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 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(!zbr->leaf);
325 ubifs_assert(zbr->len != 0);
326 ubifs_assert(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(!zbr->leaf);
359 ubifs_assert(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(is_hash_key(c, &zbr->key));
404 /* Read from the leaf node cache */
405 ubifs_assert(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 * @len: node length (not aligned)
437 * @lnum: LEB number of node to read
438 * @offs: offset of node to read
440 * This function tries to read a node of known type and length, checks it and
441 * stores it in @buf. This function returns %1 if a node is present and %0 if
442 * a node is not present. A negative error code is returned for I/O errors.
443 * This function performs that same function as ubifs_read_node except that
444 * it does not require that there is actually a node present and instead
445 * the return code indicates if a node was read.
447 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
448 * is true (it is controlled by corresponding mount option). However, if
449 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
450 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
451 * because during mounting or re-mounting from R/O mode to R/W mode we may read
452 * journal nodes (when replying the journal or doing the recovery) and the
453 * journal nodes may potentially be corrupted, so checking is required.
455 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
456 int len, int lnum, int offs)
459 struct ubifs_ch *ch = buf;
460 uint32_t crc, node_crc;
462 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
464 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
466 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
467 type, lnum, offs, err);
471 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
474 if (ch->node_type != type)
477 node_len = le32_to_cpu(ch->len);
481 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
485 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
486 node_crc = le32_to_cpu(ch->crc);
494 * fallible_read_node - try to read a leaf node.
495 * @c: UBIFS file-system description object
496 * @key: key of node to read
497 * @zbr: position of node
498 * @node: node returned
500 * This function tries to read a node and returns %1 if the node is read, %0
501 * if the node is not present, and a negative error code in the case of error.
503 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
504 struct ubifs_zbranch *zbr, void *node)
508 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
510 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
513 union ubifs_key node_key;
514 struct ubifs_dent_node *dent = node;
516 /* All nodes have key in the same place */
517 key_read(c, &dent->key, &node_key);
518 if (keys_cmp(c, key, &node_key) != 0)
521 if (ret == 0 && c->replaying)
522 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
523 zbr->lnum, zbr->offs, zbr->len);
528 * matches_name - determine if a direntry or xattr entry matches a given name.
529 * @c: UBIFS file-system description object
530 * @zbr: zbranch of dent
533 * This function checks if xentry/direntry referred by zbranch @zbr matches name
534 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
535 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
536 * of failure, a negative error code is returned.
538 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
539 const struct fscrypt_name *nm)
541 struct ubifs_dent_node *dent;
544 /* If possible, match against the dent in the leaf node cache */
546 dent = kmalloc(zbr->len, GFP_NOFS);
550 err = ubifs_tnc_read_node(c, zbr, dent);
554 /* Add the node to the leaf node cache */
555 err = lnc_add_directly(c, zbr, dent);
561 nlen = le16_to_cpu(dent->nlen);
562 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
564 if (nlen == fname_len(nm))
566 else if (nlen < fname_len(nm))
581 * get_znode - get a TNC znode that may not be loaded yet.
582 * @c: UBIFS file-system description object
583 * @znode: parent znode
584 * @n: znode branch slot number
586 * This function returns the znode or a negative error code.
588 static struct ubifs_znode *get_znode(struct ubifs_info *c,
589 struct ubifs_znode *znode, int n)
591 struct ubifs_zbranch *zbr;
593 zbr = &znode->zbranch[n];
597 znode = ubifs_load_znode(c, zbr, znode, n);
602 * tnc_next - find next TNC entry.
603 * @c: UBIFS file-system description object
604 * @zn: znode is passed and returned here
605 * @n: znode branch slot number is passed and returned here
607 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
608 * no next entry, or a negative error code otherwise.
610 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
612 struct ubifs_znode *znode = *zn;
616 if (nn < znode->child_cnt) {
621 struct ubifs_znode *zp;
628 if (nn < znode->child_cnt) {
629 znode = get_znode(c, znode, nn);
631 return PTR_ERR(znode);
632 while (znode->level != 0) {
633 znode = get_znode(c, znode, 0);
635 return PTR_ERR(znode);
647 * tnc_prev - find previous TNC entry.
648 * @c: UBIFS file-system description object
649 * @zn: znode is returned here
650 * @n: znode branch slot number is passed and returned here
652 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
653 * there is no next entry, or a negative error code otherwise.
655 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
657 struct ubifs_znode *znode = *zn;
665 struct ubifs_znode *zp;
673 znode = get_znode(c, znode, nn);
675 return PTR_ERR(znode);
676 while (znode->level != 0) {
677 nn = znode->child_cnt - 1;
678 znode = get_znode(c, znode, nn);
680 return PTR_ERR(znode);
682 nn = znode->child_cnt - 1;
692 * resolve_collision - resolve a collision.
693 * @c: UBIFS file-system description object
694 * @key: key of a directory or extended attribute entry
695 * @zn: znode is returned here
696 * @n: zbranch number is passed and returned here
697 * @nm: name of the entry
699 * This function is called for "hashed" keys to make sure that the found key
700 * really corresponds to the looked up node (directory or extended attribute
701 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
702 * %0 is returned if @nm is not found and @zn and @n are set to the previous
703 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
704 * This means that @n may be set to %-1 if the leftmost key in @zn is the
705 * previous one. A negative error code is returned on failures.
