error code otherwise.
615 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
617 struct ubifs_znode *znode = *zn;
621 if (nn < znode->child_cnt) {
626 struct ubifs_znode *zp;
633 if (nn < znode->child_cnt) {
634 znode = get_znode(c, znode, nn);
636 return PTR_ERR(znode);
637 while (znode->level != 0) {
638 znode = get_znode(c, znode, 0);
640 return PTR_ERR(znode);
652 * tnc_prev - find previous TNC entry.
653 * @c: UBIFS file-system description object
654 * @zn: znode is returned here
655 * @n: znode branch slot number is passed and returned here
657 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
658 * there is no next entry, or a negative error code otherwise.
660 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
662 struct ubifs_znode *znode = *zn;
670 struct ubifs_znode *zp;
678 znode = get_znode(c, znode, nn);
680 return PTR_ERR(znode);
681 while (znode->level != 0) {
682 nn = znode->child_cnt - 1;
683 znode = get_znode(c, znode, nn);
685 return PTR_ERR(znode);
687 nn = znode->child_cnt - 1;
697 * resolve_collision - resolve a collision.
698 * @c: UBIFS file-system description object
699 * @key: key of a directory or extended attribute entry
700 * @zn: znode is returned here
701 * @n: zbranch number is passed and returned here
702 * @nm: name of the entry
704 * This function is called for "hashed" keys to make sure that the found key
705 * really corresponds to the looked up node (directory or extended attribute
706 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
707 * %0 is returned if @nm is not found and @zn and @n are set to the previous
708 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
709 * This means that @n may be set to %-1 if the leftmost key in @zn is the
710 * previous one. A negative error code is returned on failures.
712 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
713 struct ubifs_znode **zn, int *n,
714 const struct fscrypt_name *nm)
718 err = matches_name(c, &(*zn)->zbranch[*n], nm);
719 if (unlikely(err < 0))
721 if (err == NAME_MATCHES)
724 if (err == NAME_GREATER) {
727 err = tnc_prev(c, zn, n);
728 if (err == -ENOENT) {
729 ubifs_assert(c, *n == 0);
735 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
737 * We have found the branch after which we would
738 * like to insert, but inserting in this znode
739 * may still be wrong. Consider the following 3
740 * znodes, in the case where we are resolving a
741 * collision with Key2.
744 * ----------------------
745 * level 1 | Key0 | Key1 |
746 * -----------------------
748 * znode za | | znode zb
749 * ------------ ------------
750 * level 0 | Key0 | | Key2 |
751 * ------------ ------------
753 * The lookup finds Key2 in znode zb. Lets say
754 * there is no match and the name is greater so
755 * we look left. When we find Key0, we end up
756 * here. If we return now, we will insert into
757 * znode za at slot n = 1. But that is invalid
758 * according to the parent's keys. Key2 must
759 * be inserted into znode zb.
761 * Note, this problem is not relevant for the
762 * case when we go right, because
763 * 'tnc_insert()' would correct the parent key.
765 if (*n == (*zn)->child_cnt - 1) {
766 err = tnc_next(c, zn, n);
768 /* Should be impossible */
774 ubifs_assert(c, *n == 0);
779 err = matches_name(c, &(*zn)->zbranch[*n], nm);
782 if (err == NAME_LESS)
784 if (err == NAME_MATCHES)
786 ubifs_assert(c, err == NAME_GREATER);
790 struct ubifs_znode *znode = *zn;
794 err = tnc_next(c, &znode, &nn);
799 if (keys_cmp(c, &znode->zbranch[nn].key, key))
801 err = matches_name(c, &znode->zbranch[nn], nm);
804 if (err == NAME_GREATER)
808 if (err == NAME_MATCHES)
810 ubifs_assert(c, err == NAME_LESS);
816 * fallible_matches_name - determine if a dent matches a given name.
817 * @c: UBIFS file-system description object
818 * @zbr: zbranch of dent
821 * This is a "fallible" version of 'matches_name()' function which does not
822 * panic if the direntry/xentry referred by @zbr does not exist on the media.
824 * This function checks if xentry/direntry referred by zbranch @zbr matches name
825 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
826 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
827 * if xentry/direntry referred by @zbr does not exist on the media. A negative
828 * error code is returned in case of failure.
830 static int fallible_matches_name(struct ubifs_info *c,
831 struct ubifs_zbranch *zbr,
832 const struct fscrypt_name *nm)
834 struct ubifs_dent_node *dent;
837 /* If possible, match against the dent in the leaf node cache */
839 dent = kmalloc(zbr->len, GFP_NOFS);
843 err = fallible_read_node(c, &zbr->key, zbr, dent);
847 /* The node was not present */
851 ubifs_assert(c, err == 1);
853 err = lnc_add_directly(c, zbr, dent);
859 nlen = le16_to_cpu(dent->nlen);
860 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
862 if (nlen == fname_len(nm))
864 else if (nlen < fname_len(nm))
879 * fallible_resolve_collision - resolve a collision even if nodes are missing.
880 * @c: UBIFS file-system description object
882 * @zn: znode is returned here
883 * @n: branch number is passed and returned here
884 * @nm: name of directory entry
885 * @adding: indicates caller is adding a key to the TNC
887 * This is a "fallible" version of the 'resolve_collision()' function which
888 * does not panic if one of the nodes referred to by TNC does not exist on the
889 * media. This may happen when replaying the journal if a deleted node was
890 * Garbage-collected and the commit was not done. A branch that refers to a node
891 * that is not present is called a dangling branch. The following are the return
892 * codes for this function:
893 * o if @nm was found, %1 is returned and @zn and @n are set to the found
895 * o if we are @adding and @nm was not found, %0 is returned;
896 * o if we are not @adding and @nm was not found, but a dangling branch was
897 * found, then %1 is returned and @zn and @n are set to the dangling branch;
898 * o a negative error code is returned in case of failure.
900 static int fallible_resolve_collision(struct ubifs_info *c,
901 const union ubifs_key *key,
902 struct ubifs_znode **zn, int *n,
903 const struct fscrypt_name *nm,
906 struct ubifs_znode *o_znode = NULL, *znode = *zn;
907 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
909 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
910 if (unlikely(cmp < 0))
912 if (cmp == NAME_MATCHES)
914 if (cmp == NOT_ON_MEDIA) {
918 * We are unlucky and hit a dangling branch straight away.
919 * Now we do not really know where to go to find the needed
920 * branch - to the left or to the right. Well, let's try left.
924 unsure = 1; /* Remove a dangling branch wherever it is */
926 if (cmp == NAME_GREATER || unsure) {
929 err = tnc_prev(c, zn, n);
930 if (err == -ENOENT) {
931 ubifs_assert(c, *n == 0);
937 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
938 /* See comments in 'resolve_collision()' */
939 if (*n == (*zn)->child_cnt - 1) {
940 err = tnc_next(c, zn, n);
942 /* Should be impossible */
948 ubifs_assert(c, *n == 0);
953 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
956 if (err == NAME_MATCHES)
958 if (err == NOT_ON_MEDIA) {
965 if (err == NAME_LESS)
972 if (cmp == NAME_LESS || unsure) {
977 err = tnc_next(c, &znode, &nn);
982 if (keys_cmp(c, &znode->zbranch[nn].key, key))
984 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
987 if (err == NAME_GREATER)
991 if (err == NAME_MATCHES)
993 if (err == NOT_ON_MEDIA) {
1000 /* Never match a dangling branch when adding */
1001 if (adding || !o_znode)
1004 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
1005 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1006 o_znode->zbranch[o_n].len);
1013 * matches_position - determine if a zbranch matches a given position.