707 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
708 struct ubifs_znode **zn, int *n,
709 const struct fscrypt_name *nm)
713 err = matches_name(c, &(*zn)->zbranch[*n], nm);
714 if (unlikely(err < 0))
716 if (err == NAME_MATCHES)
719 if (err == NAME_GREATER) {
722 err = tnc_prev(c, zn, n);
723 if (err == -ENOENT) {
724 ubifs_assert(*n == 0);
730 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
732 * We have found the branch after which we would
733 * like to insert, but inserting in this znode
734 * may still be wrong. Consider the following 3
735 * znodes, in the case where we are resolving a
736 * collision with Key2.
739 * ----------------------
740 * level 1 | Key0 | Key1 |
741 * -----------------------
743 * znode za | | znode zb
744 * ------------ ------------
745 * level 0 | Key0 | | Key2 |
746 * ------------ ------------
748 * The lookup finds Key2 in znode zb. Lets say
749 * there is no match and the name is greater so
750 * we look left. When we find Key0, we end up
751 * here. If we return now, we will insert into
752 * znode za at slot n = 1. But that is invalid
753 * according to the parent's keys. Key2 must
754 * be inserted into znode zb.
756 * Note, this problem is not relevant for the
757 * case when we go right, because
758 * 'tnc_insert()' would correct the parent key.
760 if (*n == (*zn)->child_cnt - 1) {
761 err = tnc_next(c, zn, n);
763 /* Should be impossible */
769 ubifs_assert(*n == 0);
774 err = matches_name(c, &(*zn)->zbranch[*n], nm);
777 if (err == NAME_LESS)
779 if (err == NAME_MATCHES)
781 ubifs_assert(err == NAME_GREATER);
785 struct ubifs_znode *znode = *zn;
789 err = tnc_next(c, &znode, &nn);
794 if (keys_cmp(c, &znode->zbranch[nn].key, key))
796 err = matches_name(c, &znode->zbranch[nn], nm);
799 if (err == NAME_GREATER)
803 if (err == NAME_MATCHES)
805 ubifs_assert(err == NAME_LESS);
811 * fallible_matches_name - determine if a dent matches a given name.
812 * @c: UBIFS file-system description object
813 * @zbr: zbranch of dent
816 * This is a "fallible" version of 'matches_name()' function which does not
817 * panic if the direntry/xentry referred by @zbr does not exist on the media.
819 * This function checks if xentry/direntry referred by zbranch @zbr matches name
820 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
821 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
822 * if xentry/direntry referred by @zbr does not exist on the media. A negative
823 * error code is returned in case of failure.
825 static int fallible_matches_name(struct ubifs_info *c,
826 struct ubifs_zbranch *zbr,
827 const struct fscrypt_name *nm)
829 struct ubifs_dent_node *dent;
832 /* If possible, match against the dent in the leaf node cache */
834 dent = kmalloc(zbr->len, GFP_NOFS);
838 err = fallible_read_node(c, &zbr->key, zbr, dent);
842 /* The node was not present */
846 ubifs_assert(err == 1);
848 err = lnc_add_directly(c, zbr, dent);
854 nlen = le16_to_cpu(dent->nlen);
855 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
857 if (nlen == fname_len(nm))
859 else if (nlen < fname_len(nm))
874 * fallible_resolve_collision - resolve a collision even if nodes are missing.
875 * @c: UBIFS file-system description object
877 * @zn: znode is returned here
878 * @n: branch number is passed and returned here
879 * @nm: name of directory entry
880 * @adding: indicates caller is adding a key to the TNC
882 * This is a "fallible" version of the 'resolve_collision()' function which
883 * does not panic if one of the nodes referred to by TNC does not exist on the
884 * media. This may happen when replaying the journal if a deleted node was
885 * Garbage-collected and the commit was not done. A branch that refers to a node
886 * that is not present is called a dangling branch. The following are the return
887 * codes for this function:
888 * o if @nm was found, %1 is returned and @zn and @n are set to the found
890 * o if we are @adding and @nm was not found, %0 is returned;
891 * o if we are not @adding and @nm was not found, but a dangling branch was
892 * found, then %1 is returned and @zn and @n are set to the dangling branch;
893 * o a negative error code is returned in case of failure.
895 static int fallible_resolve_collision(struct ubifs_info *c,
896 const union ubifs_key *key,
897 struct ubifs_znode **zn, int *n,
898 const struct fscrypt_name *nm,
901 struct ubifs_znode *o_znode = NULL, *znode = *zn;
902 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
904 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
905 if (unlikely(cmp < 0))
907 if (cmp == NAME_MATCHES)
909 if (cmp == NOT_ON_MEDIA) {
913 * We are unlucky and hit a dangling branch straight away.
914 * Now we do not really know where to go to find the needed
915 * branch - to the left or to the right. Well, let's try left.
919 unsure = 1; /* Remove a dangling branch wherever it is */
921 if (cmp == NAME_GREATER || unsure) {
924 err = tnc_prev(c, zn, n);
925 if (err == -ENOENT) {
926 ubifs_assert(*n == 0);
932 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
933 /* See comments in 'resolve_collision()' */
934 if (*n == (*zn)->child_cnt - 1) {
935 err = tnc_next(c, zn, n);
937 /* Should be impossible */
943 ubifs_assert(*n == 0);
948 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
951 if (err == NAME_MATCHES)
953 if (err == NOT_ON_MEDIA) {
960 if (err == NAME_LESS)
967 if (cmp == NAME_LESS || unsure) {
972 err = tnc_next(c, &znode, &nn);
977 if (keys_cmp(c, &znode->zbranch[nn].key, key))
979 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
982 if (err == NAME_GREATER)
986 if (err == NAME_MATCHES)
988 if (err == NOT_ON_MEDIA) {
995 /* Never match a dangling branch when adding */
996 if (adding || !o_znode)
999 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
1000 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1001 o_znode->zbranch[o_n].len);
1008 * matches_position - determine if a zbranch matches a given position.
1009 * @zbr: zbranch of dent
1010 * @lnum: LEB number of dent to match
1011 * @offs: offset of dent to match
1013 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1015 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1017 if (zbr->lnum == lnum && zbr->offs == offs)
1024 * resolve_collision_directly - resolve a collision directly.
1025 * @c: UBIFS file-system description object
1026 * @key: key of directory entry
1027 * @zn: znode is passed and returned here
1028 * @n: zbranch number is passed and returned here
1029 * @lnum: LEB number of dent node to match
1030 * @offs: offset of dent node to match
1032 * This function is used for "hashed" keys to make sure the found directory or
1033 * extended attribute entry node is what was looked for. It is used when the
1034 * flash address of the right node is known (@lnum:@offs) which makes it much
1035 * easier to resolve collisions (no need to read entries and match full
1036 * names). This function returns %1 and sets @zn and @n if the collision is
1037 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1038 * previous directory entry. Otherwise a negative error code is returned.
1040 static int resolve_collision_directly(struct ubifs_info *c,
1041 const union ubifs_key *key,
1042 struct ubifs_znode **zn, int *n,
1045 struct ubifs_znode *znode;
1050 if (matches_position(&znode->zbranch[nn], lnum, offs))
1055 err = tnc_prev(c, &znode, &nn);
1060 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1062 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1073 err = tnc_next(c, &znode, &nn);
1078 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1082 if (matches_position(&znode->zbranch[nn], lnum, offs))
1088 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1089 * @c: UBIFS file-system description object
1090 * @znode: znode to dirty
1092 * If we do not have a unique key that resides in a znode, then we cannot
1093 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1094 * This function records the path back to the last dirty ancestor, and then
1095 * dirties the znodes on that path.
1097 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1098 struct ubifs_znode *znode)
1100 struct ubifs_znode *zp;
1101 int *path = c->bottom_up_buf, p = 0;
1103 ubifs_assert(c->zroot.znode);
1104 ubifs_assert(znode);
1105 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1106 kfree(c->bottom_up_buf);
1107 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1109 if (!c->bottom_up_buf)
1110 return ERR_PTR(-ENOMEM);
1111 path = c->bottom_up_buf;
1113 if (c->zroot.znode->level) {
1114 /* Go up until parent is dirty */
1122 ubifs_assert(p < c->zroot.znode->level);
1124 if (!zp->cnext && ubifs_zn_dirty(znode))
1130 /* Come back down, dirtying as we go */
1132 struct ubifs_zbranch *zbr;
1136 ubifs_assert(path[p - 1] >= 0);
1137 ubifs_assert(path[p - 1] < zp->child_cnt);
1138 zbr = &zp->zbranch[path[--p]];
1139 znode = dirty_cow_znode(c, zbr);
1141 ubifs_assert(znode == c->zroot.znode);
1142 znode = dirty_cow_znode(c, &c->zroot);
1144 if (IS_ERR(znode) || !p)
1146 ubifs_assert(path[p - 1] >= 0);
1147 ubifs_assert(path[p - 1] < znode->child_cnt);
1148 znode = znode->zbranch[path[p - 1]].znode;
1155 * ubifs_lookup_level0 - search for zero-level znode.
1156 * @c: UBIFS file-system description object
1157 * @key: key to lookup
1158 * @zn: znode is returned here
1159 * @n: znode branch slot number is returned here
1161 * This function looks up the TNC tree and search for zero-level znode which
1162 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1164 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1165 * is returned and slot number of the matched branch is stored in @n;
1166 * o not exact match, which means that zero-level znode does not contain
1167 * @key, then %0 is returned and slot number of the closest branch or %-1
1168 * is stored in @n; In this case calling tnc_next() is mandatory.
1169 * o @key is so small that it is even less than the lowest key of the
1170 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1172 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1173 * function reads corresponding indexing nodes and inserts them to TNC. In
1174 * case of failure, a negative error code is returned.
1176 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1177 struct ubifs_znode **zn, int *n)
1180 struct ubifs_znode *znode;
1181 unsigned long time = get_seconds();
1183 dbg_tnck(key, "search key ");
1184 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1186 znode = c->zroot.znode;
1187 if (unlikely(!znode)) {
1188 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1190 return PTR_ERR(znode);
1196 struct ubifs_zbranch *zbr;
1198 exact = ubifs_search_zbranch(c, znode, key, n);
1200 if (znode->level == 0)
1205 zbr = &znode->zbranch[*n];
1213 /* znode is not in TNC cache, load it from the media */
1214 znode = ubifs_load_znode(c, zbr, znode, *n);
1216 return PTR_ERR(znode);
1220 if (exact || !is_hash_key(c, key) || *n != -1) {
1221 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1226 * Here is a tricky place. We have not found the key and this is a
1227 * "hashed" key, which may collide. The rest of the code deals with
1228 * situations like this:
1232 * | 3 | 5 | | 6 | 7 | (x)
1234 * Or more a complex example:
1238 * | 1 | 3 | | 5 | 8 |
1240 * | 5 | 5 | | 6 | 7 | (x)
1242 * In the examples, if we are looking for key "5", we may reach nodes
1243 * marked with "(x)". In this case what we have do is to look at the
1244 * left and see if there is "5" key there. If there is, we have to
1247 * Note, this whole situation is possible because we allow to have
1248 * elements which are equivalent to the next key in the parent in the
1249 * children of current znode. For example, this happens if we split a
1250 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1254 * | 3 | 5 | | 5 | 6 | 7 |
1256 * And this becomes what is at the first "picture" after key "5" marked
1257 * with "^" is removed. What could be done is we could prohibit
1258 * splitting in the middle of the colliding sequence. Also, when
1259 * removing the leftmost key, we would have to correct the key of the
1260 * parent node, which would introduce additional complications. Namely,
1261 * if we changed the leftmost key of the parent znode, the garbage
1262 * collector would be unable to find it (GC is doing this when GC'ing
1263 * indexing LEBs). Although we already have an additional RB-tree where
1264 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1265 * after the commit. But anyway, this does not look easy to implement
1266 * so we did not try this.