1014 * @zbr: zbranch of dent
1015 * @lnum: LEB number of dent to match
1016 * @offs: offset of dent to match
1018 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1020 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1022 if (zbr->lnum == lnum && zbr->offs == offs)
1029 * resolve_collision_directly - resolve a collision directly.
1030 * @c: UBIFS file-system description object
1031 * @key: key of directory entry
1032 * @zn: znode is passed and returned here
1033 * @n: zbranch number is passed and returned here
1034 * @lnum: LEB number of dent node to match
1035 * @offs: offset of dent node to match
1037 * This function is used for "hashed" keys to make sure the found directory or
1038 * extended attribute entry node is what was looked for. It is used when the
1039 * flash address of the right node is known (@lnum:@offs) which makes it much
1040 * easier to resolve collisions (no need to read entries and match full
1041 * names). This function returns %1 and sets @zn and @n if the collision is
1042 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1043 * previous directory entry. Otherwise a negative error code is returned.
1045 static int resolve_collision_directly(struct ubifs_info *c,
1046 const union ubifs_key *key,
1047 struct ubifs_znode **zn, int *n,
1050 struct ubifs_znode *znode;
1055 if (matches_position(&znode->zbranch[nn], lnum, offs))
1060 err = tnc_prev(c, &znode, &nn);
1065 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1067 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1078 err = tnc_next(c, &znode, &nn);
1083 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1087 if (matches_position(&znode->zbranch[nn], lnum, offs))
1093 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1094 * @c: UBIFS file-system description object
1095 * @znode: znode to dirty
1097 * If we do not have a unique key that resides in a znode, then we cannot
1098 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1099 * This function records the path back to the last dirty ancestor, and then
1100 * dirties the znodes on that path.
1102 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1103 struct ubifs_znode *znode)
1105 struct ubifs_znode *zp;
1106 int *path = c->bottom_up_buf, p = 0;
1108 ubifs_assert(c, c->zroot.znode);
1109 ubifs_assert(c, znode);
1110 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1111 kfree(c->bottom_up_buf);
1112 c->bottom_up_buf = kmalloc_array(c->zroot.znode->level,
1115 if (!c->bottom_up_buf)
1116 return ERR_PTR(-ENOMEM);
1117 path = c->bottom_up_buf;
1119 if (c->zroot.znode->level) {
1120 /* Go up until parent is dirty */
1128 ubifs_assert(c, p < c->zroot.znode->level);
1130 if (!zp->cnext && ubifs_zn_dirty(znode))
1136 /* Come back down, dirtying as we go */
1138 struct ubifs_zbranch *zbr;
1142 ubifs_assert(c, path[p - 1] >= 0);
1143 ubifs_assert(c, path[p - 1] < zp->child_cnt);
1144 zbr = &zp->zbranch[path[--p]];
1145 znode = dirty_cow_znode(c, zbr);
1147 ubifs_assert(c, znode == c->zroot.znode);
1148 znode = dirty_cow_znode(c, &c->zroot);
1150 if (IS_ERR(znode) || !p)
1152 ubifs_assert(c, path[p - 1] >= 0);
1153 ubifs_assert(c, path[p - 1] < znode->child_cnt);
1154 znode = znode->zbranch[path[p - 1]].znode;
1161 * ubifs_lookup_level0 - search for zero-level znode.
1162 * @c: UBIFS file-system description object
1163 * @key: key to lookup
1164 * @zn: znode is returned here
1165 * @n: znode branch slot number is returned here
1167 * This function looks up the TNC tree and search for zero-level znode which
1168 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1170 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1171 * is returned and slot number of the matched branch is stored in @n;
1172 * o not exact match, which means that zero-level znode does not contain
1173 * @key, then %0 is returned and slot number of the closest branch is stored
1175 * o @key is so small that it is even less than the lowest key of the
1176 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1178 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1179 * function reads corresponding indexing nodes and inserts them to TNC. In
1180 * case of failure, a negative error code is returned.
1182 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1183 struct ubifs_znode **zn, int *n)
1186 struct ubifs_znode *znode;
1187 time64_t time = ktime_get_seconds();
1189 dbg_tnck(key, "search key ");
1190 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
1192 znode = c->zroot.znode;
1193 if (unlikely(!znode)) {
1194 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1196 return PTR_ERR(znode);
1202 struct ubifs_zbranch *zbr;
1204 exact = ubifs_search_zbranch(c, znode, key, n);
1206 if (znode->level == 0)
1211 zbr = &znode->zbranch[*n];
1219 /* znode is not in TNC cache, load it from the media */
1220 znode = ubifs_load_znode(c, zbr, znode, *n);
1222 return PTR_ERR(znode);
1226 if (exact || !is_hash_key(c, key) || *n != -1) {
1227 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1232 * Here is a tricky place. We have not found the key and this is a
1233 * "hashed" key, which may collide. The rest of the code deals with
1234 * situations like this:
1238 * | 3 | 5 | | 6 | 7 | (x)
1240 * Or more a complex example:
1244 * | 1 | 3 | | 5 | 8 |
1246 * | 5 | 5 | | 6 | 7 | (x)
1248 * In the examples, if we are looking for key "5", we may reach nodes
1249 * marked with "(x)". In this case what we have do is to look at the
1250 * left and see if there is "5" key there. If there is, we have to
1253 * Note, this whole situation is possible because we allow to have
1254 * elements which are equivalent to the next key in the parent in the
1255 * children of current znode. For example, this happens if we split a
1256 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1260 * | 3 | 5 | | 5 | 6 | 7 |
1262 * And this becomes what is at the first "picture" after key "5" marked
1263 * with "^" is removed. What could be done is we could prohibit
1264 * splitting in the middle of the colliding sequence. Also, when
1265 * removing the leftmost key, we would have to correct the key of the
1266 * parent node, which would introduce additional complications. Namely,
1267 * if we changed the leftmost key of the parent znode, the garbage
1268 * collector would be unable to find it (GC is doing this when GC'ing
1269 * indexing LEBs). Although we already have an additional RB-tree where
1270 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1271 * after the commit. But anyway, this does not look easy to implement
1272 * so we did not try this.
1274 err = tnc_prev(c, &znode, n);
1275 if (err == -ENOENT) {
1276 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1280 if (unlikely(err < 0))
1282 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1283 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1288 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1294 * lookup_level0_dirty - search for zero-level znode dirtying.