1268 err = tnc_prev(c, &znode, n);
1269 if (err == -ENOENT) {
1270 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1274 if (unlikely(err < 0))
1276 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1277 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1282 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1288 * lookup_level0_dirty - search for zero-level znode dirtying.
1289 * @c: UBIFS file-system description object
1290 * @key: key to lookup
1291 * @zn: znode is returned here
1292 * @n: znode branch slot number is returned here
1294 * This function looks up the TNC tree and search for zero-level znode which
1295 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1297 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1298 * is returned and slot number of the matched branch is stored in @n;
1299 * o not exact match, which means that zero-level znode does not contain @key
1300 * then %0 is returned and slot number of the closed branch is stored in
1302 * o @key is so small that it is even less than the lowest key of the
1303 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1305 * Additionally all znodes in the path from the root to the located zero-level
1306 * znode are marked as dirty.
1308 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1309 * function reads corresponding indexing nodes and inserts them to TNC. In
1310 * case of failure, a negative error code is returned.
1312 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1313 struct ubifs_znode **zn, int *n)
1316 struct ubifs_znode *znode;
1317 unsigned long time = get_seconds();
1319 dbg_tnck(key, "search and dirty key ");
1321 znode = c->zroot.znode;
1322 if (unlikely(!znode)) {
1323 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1325 return PTR_ERR(znode);
1328 znode = dirty_cow_znode(c, &c->zroot);
1330 return PTR_ERR(znode);
1335 struct ubifs_zbranch *zbr;
1337 exact = ubifs_search_zbranch(c, znode, key, n);
1339 if (znode->level == 0)
1344 zbr = &znode->zbranch[*n];
1348 znode = dirty_cow_znode(c, zbr);
1350 return PTR_ERR(znode);
1354 /* znode is not in TNC cache, load it from the media */
1355 znode = ubifs_load_znode(c, zbr, znode, *n);
1357 return PTR_ERR(znode);
1358 znode = dirty_cow_znode(c, zbr);
1360 return PTR_ERR(znode);
1364 if (exact || !is_hash_key(c, key) || *n != -1) {
1365 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1370 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1373 err = tnc_prev(c, &znode, n);
1374 if (err == -ENOENT) {
1376 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1379 if (unlikely(err < 0))
1381 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1383 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1387 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1388 znode = dirty_cow_bottom_up(c, znode);
1390 return PTR_ERR(znode);
1393 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1399 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1400 * @c: UBIFS file-system description object
1402 * @gc_seq1: garbage collection sequence number
1404 * This function determines if @lnum may have been garbage collected since
1405 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1408 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1410 int gc_seq2, gced_lnum;
1412 gced_lnum = c->gced_lnum;
1414 gc_seq2 = c->gc_seq;
1415 /* Same seq means no GC */
1416 if (gc_seq1 == gc_seq2)
1418 /* Different by more than 1 means we don't know */
1419 if (gc_seq1 + 1 != gc_seq2)
1422 * We have seen the sequence number has increased by 1. Now we need to
1423 * be sure we read the right LEB number, so read it again.
1426 if (gced_lnum != c->gced_lnum)
1428 /* Finally we can check lnum */
1429 if (gced_lnum == lnum)
1435 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1436 * @c: UBIFS file-system description object
1437 * @key: node key to lookup
1438 * @node: the node is returned here
1439 * @lnum: LEB number is returned here
1440 * @offs: offset is returned here
1442 * This function looks up and reads node with key @key. The caller has to make
1443 * sure the @node buffer is large enough to fit the node. Returns zero in case
1444 * of success, %-ENOENT if the node was not found, and a negative error code in
1445 * case of failure. The node location can be returned in @lnum and @offs.
1447 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1448 void *node, int *lnum, int *offs)
1450 int found, n, err, safely = 0, gc_seq1;
1451 struct ubifs_znode *znode;
1452 struct ubifs_zbranch zbr, *zt;
1455 mutex_lock(&c->tnc_mutex);
1456 found = ubifs_lookup_level0(c, key, &znode, &n);
1460 } else if (found < 0) {
1464 zt = &znode->zbranch[n];
1469 if (is_hash_key(c, key)) {
1471 * In this case the leaf node cache gets used, so we pass the
1472 * address of the zbranch and keep the mutex locked
1474 err = tnc_read_hashed_node(c, zt, node);
1478 err = ubifs_tnc_read_node(c, zt, node);
1481 /* Drop the TNC mutex prematurely and race with garbage collection */
1482 zbr = znode->zbranch[n];
1483 gc_seq1 = c->gc_seq;
1484 mutex_unlock(&c->tnc_mutex);
1486 if (ubifs_get_wbuf(c, zbr.lnum)) {
1487 /* We do not GC journal heads */
1488 err = ubifs_tnc_read_node(c, &zbr, node);
1492 err = fallible_read_node(c, key, &zbr, node);
1493 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1495 * The node may have been GC'ed out from under us so try again
1496 * while keeping the TNC mutex locked.
1504 mutex_unlock(&c->tnc_mutex);
1509 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1510 * @c: UBIFS file-system description object
1511 * @bu: bulk-read parameters and results
1513 * Lookup consecutive data node keys for the same inode that reside
1514 * consecutively in the same LEB. This function returns zero in case of success
1515 * and a negative error code in case of failure.
1517 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1518 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1519 * maximum possible amount of nodes for bulk-read.
1521 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1523 int n, err = 0, lnum = -1, uninitialized_var(offs);
1524 int uninitialized_var(len);
1525 unsigned int block = key_block(c, &bu->key);
1526 struct ubifs_znode *znode;
1532 mutex_lock(&c->tnc_mutex);
1533 /* Find first key */
1534 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1539 len = znode->zbranch[n].len;
1540 /* The buffer must be big enough for at least 1 node */
1541 if (len > bu->buf_len) {
1546 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1548 lnum = znode->zbranch[n].lnum;
1549 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1552 struct ubifs_zbranch *zbr;
1553 union ubifs_key *key;
1554 unsigned int next_block;
1557 err = tnc_next(c, &znode, &n);
1560 zbr = &znode->zbranch[n];
1562 /* See if there is another data key for this file */
1563 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1564 key_type(c, key) != UBIFS_DATA_KEY) {
1569 /* First key found */
1571 offs = ALIGN(zbr->offs + zbr->len, 8);
1573 if (len > bu->buf_len) {
1579 * The data nodes must be in consecutive positions in
1582 if (zbr->lnum != lnum || zbr->offs != offs)
1584 offs += ALIGN(zbr->len, 8);
1585 len = ALIGN(len, 8) + zbr->len;
1586 /* Must not exceed buffer length */
1587 if (len > bu->buf_len)
1590 /* Allow for holes */
1591 next_block = key_block(c, key);
1592 bu->blk_cnt += (next_block - block - 1);
1593 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1597 bu->zbranch[bu->cnt++] = *zbr;
1599 /* See if we have room for more */
1600 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1602 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1606 if (err == -ENOENT) {
1610 bu->gc_seq = c->gc_seq;
1611 mutex_unlock(&c->tnc_mutex);
1615 * An enormous hole could cause bulk-read to encompass too many
1616 * page cache pages, so limit the number here.
1618 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1619 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1621 * Ensure that bulk-read covers a whole number of page cache
1624 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1625 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1628 /* At the end of file we can round up */
1629 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1632 /* Exclude data nodes that do not make up a whole page cache page */
1633 block = key_block(c, &bu->key) + bu->blk_cnt;
1634 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1636 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1644 * read_wbuf - bulk-read from a LEB with a wbuf.
1645 * @wbuf: wbuf that may overlap the read
1646 * @buf: buffer into which to read
1648 * @lnum: LEB number from which to read
1649 * @offs: offset from which to read
1651 * This functions returns %0 on success or a negative error code on failure.
1653 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1656 const struct ubifs_info *c = wbuf->c;
1659 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1660 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1661 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1662 ubifs_assert(offs + len <= c->leb_size);
1664 spin_lock(&wbuf->lock);
1665 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1667 /* We may safely unlock the write-buffer and read the data */
1668 spin_unlock(&wbuf->lock);
1669 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1672 /* Don't read under wbuf */
1673 rlen = wbuf->offs - offs;
1677 /* Copy the rest from the write-buffer */
1678 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1679 spin_unlock(&wbuf->lock);
1682 /* Read everything that goes before write-buffer */
1683 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1689 * validate_data_node - validate data nodes for bulk-read.
1690 * @c: UBIFS file-system description object
1691 * @buf: buffer containing data node to validate
1692 * @zbr: zbranch of data node to validate
1694 * This functions returns %0 on success or a negative error code on failure.
1696 static int validate_data_node(struct ubifs_info *c, void *buf,
1697 struct ubifs_zbranch *zbr)
1699 union ubifs_key key1;
1700 struct ubifs_ch *ch = buf;
1703 if (ch->node_type != UBIFS_DATA_NODE) {
1704 ubifs_err(c, "bad node type (%d but expected %d)",
1705 ch->node_type, UBIFS_DATA_NODE);
1709 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1711 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1715 len = le32_to_cpu(ch->len);
1716 if (len != zbr->len) {
1717 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1721 /* Make sure the key of the read node is correct */
1722 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1723 if (!keys_eq(c, &zbr->key, &key1)) {
1724 ubifs_err(c, "bad key in node at LEB %d:%d",
1725 zbr->lnum, zbr->offs);
1726 dbg_tnck(&zbr->key, "looked for key ");
1727 dbg_tnck(&key1, "found node's key ");
1736 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1737 ubifs_dump_node(c, buf);
1743 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1744 * @c: UBIFS file-system description object
1745 * @bu: bulk-read parameters and results
1747 * This functions reads and validates the data nodes that were identified by the
1748 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1749 * -EAGAIN to indicate a race with GC, or another negative error code on
1752 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1754 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1755 struct ubifs_wbuf *wbuf;
1758 len = bu->zbranch[bu->cnt - 1].offs;
1759 len += bu->zbranch[bu->cnt - 1].len - offs;
1760 if (len > bu->buf_len) {
1761 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1766 wbuf = ubifs_get_wbuf(c, lnum);
1768 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1770 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1772 /* Check for a race with GC */
1773 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1776 if (err && err != -EBADMSG) {
1777 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1780 dbg_tnck(&bu->key, "key ");
1784 /* Validate the nodes read */
1786 for (i = 0; i < bu->cnt; i++) {
1787 err = validate_data_node(c, buf, &bu->zbranch[i]);
1790 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1797 * do_lookup_nm- look up a "hashed" node.