1295 * @c: UBIFS file-system description object
1296 * @key: key to lookup
1297 * @zn: znode is returned here
1298 * @n: znode branch slot number is returned here
1300 * This function looks up the TNC tree and search for zero-level znode which
1301 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1303 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1304 * is returned and slot number of the matched branch is stored in @n;
1305 * o not exact match, which means that zero-level znode does not contain @key
1306 * then %0 is returned and slot number of the closed branch is stored in
1308 * o @key is so small that it is even less than the lowest key of the
1309 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1311 * Additionally all znodes in the path from the root to the located zero-level
1312 * znode are marked as dirty.
1314 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1315 * function reads corresponding indexing nodes and inserts them to TNC. In
1316 * case of failure, a negative error code is returned.
1318 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1319 struct ubifs_znode **zn, int *n)
1322 struct ubifs_znode *znode;
1323 time64_t time = ktime_get_seconds();
1325 dbg_tnck(key, "search and dirty key ");
1327 znode = c->zroot.znode;
1328 if (unlikely(!znode)) {
1329 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1331 return PTR_ERR(znode);
1334 znode = dirty_cow_znode(c, &c->zroot);
1336 return PTR_ERR(znode);
1341 struct ubifs_zbranch *zbr;
1343 exact = ubifs_search_zbranch(c, znode, key, n);
1345 if (znode->level == 0)
1350 zbr = &znode->zbranch[*n];
1354 znode = dirty_cow_znode(c, zbr);
1356 return PTR_ERR(znode);
1360 /* znode is not in TNC cache, load it from the media */
1361 znode = ubifs_load_znode(c, zbr, znode, *n);
1363 return PTR_ERR(znode);
1364 znode = dirty_cow_znode(c, zbr);
1366 return PTR_ERR(znode);
1370 if (exact || !is_hash_key(c, key) || *n != -1) {
1371 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1376 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1379 err = tnc_prev(c, &znode, n);
1380 if (err == -ENOENT) {
1382 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1385 if (unlikely(err < 0))
1387 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1389 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1393 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1394 znode = dirty_cow_bottom_up(c, znode);
1396 return PTR_ERR(znode);
1399 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1405 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1406 * @c: UBIFS file-system description object
1408 * @gc_seq1: garbage collection sequence number
1410 * This function determines if @lnum may have been garbage collected since
1411 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1414 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1416 int gc_seq2, gced_lnum;
1418 gced_lnum = c->gced_lnum;
1420 gc_seq2 = c->gc_seq;
1421 /* Same seq means no GC */
1422 if (gc_seq1 == gc_seq2)
1424 /* Different by more than 1 means we don't know */
1425 if (gc_seq1 + 1 != gc_seq2)
1428 * We have seen the sequence number has increased by 1. Now we need to
1429 * be sure we read the right LEB number, so read it again.
1432 if (gced_lnum != c->gced_lnum)
1434 /* Finally we can check lnum */
1435 if (gced_lnum == lnum)
1441 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1442 * @c: UBIFS file-system description object
1443 * @key: node key to lookup
1444 * @node: the node is returned here
1445 * @lnum: LEB number is returned here
1446 * @offs: offset is returned here
1448 * This function looks up and reads node with key @key. The caller has to make
1449 * sure the @node buffer is large enough to fit the node. Returns zero in case
1450 * of success, %-ENOENT if the node was not found, and a negative error code in
1451 * case of failure. The node location can be returned in @lnum and @offs.
1453 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1454 void *node, int *lnum, int *offs)
1456 int found, n, err, safely = 0, gc_seq1;
1457 struct ubifs_znode *znode;
1458 struct ubifs_zbranch zbr, *zt;
1461 mutex_lock(&c->tnc_mutex);
1462 found = ubifs_lookup_level0(c, key, &znode, &n);
1466 } else if (found < 0) {
1470 zt = &znode->zbranch[n];
1475 if (is_hash_key(c, key)) {
1477 * In this case the leaf node cache gets used, so we pass the
1478 * address of the zbranch and keep the mutex locked
1480 err = tnc_read_hashed_node(c, zt, node);
1484 err = ubifs_tnc_read_node(c, zt, node);
1487 /* Drop the TNC mutex prematurely and race with garbage collection */
1488 zbr = znode->zbranch[n];
1489 gc_seq1 = c->gc_seq;
1490 mutex_unlock(&c->tnc_mutex);
1492 if (ubifs_get_wbuf(c, zbr.lnum)) {
1493 /* We do not GC journal heads */
1494 err = ubifs_tnc_read_node(c, &zbr, node);
1498 err = fallible_read_node(c, key, &zbr, node);
1499 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1501 * The node may have been GC'ed out from under us so try again
1502 * while keeping the TNC mutex locked.
1510 mutex_unlock(&c->tnc_mutex);
1515 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1516 * @c: UBIFS file-system description object
1517 * @bu: bulk-read parameters and results
1519 * Lookup consecutive data node keys for the same inode that reside
1520 * consecutively in the same LEB. This function returns zero in case of success
1521 * and a negative error code in case of failure.
1523 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1524 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1525 * maximum possible amount of nodes for bulk-read.
1527 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1529 int n, err = 0, lnum = -1, uninitialized_var(offs);
1530 int uninitialized_var(len);
1531 unsigned int block = key_block(c, &bu->key);
1532 struct ubifs_znode *znode;
1538 mutex_lock(&c->tnc_mutex);
1539 /* Find first key */
1540 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1545 len = znode->zbranch[n].len;
1546 /* The buffer must be big enough for at least 1 node */
1547 if (len > bu->buf_len) {
1552 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1554 lnum = znode->zbranch[n].lnum;
1555 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1558 struct ubifs_zbranch *zbr;
1559 union ubifs_key *key;
1560 unsigned int next_block;
1563 err = tnc_next(c, &znode, &n);
1566 zbr = &znode->zbranch[n];
1568 /* See if there is another data key for this file */
1569 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1570 key_type(c, key) != UBIFS_DATA_KEY) {
1575 /* First key found */
1577 offs = ALIGN(zbr->offs + zbr->len, 8);
1579 if (len > bu->buf_len) {
1585 * The data nodes must be in consecutive positions in
1588 if (zbr->lnum != lnum || zbr->offs != offs)
1590 offs += ALIGN(zbr->len, 8);
1591 len = ALIGN(len, 8) + zbr->len;
1592 /* Must not exceed buffer length */
1593 if (len > bu->buf_len)
1596 /* Allow for holes */
1597 next_block = key_block(c, key);
1598 bu->blk_cnt += (next_block - block - 1);
1599 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1603 bu->zbranch[bu->cnt++] = *zbr;
1605 /* See if we have room for more */
1606 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1608 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1612 if (err == -ENOENT) {
1616 bu->gc_seq = c->gc_seq;
1617 mutex_unlock(&c->tnc_mutex);
1621 * An enormous hole could cause bulk-read to encompass too many
1622 * page cache pages, so limit the number here.
1624 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1625 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1627 * Ensure that bulk-read covers a whole number of page cache
1630 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1631 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1634 /* At the end of file we can round up */
1635 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1638 /* Exclude data nodes that do not make up a whole page cache page */
1639 block = key_block(c, &bu->key) + bu->blk_cnt;
1640 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1642 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1650 * read_wbuf - bulk-read from a LEB with a wbuf.
1651 * @wbuf: wbuf that may overlap the read
1652 * @buf: buffer into which to read
1654 * @lnum: LEB number from which to read
1655 * @offs: offset from which to read
1657 * This functions returns %0 on success or a negative error code on failure.
1659 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1662 const struct ubifs_info *c = wbuf->c;
1665 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1666 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1667 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1668 ubifs_assert(c, offs + len <= c->leb_size);
1670 spin_lock(&wbuf->lock);
1671 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1673 /* We may safely unlock the write-buffer and read the data */
1674 spin_unlock(&wbuf->lock);
1675 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1678 /* Don't read under wbuf */
1679 rlen = wbuf->offs - offs;
1683 /* Copy the rest from the write-buffer */
1684 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1685 spin_unlock(&wbuf->lock);
1688 /* Read everything that goes before write-buffer */
1689 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1695 * validate_data_node - validate data nodes for bulk-read.
1696 * @c: UBIFS file-system description object
1697 * @buf: buffer containing data node to validate
1698 * @zbr: zbranch of data node to validate
1700 * This functions returns %0 on success or a negative error code on failure.
1702 static int validate_data_node(struct ubifs_info *c, void *buf,
1703 struct ubifs_zbranch *zbr)
1705 union ubifs_key key1;
1706 struct ubifs_ch *ch = buf;
1709 if (ch->node_type != UBIFS_DATA_NODE) {
1710 ubifs_err(c, "bad node type (%d but expected %d)",
1711 ch->node_type, UBIFS_DATA_NODE);
1715 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1717 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1721 err = ubifs_node_check_hash(c, buf, zbr->hash);
1723 ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs);
1727 len = le32_to_cpu(ch->len);
1728 if (len != zbr->len) {
1729 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1733 /* Make sure the key of the read node is correct */
1734 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1735 if (!keys_eq(c, &zbr->key, &key1)) {
1736 ubifs_err(c, "bad key in node at LEB %d:%d",
1737 zbr->lnum, zbr->offs);
1738 dbg_tnck(&zbr->key, "looked for key ");
1739 dbg_tnck(&key1, "found node's key ");
1748 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1749 ubifs_dump_node(c, buf);
1755 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1756 * @c: UBIFS file-system description object
1757 * @bu: bulk-read parameters and results
1759 * This functions reads and validates the data nodes that were identified by the
1760 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1761 * -EAGAIN to indicate a race with GC, or another negative error code on
1764 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1766 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1767 struct ubifs_wbuf *wbuf;
1770 len = bu->zbranch[bu->cnt - 1].offs;
1771 len += bu->zbranch[bu->cnt - 1].len - offs;
1772 if (len > bu->buf_len) {
1773 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1778 wbuf = ubifs_get_wbuf(c, lnum);
1780 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1782 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1784 /* Check for a race with GC */
1785 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1788 if (err && err != -EBADMSG) {
1789 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1792 dbg_tnck(&bu->key, "key ");
1796 /* Validate the nodes read */
1798 for (i = 0; i < bu->cnt; i++) {
1799 err = validate_data_node(c, buf, &bu->zbranch[i]);
1802 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1809 * do_lookup_nm- look up a "hashed" node.
1810 * @c: UBIFS file-system description object
1811 * @key: node key to lookup
1812 * @node: the node is returned here
1815 * This function looks up and reads a node which contains name hash in the key.
1816 * Since the hash may have collisions, there may be many nodes with the same
1817 * key, so we have to sequentially look to all of them until the needed one is
1818 * found. This function returns zero in case of success, %-ENOENT if the node
1819 * was not found, and a negative error code in case of failure.
1821 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1822 void *node, const struct fscrypt_name *nm)
1825 struct ubifs_znode *znode;
1827 dbg_tnck(key, "key ");
1828 mutex_lock(&c->tnc_mutex);
1829 found = ubifs_lookup_level0(c, key, &znode, &n);
1833 } else if (found < 0) {
1838 ubifs_assert(c, n >= 0);
1840 err = resolve_collision(c, key, &znode, &n, nm);
1841 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1842 if (unlikely(err < 0))
1849 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1852 mutex_unlock(&c->tnc_mutex);
1857 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1858 * @c: UBIFS file-system description object
1859 * @key: node key to lookup
1860 * @node: the node is returned here
1863 * This function looks up and reads a node which contains name hash in the key.
1864 * Since the hash may have collisions, there may be many nodes with the same
1865 * key, so we have to sequentially look to all of them until the needed one is
1866 * found. This function returns zero in case of success, %-ENOENT if the node
1867 * was not found, and a negative error code in case of failure.
1869 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1870 void *node, const struct fscrypt_name *nm)
1873 const struct ubifs_dent_node *dent = node;
1876 * We assume that in most of the cases there are no name collisions and
1877 * 'ubifs_tnc_lookup()' returns us the right direntry.
1879 err = ubifs_tnc_lookup(c, key, node);
1883 len = le16_to_cpu(dent->nlen);
1884 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1888 * Unluckily, there are hash collisions and we have to iterate over
1889 * them look at each direntry with colliding name hash sequentially.
1892 return do_lookup_nm(c, key, node, nm);
1895 static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1896 struct ubifs_dent_node *dent, uint32_t cookie,
1897 struct ubifs_znode **zn, int *n)
1900 struct ubifs_znode *znode = *zn;
1901 struct ubifs_zbranch *zbr;
1902 union ubifs_key *dkey;
1905 zbr = &znode->zbranch[*n];
1908 if (key_inum(c, dkey) != key_inum(c, key) ||
1909 key_type(c, dkey) != key_type(c, key)) {
1913 err = tnc_read_hashed_node(c, zbr, dent);
1917 if (key_hash(c, key) == key_hash(c, dkey) &&
1918 le32_to_cpu(dent->cookie) == cookie) {
1923 err = tnc_next(c, &znode, n);
1929 static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1930 struct ubifs_dent_node *dent, uint32_t cookie)
1933 struct ubifs_znode *znode;
1934 union ubifs_key start_key;
1936 ubifs_assert(c, is_hash_key(c, key));
1938 lowest_dent_key(c, &start_key, key_inum(c, key));
1940 mutex_lock(&c->tnc_mutex);
1941 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1942 if (unlikely(err < 0))
1945 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
1948 mutex_unlock(&c->tnc_mutex);
1953 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1954 * @c: UBIFS file-system description object
1955 * @key: node key to lookup
1956 * @node: the node is returned here
1957 * @cookie: node cookie for collision resolution
1959 * This function looks up and reads a node which contains name hash in the key.
1960 * Since the hash may have collisions, there may be many nodes with the same
1961 * key, so we have to sequentially look to all of them until the needed one
1962 * with the same cookie value is found.
1963 * This function returns zero in case of success, %-ENOENT if the node
1964 * was not found, and a negative error code in case of failure.
1966 int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1967 void *node, uint32_t cookie)
1970 const struct ubifs_dent_node *dent = node;
1972 if (!c->double_hash)
1976 * We assume that in most of the cases there are no name collisions and
1977 * 'ubifs_tnc_lookup()' returns us the right direntry.