1798 * @c: UBIFS file-system description object
1799 * @key: node key to lookup
1800 * @node: the node is returned here
1803 * This function looks up and reads a node which contains name hash in the key.
1804 * Since the hash may have collisions, there may be many nodes with the same
1805 * key, so we have to sequentially look to all of them until the needed one is
1806 * found. This function returns zero in case of success, %-ENOENT if the node
1807 * was not found, and a negative error code in case of failure.
1809 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1810 void *node, const struct fscrypt_name *nm)
1813 struct ubifs_znode *znode;
1815 dbg_tnck(key, "key ");
1816 mutex_lock(&c->tnc_mutex);
1817 found = ubifs_lookup_level0(c, key, &znode, &n);
1821 } else if (found < 0) {
1826 ubifs_assert(n >= 0);
1828 err = resolve_collision(c, key, &znode, &n, nm);
1829 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1830 if (unlikely(err < 0))
1837 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1840 mutex_unlock(&c->tnc_mutex);
1845 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1846 * @c: UBIFS file-system description object
1847 * @key: node key to lookup
1848 * @node: the node is returned here
1851 * This function looks up and reads a node which contains name hash in the key.
1852 * Since the hash may have collisions, there may be many nodes with the same
1853 * key, so we have to sequentially look to all of them until the needed one is
1854 * found. This function returns zero in case of success, %-ENOENT if the node
1855 * was not found, and a negative error code in case of failure.
1857 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1858 void *node, const struct fscrypt_name *nm)
1861 const struct ubifs_dent_node *dent = node;
1864 * We assume that in most of the cases there are no name collisions and
1865 * 'ubifs_tnc_lookup()' returns us the right direntry.
1867 err = ubifs_tnc_lookup(c, key, node);
1871 len = le16_to_cpu(dent->nlen);
1872 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1876 * Unluckily, there are hash collisions and we have to iterate over
1877 * them look at each direntry with colliding name hash sequentially.
1880 return do_lookup_nm(c, key, node, nm);
1883 static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1884 struct ubifs_dent_node *dent, uint32_t cookie,
1885 struct ubifs_znode **zn, int *n, int exact)
1888 struct ubifs_znode *znode = *zn;
1889 struct ubifs_zbranch *zbr;
1890 union ubifs_key *dkey;
1893 err = tnc_next(c, &znode, n);
1899 zbr = &znode->zbranch[*n];
1902 if (key_inum(c, dkey) != key_inum(c, key) ||
1903 key_type(c, dkey) != key_type(c, key)) {
1907 err = tnc_read_hashed_node(c, zbr, dent);
1911 if (key_hash(c, key) == key_hash(c, dkey) &&
1912 le32_to_cpu(dent->cookie) == cookie) {
1917 err = tnc_next(c, &znode, n);
1923 static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1924 struct ubifs_dent_node *dent, uint32_t cookie)
1927 struct ubifs_znode *znode;
1928 union ubifs_key start_key;
1930 ubifs_assert(is_hash_key(c, key));
1932 lowest_dent_key(c, &start_key, key_inum(c, key));
1934 mutex_lock(&c->tnc_mutex);
1935 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1936 if (unlikely(err < 0))
1939 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
1942 mutex_unlock(&c->tnc_mutex);
1947 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1948 * @c: UBIFS file-system description object
1949 * @key: node key to lookup
1950 * @node: the node is returned here
1951 * @cookie: node cookie for collision resolution
1953 * This function looks up and reads a node which contains name hash in the key.
1954 * Since the hash may have collisions, there may be many nodes with the same
1955 * key, so we have to sequentially look to all of them until the needed one
1956 * with the same cookie value is found.
1957 * This function returns zero in case of success, %-ENOENT if the node
1958 * was not found, and a negative error code in case of failure.
1960 int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1961 void *node, uint32_t cookie)
1964 const struct ubifs_dent_node *dent = node;
1966 if (!c->double_hash)
1970 * We assume that in most of the cases there are no name collisions and
1971 * 'ubifs_tnc_lookup()' returns us the right direntry.
1973 err = ubifs_tnc_lookup(c, key, node);
1977 if (le32_to_cpu(dent->cookie) == cookie)
1981 * Unluckily, there are hash collisions and we have to iterate over
1982 * them look at each direntry with colliding name hash sequentially.
1984 return do_lookup_dh(c, key, node, cookie);
1988 * correct_parent_keys - correct parent znodes' keys.
1989 * @c: UBIFS file-system description object
1990 * @znode: znode to correct parent znodes for
1992 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1993 * zbranch changes, keys of parent znodes have to be corrected. This helper
1994 * function is called in such situations and corrects the keys if needed.