1979 err = ubifs_tnc_lookup(c, key, node);
1983 if (le32_to_cpu(dent->cookie) == cookie)
1987 * Unluckily, there are hash collisions and we have to iterate over
1988 * them look at each direntry with colliding name hash sequentially.
1990 return do_lookup_dh(c, key, node, cookie);
1994 * correct_parent_keys - correct parent znodes' keys.
1995 * @c: UBIFS file-system description object
1996 * @znode: znode to correct parent znodes for
1998 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1999 * zbranch changes, keys of parent znodes have to be corrected. This helper
2000 * function is called in such situations and corrects the keys if needed.
2002 static void correct_parent_keys(const struct ubifs_info *c,
2003 struct ubifs_znode *znode)
2005 union ubifs_key *key, *key1;
2007 ubifs_assert(c, znode->parent);
2008 ubifs_assert(c, znode->iip == 0);
2010 key = &znode->zbranch[0].key;
2011 key1 = &znode->parent->zbranch[0].key;
2013 while (keys_cmp(c, key, key1) < 0) {
2014 key_copy(c, key, key1);
2015 znode = znode->parent;
2017 if (!znode->parent || znode->iip)
2019 key1 = &znode->parent->zbranch[0].key;
2024 * insert_zbranch - insert a zbranch into a znode.
2025 * @c: UBIFS file-system description object
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_info *c, struct ubifs_znode *znode,
2036 const struct ubifs_zbranch *zbr, int n)
2040 ubifs_assert(c, 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(c, n >= 0 && n <= c->fanout);
2096 /* Implement naive insert for now */
2099 if (znode->child_cnt < c->fanout) {
2100 ubifs_assert(c, n != c->fanout);
2101 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2103 insert_zbranch(c, 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(c, 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
2274 * @hash: The hash over the node
2276 * This function adds a node with key @key to TNC. The node may be new or it may
2277 * obsolete some existing one. Returns %0 on success or negative error code on
2280 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2281 int offs, int len, const u8 *hash)
2283 int found, n, err = 0;
2284 struct ubifs_znode *znode;
2286 mutex_lock(&c->tnc_mutex);
2287 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2288 found = lookup_level0_dirty(c, key, &znode, &n);
2290 struct ubifs_zbranch zbr;
2296 ubifs_copy_hash(c, hash, zbr.hash);
2297 key_copy(c, key, &zbr.key);
2298 err = tnc_insert(c, znode, &zbr, n + 1);
2299 } else if (found == 1) {
2300 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2303 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2307 ubifs_copy_hash(c, hash, zbr->hash);
2311 err = dbg_check_tnc(c, 0);
2312 mutex_unlock(&c->tnc_mutex);
2318 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2319 * @c: UBIFS file-system description object
2321 * @old_lnum: LEB number of old node
2322 * @old_offs: old node offset
2323 * @lnum: LEB number of node
2324 * @offs: node offset
2327 * This function replaces a node with key @key in the TNC only if the old node
2328 * is found. This function is called by garbage collection when node are moved.
2329 * Returns %0 on success or negative error code on failure.
2331 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2332 int old_lnum, int old_offs, int lnum, int offs, int len)
2334 int found, n, err = 0;
2335 struct ubifs_znode *znode;
2337 mutex_lock(&c->tnc_mutex);
2338 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2339 old_offs, lnum, offs, len);
2340 found = lookup_level0_dirty(c, key, &znode, &n);
2347 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2350 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2352 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2359 } else if (is_hash_key(c, key)) {
2360 found = resolve_collision_directly(c, key, &znode, &n,
2361 old_lnum, old_offs);
2362 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2363 found, znode, n, old_lnum, old_offs);
2370 /* Ensure the znode is dirtied */
2371 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2372 znode = dirty_cow_bottom_up(c, znode);
2373 if (IS_ERR(znode)) {
2374 err = PTR_ERR(znode);
2378 zbr = &znode->zbranch[n];
2380 err = ubifs_add_dirt(c, zbr->lnum,
2392 err = ubifs_add_dirt(c, lnum, len);
2395 err = dbg_check_tnc(c, 0);
2398 mutex_unlock(&c->tnc_mutex);
2403 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2404 * @c: UBIFS file-system description object
2406 * @lnum: LEB number of node
2407 * @offs: node offset
2409 * @hash: The hash over the node
2412 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2413 * may have collisions, like directory entry keys.
2415 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2416 int lnum, int offs, int len, const u8 *hash,
2417 const struct fscrypt_name *nm)
2419 int found, n, err = 0;
2420 struct ubifs_znode *znode;
2422 mutex_lock(&c->tnc_mutex);
2423 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2424 found = lookup_level0_dirty(c, key, &znode, &n);
2432 found = fallible_resolve_collision(c, key, &znode, &n,
2435 found = resolve_collision(c, key, &znode, &n, nm);
2436 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2442 /* Ensure the znode is dirtied */
2443 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2444 znode = dirty_cow_bottom_up(c, znode);
2445 if (IS_ERR(znode)) {
2446 err = PTR_ERR(znode);
2452 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2455 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2459 ubifs_copy_hash(c, hash, zbr->hash);
2465 struct ubifs_zbranch zbr;
2471 ubifs_copy_hash(c, hash, zbr.hash);
2472 key_copy(c, key, &zbr.key);
2473 err = tnc_insert(c, znode, &zbr, n + 1);
2478 * We did not find it in the index so there may be a
2479 * dangling branch still in the index. So we remove it
2480 * by passing 'ubifs_tnc_remove_nm()' the same key but
2481 * an unmatchable name.
2483 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2485 err = dbg_check_tnc(c, 0);
2486 mutex_unlock(&c->tnc_mutex);
2489 return ubifs_tnc_remove_nm(c, key, &noname);
2495 err = dbg_check_tnc(c, 0);
2496 mutex_unlock(&c->tnc_mutex);
2501 * tnc_delete - delete a znode form TNC.
2502 * @c: UBIFS file-system description object
2503 * @znode: znode to delete from
2504 * @n: zbranch slot number to delete
2506 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2507 * case of success and a negative error code in case of failure.
2509 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2511 struct ubifs_zbranch *zbr;
2512 struct ubifs_znode *zp;
2515 /* Delete without merge for now */
2516 ubifs_assert(c, znode->level == 0);
2517 ubifs_assert(c, n >= 0 && n < c->fanout);
2518 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2520 zbr = &znode->zbranch[n];
2523 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2525 ubifs_dump_znode(c, znode);
2529 /* We do not "gap" zbranch slots */
2530 for (i = n; i < znode->child_cnt - 1; i++)
2531 znode->zbranch[i] = znode->zbranch[i + 1];
2532 znode->child_cnt -= 1;
2534 if (znode->child_cnt > 0)
2538 * This was the last zbranch, we have to delete this znode from the
2543 ubifs_assert(c, !ubifs_zn_obsolete(znode));
2544 ubifs_assert(c, ubifs_zn_dirty(znode));
2549 atomic_long_dec(&c->dirty_zn_cnt);
2551 err = insert_old_idx_znode(c, znode);
2556 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2557 atomic_long_inc(&c->clean_zn_cnt);
2558 atomic_long_inc(&ubifs_clean_zn_cnt);
2562 } while (znode->child_cnt == 1); /* while removing last child */
2564 /* Remove from znode, entry n - 1 */
2565 znode->child_cnt -= 1;
2566 ubifs_assert(c, znode->level != 0);
2567 for (i = n; i < znode->child_cnt; i++) {
2568 znode->zbranch[i] = znode->zbranch[i + 1];
2569 if (znode->zbranch[i].znode)
2570 znode->zbranch[i].znode->iip = i;
2574 * If this is the root and it has only 1 child then
2575 * collapse the tree.