1996 static void correct_parent_keys(const struct ubifs_info *c,
1997 struct ubifs_znode *znode)
1999 union ubifs_key *key, *key1;
2001 ubifs_assert(znode->parent);
2002 ubifs_assert(znode->iip == 0);
2004 key = &znode->zbranch[0].key;
2005 key1 = &znode->parent->zbranch[0].key;
2007 while (keys_cmp(c, key, key1) < 0) {
2008 key_copy(c, key, key1);
2009 znode = znode->parent;
2011 if (!znode->parent || znode->iip)
2013 key1 = &znode->parent->zbranch[0].key;
2018 * insert_zbranch - insert a zbranch into a znode.
2019 * @znode: znode into which to insert
2020 * @zbr: zbranch to insert
2021 * @n: slot number to insert to
2023 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2024 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2025 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2026 * slot, zbranches starting from @n have to be moved right.
2028 static void insert_zbranch(struct ubifs_znode *znode,
2029 const struct ubifs_zbranch *zbr, int n)
2033 ubifs_assert(ubifs_zn_dirty(znode));
2036 for (i = znode->child_cnt; i > n; i--) {
2037 znode->zbranch[i] = znode->zbranch[i - 1];
2038 if (znode->zbranch[i].znode)
2039 znode->zbranch[i].znode->iip = i;
2042 zbr->znode->iip = n;
2044 for (i = znode->child_cnt; i > n; i--)
2045 znode->zbranch[i] = znode->zbranch[i - 1];
2047 znode->zbranch[n] = *zbr;
2048 znode->child_cnt += 1;
2051 * After inserting at slot zero, the lower bound of the key range of
2052 * this znode may have changed. If this znode is subsequently split
2053 * then the upper bound of the key range may change, and furthermore
2054 * it could change to be lower than the original lower bound. If that
2055 * happens, then it will no longer be possible to find this znode in the
2056 * TNC using the key from the index node on flash. That is bad because
2057 * if it is not found, we will assume it is obsolete and may overwrite
2058 * it. Then if there is an unclean unmount, we will start using the
2059 * old index which will be broken.
2061 * So we first mark znodes that have insertions at slot zero, and then
2062 * if they are split we add their lnum/offs to the old_idx tree.
2069 * tnc_insert - insert a node into TNC.
2070 * @c: UBIFS file-system description object
2071 * @znode: znode to insert into
2072 * @zbr: branch to insert
2073 * @n: slot number to insert new zbranch to
2075 * This function inserts a new node described by @zbr into znode @znode. If
2076 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2077 * are splat as well if needed. Returns zero in case of success or a negative
2078 * error code in case of failure.
2080 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2081 struct ubifs_zbranch *zbr, int n)
2083 struct ubifs_znode *zn, *zi, *zp;
2084 int i, keep, move, appending = 0;
2085 union ubifs_key *key = &zbr->key, *key1;
2087 ubifs_assert(n >= 0 && n <= c->fanout);
2089 /* Implement naive insert for now */
2092 if (znode->child_cnt < c->fanout) {
2093 ubifs_assert(n != c->fanout);
2094 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2096 insert_zbranch(znode, zbr, n);
2098 /* Ensure parent's key is correct */
2099 if (n == 0 && zp && znode->iip == 0)
2100 correct_parent_keys(c, znode);
2106 * Unfortunately, @znode does not have more empty slots and we have to
2109 dbg_tnck(key, "splitting level %d, key ", znode->level);
2113 * We can no longer be sure of finding this znode by key, so we
2114 * record it in the old_idx tree.
2116 ins_clr_old_idx_znode(c, znode);
2118 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2122 zn->level = znode->level;
2124 /* Decide where to split */
2125 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2126 /* Try not to split consecutive data keys */
2127 if (n == c->fanout) {
2128 key1 = &znode->zbranch[n - 1].key;
2129 if (key_inum(c, key1) == key_inum(c, key) &&
2130 key_type(c, key1) == UBIFS_DATA_KEY)
2134 } else if (appending && n != c->fanout) {
2135 /* Try not to split consecutive data keys */
2138 if (n >= (c->fanout + 1) / 2) {
2139 key1 = &znode->zbranch[0].key;
2140 if (key_inum(c, key1) == key_inum(c, key) &&
2141 key_type(c, key1) == UBIFS_DATA_KEY) {
2142 key1 = &znode->zbranch[n].key;
2143 if (key_inum(c, key1) != key_inum(c, key) ||
2144 key_type(c, key1) != UBIFS_DATA_KEY) {
2146 move = c->fanout - keep;
2158 keep = (c->fanout + 1) / 2;
2159 move = c->fanout - keep;
2163 * Although we don't at present, we could look at the neighbors and see
2164 * if we can move some zbranches there.
2168 /* Insert into existing znode */
2173 /* Insert into new znode */
2178 zbr->znode->parent = zn;
2183 __set_bit(DIRTY_ZNODE, &zn->flags);
2184 atomic_long_inc(&c->dirty_zn_cnt);
2186 zn->child_cnt = move;
2187 znode->child_cnt = keep;
2189 dbg_tnc("moving %d, keeping %d", move, keep);
2192 for (i = 0; i < move; i++) {
2193 zn->zbranch[i] = znode->zbranch[keep + i];
2196 if (zn->zbranch[i].znode) {
2197 zn->zbranch[i].znode->parent = zn;
2198 zn->zbranch[i].znode->iip = i;
2202 /* Insert new key and branch */
2203 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2205 insert_zbranch(zi, zbr, n);
2207 /* Insert new znode (produced by spitting) into the parent */
2209 if (n == 0 && zi == znode && znode->iip == 0)
2210 correct_parent_keys(c, znode);
2212 /* Locate insertion point */
2215 /* Tail recursion */
2216 zbr->key = zn->zbranch[0].key;
2226 /* We have to split root znode */
2227 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2229 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2234 zi->level = znode->level + 1;
2236 __set_bit(DIRTY_ZNODE, &zi->flags);
2237 atomic_long_inc(&c->dirty_zn_cnt);
2239 zi->zbranch[0].key = znode->zbranch[0].key;
2240 zi->zbranch[0].znode = znode;
2241 zi->zbranch[0].lnum = c->zroot.lnum;
2242 zi->zbranch[0].offs = c->zroot.offs;
2243 zi->zbranch[0].len = c->zroot.len;
2244 zi->zbranch[1].key = zn->zbranch[0].key;
2245 zi->zbranch[1].znode = zn;
2250 c->zroot.znode = zi;
2261 * ubifs_tnc_add - add a node to TNC.