2577 if (!znode->parent) {
2578 while (znode->child_cnt == 1 && znode->level != 0) {
2580 zbr = &znode->zbranch[0];
2581 znode = get_znode(c, znode, 0);
2583 return PTR_ERR(znode);
2584 znode = dirty_cow_znode(c, zbr);
2586 return PTR_ERR(znode);
2587 znode->parent = NULL;
2590 err = insert_old_idx(c, c->zroot.lnum,
2595 c->zroot.lnum = zbr->lnum;
2596 c->zroot.offs = zbr->offs;
2597 c->zroot.len = zbr->len;
2598 c->zroot.znode = znode;
2599 ubifs_assert(c, !ubifs_zn_obsolete(zp));
2600 ubifs_assert(c, ubifs_zn_dirty(zp));
2601 atomic_long_dec(&c->dirty_zn_cnt);
2604 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2605 atomic_long_inc(&c->clean_zn_cnt);
2606 atomic_long_inc(&ubifs_clean_zn_cnt);
2616 * ubifs_tnc_remove - remove an index entry of a node.
2617 * @c: UBIFS file-system description object
2620 * Returns %0 on success or negative error code on failure.
2622 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2624 int found, n, err = 0;
2625 struct ubifs_znode *znode;
2627 mutex_lock(&c->tnc_mutex);
2628 dbg_tnck(key, "key ");
2629 found = lookup_level0_dirty(c, key, &znode, &n);
2635 err = tnc_delete(c, znode, n);
2637 err = dbg_check_tnc(c, 0);
2640 mutex_unlock(&c->tnc_mutex);
2645 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2646 * @c: UBIFS file-system description object
2648 * @nm: directory entry name
2650 * Returns %0 on success or negative error code on failure.
2652 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2653 const struct fscrypt_name *nm)
2656 struct ubifs_znode *znode;
2658 mutex_lock(&c->tnc_mutex);
2659 dbg_tnck(key, "key ");
2660 err = lookup_level0_dirty(c, key, &znode, &n);
2666 err = fallible_resolve_collision(c, key, &znode, &n,
2669 err = resolve_collision(c, key, &znode, &n, nm);
2670 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2674 /* Ensure the znode is dirtied */
2675 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2676 znode = dirty_cow_bottom_up(c, znode);
2677 if (IS_ERR(znode)) {
2678 err = PTR_ERR(znode);
2682 err = tnc_delete(c, znode, n);
2688 err = dbg_check_tnc(c, 0);
2689 mutex_unlock(&c->tnc_mutex);
2694 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2695 * @c: UBIFS file-system description object
2697 * @cookie: node cookie for collision resolution
2699 * Returns %0 on success or negative error code on failure.
2701 int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2705 struct ubifs_znode *znode;
2706 struct ubifs_dent_node *dent;
2707 struct ubifs_zbranch *zbr;
2709 if (!c->double_hash)
2712 mutex_lock(&c->tnc_mutex);
2713 err = lookup_level0_dirty(c, key, &znode, &n);
2717 zbr = &znode->zbranch[n];
2718 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2724 err = tnc_read_hashed_node(c, zbr, dent);
2728 /* If the cookie does not match, we're facing a hash collision. */
2729 if (le32_to_cpu(dent->cookie) != cookie) {
2730 union ubifs_key start_key;
2732 lowest_dent_key(c, &start_key, key_inum(c, key));
2734 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2735 if (unlikely(err < 0))
2738 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
2743 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2744 znode = dirty_cow_bottom_up(c, znode);
2745 if (IS_ERR(znode)) {
2746 err = PTR_ERR(znode);
2750 err = tnc_delete(c, znode, n);
2756 err = dbg_check_tnc(c, 0);
2757 mutex_unlock(&c->tnc_mutex);
2762 * key_in_range - determine if a key falls within a range of keys.
2763 * @c: UBIFS file-system description object
2764 * @key: key to check
2765 * @from_key: lowest key in range
2766 * @to_key: highest key in range
2768 * This function returns %1 if the key is in range and %0 otherwise.
2770 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2771 union ubifs_key *from_key, union ubifs_key *to_key)
2773 if (keys_cmp(c, key, from_key) < 0)
2775 if (keys_cmp(c, key, to_key) > 0)
2781 * ubifs_tnc_remove_range - remove index entries in range.
2782 * @c: UBIFS file-system description object
2783 * @from_key: lowest key to remove
2784 * @to_key: highest key to remove
2786 * This function removes index entries starting at @from_key and ending at
2787 * @to_key. This function returns zero in case of success and a negative error
2788 * code in case of failure.
2790 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2791 union ubifs_key *to_key)
2793 int i, n, k, err = 0;
2794 struct ubifs_znode *znode;
2795 union ubifs_key *key;
2797 mutex_lock(&c->tnc_mutex);
2799 /* Find first level 0 znode that contains keys to remove */
2800 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2807 err = tnc_next(c, &znode, &n);
2808 if (err == -ENOENT) {
2814 key = &znode->zbranch[n].key;
2815 if (!key_in_range(c, key, from_key, to_key)) {
2821 /* Ensure the znode is dirtied */
2822 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2823 znode = dirty_cow_bottom_up(c, znode);
2824 if (IS_ERR(znode)) {
2825 err = PTR_ERR(znode);
2830 /* Remove all keys in range except the first */
2831 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2832 key = &znode->zbranch[i].key;
2833 if (!key_in_range(c, key, from_key, to_key))
2835 lnc_free(&znode->zbranch[i]);
2836 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2837 znode->zbranch[i].len);
2839 ubifs_dump_znode(c, znode);
2842 dbg_tnck(key, "removing key ");
2845 for (i = n + 1 + k; i < znode->child_cnt; i++)
2846 znode->zbranch[i - k] = znode->zbranch[i];
2847 znode->child_cnt -= k;
2850 /* Now delete the first */
2851 err = tnc_delete(c, znode, n);
2858 err = dbg_check_tnc(c, 0);
2859 mutex_unlock(&c->tnc_mutex);
2864 * ubifs_tnc_remove_ino - remove an inode from TNC.
2865 * @c: UBIFS file-system description object
2866 * @inum: inode number to remove
2868 * This function remove inode @inum and all the extended attributes associated
2869 * with the anode from TNC and returns zero in case of success or a negative
2870 * error code in case of failure.
2872 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2874 union ubifs_key key1, key2;
2875 struct ubifs_dent_node *xent, *pxent = NULL;
2876 struct fscrypt_name nm = {0};
2878 dbg_tnc("ino %lu", (unsigned long)inum);
2881 * Walk all extended attribute entries and remove them together with
2882 * corresponding extended attribute inodes.
2884 lowest_xent_key(c, &key1, inum);
2889 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2891 err = PTR_ERR(xent);
2897 xattr_inum = le64_to_cpu(xent->inum);
2898 dbg_tnc("xent '%s', ino %lu", xent->name,
2899 (unsigned long)xattr_inum);
2901 ubifs_evict_xattr_inode(c, xattr_inum);
2903 fname_name(&nm) = xent->name;
2904 fname_len(&nm) = le16_to_cpu(xent->nlen);
2905 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2911 lowest_ino_key(c, &key1, xattr_inum);
2912 highest_ino_key(c, &key2, xattr_inum);
2913 err = ubifs_tnc_remove_range(c, &key1, &key2);
2921 key_read(c, &xent->key, &key1);
2925 lowest_ino_key(c, &key1, inum);
2926 highest_ino_key(c, &key2, inum);
2928 return ubifs_tnc_remove_range(c, &key1, &key2);
2932 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2933 * @c: UBIFS file-system description object
2934 * @key: key of last entry
2935 * @nm: name of last entry found or %NULL
2937 * This function finds and reads the next directory or extended attribute entry
2938 * after the given key (@key) if there is one. @nm is used to resolve
2941 * If the name of the current entry is not known and only the key is known,
2942 * @nm->name has to be %NULL. In this case the semantics of this function is a
2943 * little bit different and it returns the entry corresponding to this key, not
2944 * the next one. If the key was not found, the closest "right" entry is
2947 * If the fist entry has to be found, @key has to contain the lowest possible
2948 * key value for this inode and @name has to be %NULL.
2950 * This function returns the found directory or extended attribute entry node
2951 * in case of success, %-ENOENT is returned if no entry was found, and a
2952 * negative error code is returned in case of failure.
2954 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2955 union ubifs_key *key,
2956 const struct fscrypt_name *nm)
2958 int n, err, type = key_type(c, key);
2959 struct ubifs_znode *znode;
2960 struct ubifs_dent_node *dent;
2961 struct ubifs_zbranch *zbr;
2962 union ubifs_key *dkey;
2964 dbg_tnck(key, "key ");
2965 ubifs_assert(c, is_hash_key(c, key));
2967 mutex_lock(&c->tnc_mutex);
2968 err = ubifs_lookup_level0(c, key, &znode, &n);
2969 if (unlikely(err < 0))
2972 if (fname_len(nm) > 0) {
2974 /* Handle collisions */
2976 err = fallible_resolve_collision(c, key, &znode, &n,
2979 err = resolve_collision(c, key, &znode, &n, nm);
2980 dbg_tnc("rc returned %d, znode %p, n %d",
2982 if (unlikely(err < 0))
2986 /* Now find next entry */
2987 err = tnc_next(c, &znode, &n);
2992 * The full name of the entry was not given, in which case the
2993 * behavior of this function is a little different and it
2994 * returns current entry, not the next one.
2998 * However, the given key does not exist in the TNC
2999 * tree and @znode/@n variables contain the closest
3000 * "preceding" element. Switch to the next one.
3002 err = tnc_next(c, &znode, &n);
3008 zbr = &znode->zbranch[n];
3009 dent = kmalloc(zbr->len, GFP_NOFS);
3010 if (unlikely(!dent)) {
3016 * The above 'tnc_next()' call could lead us to the next inode, check
3020 if (key_inum(c, dkey) != key_inum(c, key) ||
3021 key_type(c, dkey) != type) {
3026 err = tnc_read_hashed_node(c, zbr, dent);
3030 mutex_unlock(&c->tnc_mutex);
3036 mutex_unlock(&c->tnc_mutex);
3037 return ERR_PTR(err);
3041 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3042 * @c: UBIFS file-system description object
3044 * Destroy left-over obsolete znodes from a failed commit.
3046 static void tnc_destroy_cnext(struct ubifs_info *c)
3048 struct ubifs_znode *cnext;
3052 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3055 struct ubifs_znode *znode = cnext;
3057 cnext = cnext->cnext;
3058 if (ubifs_zn_obsolete(znode))
3060 } while (cnext && cnext != c->cnext);
3064 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3065 * @c: UBIFS file-system description object
3067 void ubifs_tnc_close(struct ubifs_info *c)
3069 tnc_destroy_cnext(c);
3070 if (c->zroot.znode) {
3073 n = atomic_long_read(&c->clean_zn_cnt);
3074 freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode);
3075 ubifs_assert(c, freed == n);
3076 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3084 * left_znode - get the znode to the left.
3085 * @c: UBIFS file-system description object
3088 * This function returns a pointer to the znode to the left of @znode or NULL if
3089 * there is not one. A negative error code is returned on failure.
3091 static struct ubifs_znode *left_znode(struct ubifs_info *c,
3092 struct ubifs_znode *znode)
3094 int level = znode->level;
3097 int n = znode->iip - 1;
3099 /* Go up until we can go left */
3100 znode = znode->parent;
3104 /* Now go down the rightmost branch to 'level' */
3105 znode = get_znode(c, znode, n);
3108 while (znode->level != level) {
3109 n = znode->child_cnt - 1;
3110 znode = get_znode(c, znode, n);
3121 * right_znode - get the znode to the right.
3122 * @c: UBIFS file-system description object
3125 * This function returns a pointer to the znode to the right of @znode or NULL
3126 * if there is not one. A negative error code is returned on failure.
3128 static struct ubifs_znode *right_znode(struct ubifs_info *c,
3129 struct ubifs_znode *znode)
3131 int level = znode->level;
3134 int n = znode->iip + 1;
3136 /* Go up until we can go right */
3137 znode = znode->parent;
3140 if (n < znode->child_cnt) {
3141 /* Now go down the leftmost branch to 'level' */
3142 znode = get_znode(c, znode, n);
3145 while (znode->level != level) {
3146 znode = get_znode(c, znode, 0);
3157 * lookup_znode - find a particular indexing node from TNC.
3158 * @c: UBIFS file-system description object
3159 * @key: index node key to lookup
3160 * @level: index node level
3161 * @lnum: index node LEB number
3162 * @offs: index node offset
3164 * This function searches an indexing node by its first key @key and its
3165 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3166 * nodes it traverses to TNC. This function is called for indexing nodes which
3167 * were found on the media by scanning, for example when garbage-collecting or
3168 * when doing in-the-gaps commit. This means that the indexing node which is
3169 * looked for does not have to have exactly the same leftmost key @key, because
3170 * the leftmost key may have been changed, in which case TNC will contain a
3171 * dirty znode which still refers the same @lnum:@offs. This function is clever
3172 * enough to recognize such indexing nodes.
3174 * Note, if a znode was deleted or changed too much, then this function will
3175 * not find it. For situations like this UBIFS has the old index RB-tree
3176 * (indexed by @lnum:@offs).
3178 * This function returns a pointer to the znode found or %NULL if it is not
3179 * found. A negative error code is returned on failure.
3181 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3182 union ubifs_key *key, int level,
3185 struct ubifs_znode *znode, *zn;
3188 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3191 * The arguments have probably been read off flash, so don't assume
3195 return ERR_PTR(-EINVAL);
3197 /* Get the root znode */
3198 znode = c->zroot.znode;
3200 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3204 /* Check if it is the one we are looking for */
3205 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3207 /* Descend to the parent level i.e. (level + 1) */
3208 if (level >= znode->level)
3211 ubifs_search_zbranch(c, znode, key, &n);
3214 * We reached a znode where the leftmost key is greater
3215 * than the key we are searching for. This is the same
3216 * situation as the one described in a huge comment at
3217 * the end of the 'ubifs_lookup_level0()' function. And
3218 * for exactly the same reasons we have to try to look
3219 * left before giving up.
3221 znode = left_znode(c, znode);
3226 ubifs_search_zbranch(c, znode, key, &n);
3227 ubifs_assert(c, n >= 0);
3229 if (znode->level == level + 1)
3231 znode = get_znode(c, znode, n);
3235 /* Check if the child is the one we are looking for */
3236 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3237 return get_znode(c, znode, n);
3238 /* If the key is unique, there is nowhere else to look */
3239 if (!is_hash_key(c, key))
3242 * The key is not unique and so may be also in the znodes to either
3249 /* Move one branch to the left */
3253 znode = left_znode(c, znode);
3258 n = znode->child_cnt - 1;
3261 if (znode->zbranch[n].lnum == lnum &&
3262 znode->zbranch[n].offs == offs)
3263 return get_znode(c, znode, n);
3264 /* Stop if the key is less than the one we are looking for */
3265 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3268 /* Back to the middle */
3273 /* Move one branch to the right */
3274 if (++n >= znode->child_cnt) {
3275 znode = right_znode(c, znode);
3283 if (znode->zbranch[n].lnum == lnum &&
3284 znode->zbranch[n].offs == offs)
3285 return get_znode(c, znode, n);
3286 /* Stop if the key is greater than the one we are looking for */
3287 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3294 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3295 * @c: UBIFS file-system description object
3296 * @key: key of index node
3297 * @level: index node level
3298 * @lnum: LEB number of index node
3299 * @offs: offset of index node
3301 * This function returns %0 if the index node is not referred to in the TNC, %1
3302 * if the index node is referred to in the TNC and the corresponding znode is
3303 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3304 * znode is clean, and a negative error code in case of failure.
3306 * Note, the @key argument has to be the key of the first child. Also note,
3307 * this function relies on the fact that 0:0 is never a valid LEB number and
3308 * offset for a main-area node.
3310 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3313 struct ubifs_znode *znode;
3315 znode = lookup_znode(c, key, level, lnum, offs);
3319 return PTR_ERR(znode);
3321 return ubifs_zn_dirty(znode) ? 1 : 2;
3325 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3326 * @c: UBIFS file-system description object
3328 * @lnum: node LEB number
3329 * @offs: node offset
3331 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3332 * not, and a negative error code in case of failure.
3334 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3335 * and offset for a main-area node.
3337 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3340 struct ubifs_zbranch *zbr;
3341 struct ubifs_znode *znode, *zn;
3342 int n, found, err, nn;
3343 const int unique = !is_hash_key(c, key);
3345 found = ubifs_lookup_level0(c, key, &znode, &n);
3347 return found; /* Error code */
3350 zbr = &znode->zbranch[n];
3351 if (lnum == zbr->lnum && offs == zbr->offs)
3352 return 1; /* Found it */
3356 * Because the key is not unique, we have to look left
3363 err = tnc_prev(c, &znode, &n);
3368 if (keys_cmp(c, key, &znode->zbranch[n].key))
3370 zbr = &znode->zbranch[n];
3371 if (lnum == zbr->lnum && offs == zbr->offs)
3372 return 1; /* Found it */
3378 err = tnc_next(c, &znode, &n);
3384 if (keys_cmp(c, key, &znode->zbranch[n].key))
3386 zbr = &znode->zbranch[n];
3387 if (lnum == zbr->lnum && offs == zbr->offs)
3388 return 1; /* Found it */
3394 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3395 * @c: UBIFS file-system description object
3397 * @level: index node level (if it is an index node)
3398 * @lnum: node LEB number
3399 * @offs: node offset
3400 * @is_idx: non-zero if the node is an index node
3402 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3403 * negative error code in case of failure. For index nodes, @key has to be the
3404 * key of the first child. An index node is considered to be in the TNC only if
3405 * the corresponding znode is clean or has not been loaded.
3407 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3408 int lnum, int offs, int is_idx)
3412 mutex_lock(&c->tnc_mutex);
3414 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3418 /* The index node was found but it was dirty */
3421 /* The index node was found and it was clean */
3426 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3429 mutex_unlock(&c->tnc_mutex);
3434 * ubifs_dirty_idx_node - dirty an index node.
3435 * @c: UBIFS file-system description object
3436 * @key: index node key
3437 * @level: index node level
3438 * @lnum: index node LEB number
3439 * @offs: index node offset
3441 * This function loads and dirties an index node so that it can be garbage
3442 * collected. The @key argument has to be the key of the first child. This
3443 * function relies on the fact that 0:0 is never a valid LEB number and offset
3444 * for a main-area node. Returns %0 on success and a negative error code on
3447 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3450 struct ubifs_znode *znode;
3453 mutex_lock(&c->tnc_mutex);
3454 znode = lookup_znode(c, key, level, lnum, offs);
3457 if (IS_ERR(znode)) {
3458 err = PTR_ERR(znode);
3461 znode = dirty_cow_bottom_up(c, znode);
3462 if (IS_ERR(znode)) {
3463 err = PTR_ERR(znode);
3468 mutex_unlock(&c->tnc_mutex);
3473 * dbg_check_inode_size - check if inode size is correct.
3474 * @c: UBIFS file-system description object
3475 * @inum: inode number
3478 * This function makes sure that the inode size (@size) is correct and it does
3479 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3480 * if it has a data page beyond @size, and other negative error code in case of
3483 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3487 union ubifs_key from_key, to_key, *key;
3488 struct ubifs_znode *znode;
3491 if (!S_ISREG(inode->i_mode))
3493 if (!dbg_is_chk_gen(c))
3496 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3497 data_key_init(c, &from_key, inode->i_ino, block);
3498 highest_data_key(c, &to_key, inode->i_ino);
3500 mutex_lock(&c->tnc_mutex);
3501 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3510 err = tnc_next(c, &znode, &n);
3511 if (err == -ENOENT) {
3518 ubifs_assert(c, err == 0);
3519 key = &znode->zbranch[n].key;
3520 if (!key_in_range(c, key, &from_key, &to_key))
3524 block = key_block(c, key);
3525 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3526 (unsigned long)inode->i_ino, size,
3527 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3528 mutex_unlock(&c->tnc_mutex);
3529 ubifs_dump_inode(c, inode);
3534 mutex_unlock(&c->tnc_mutex);