2262 * @c: UBIFS file-system description object
2264 * @lnum: LEB number of node
2265 * @offs: node offset
2268 * This function adds a node with key @key to TNC. The node may be new or it may
2269 * obsolete some existing one. Returns %0 on success or negative error code on
2272 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2275 int found, n, err = 0;
2276 struct ubifs_znode *znode;
2278 mutex_lock(&c->tnc_mutex);
2279 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2280 found = lookup_level0_dirty(c, key, &znode, &n);
2282 struct ubifs_zbranch zbr;
2288 key_copy(c, key, &zbr.key);
2289 err = tnc_insert(c, znode, &zbr, n + 1);
2290 } else if (found == 1) {
2291 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2294 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2301 err = dbg_check_tnc(c, 0);
2302 mutex_unlock(&c->tnc_mutex);
2308 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2309 * @c: UBIFS file-system description object
2311 * @old_lnum: LEB number of old node
2312 * @old_offs: old node offset
2313 * @lnum: LEB number of node
2314 * @offs: node offset
2317 * This function replaces a node with key @key in the TNC only if the old node
2318 * is found. This function is called by garbage collection when node are moved.
2319 * Returns %0 on success or negative error code on failure.
2321 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2322 int old_lnum, int old_offs, int lnum, int offs, int len)
2324 int found, n, err = 0;
2325 struct ubifs_znode *znode;
2327 mutex_lock(&c->tnc_mutex);
2328 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2329 old_offs, lnum, offs, len);
2330 found = lookup_level0_dirty(c, key, &znode, &n);
2337 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2340 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2342 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2349 } else if (is_hash_key(c, key)) {
2350 found = resolve_collision_directly(c, key, &znode, &n,
2351 old_lnum, old_offs);
2352 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2353 found, znode, n, old_lnum, old_offs);
2360 /* Ensure the znode is dirtied */
2361 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2362 znode = dirty_cow_bottom_up(c, znode);
2363 if (IS_ERR(znode)) {
2364 err = PTR_ERR(znode);
2368 zbr = &znode->zbranch[n];
2370 err = ubifs_add_dirt(c, zbr->lnum,
2382 err = ubifs_add_dirt(c, lnum, len);
2385 err = dbg_check_tnc(c, 0);
2388 mutex_unlock(&c->tnc_mutex);
2393 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2394 * @c: UBIFS file-system description object
2396 * @lnum: LEB number of node
2397 * @offs: node offset
2401 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2402 * may have collisions, like directory entry keys.
2404 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2405 int lnum, int offs, int len,
2406 const struct fscrypt_name *nm)
2408 int found, n, err = 0;
2409 struct ubifs_znode *znode;
2411 mutex_lock(&c->tnc_mutex);
2412 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2413 found = lookup_level0_dirty(c, key, &znode, &n);
2421 found = fallible_resolve_collision(c, key, &znode, &n,
2424 found = resolve_collision(c, key, &znode, &n, nm);
2425 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2431 /* Ensure the znode is dirtied */
2432 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2433 znode = dirty_cow_bottom_up(c, znode);
2434 if (IS_ERR(znode)) {
2435 err = PTR_ERR(znode);
2441 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2444 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2453 struct ubifs_zbranch zbr;
2459 key_copy(c, key, &zbr.key);
2460 err = tnc_insert(c, znode, &zbr, n + 1);
2465 * We did not find it in the index so there may be a
2466 * dangling branch still in the index. So we remove it
2467 * by passing 'ubifs_tnc_remove_nm()' the same key but
2468 * an unmatchable name.
2470 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2472 err = dbg_check_tnc(c, 0);
2473 mutex_unlock(&c->tnc_mutex);
2476 return ubifs_tnc_remove_nm(c, key, &noname);
2482 err = dbg_check_tnc(c, 0);
2483 mutex_unlock(&c->tnc_mutex);
2488 * tnc_delete - delete a znode form TNC.
2489 * @c: UBIFS file-system description object
2490 * @znode: znode to delete from
2491 * @n: zbranch slot number to delete
2493 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2494 * case of success and a negative error code in case of failure.
2496 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2498 struct ubifs_zbranch *zbr;
2499 struct ubifs_znode *zp;
2502 /* Delete without merge for now */
2503 ubifs_assert(znode->level == 0);
2504 ubifs_assert(n >= 0 && n < c->fanout);
2505 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2507 zbr = &znode->zbranch[n];
2510 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2512 ubifs_dump_znode(c, znode);
2516 /* We do not "gap" zbranch slots */
2517 for (i = n; i < znode->child_cnt - 1; i++)
2518 znode->zbranch[i] = znode->zbranch[i + 1];
2519 znode->child_cnt -= 1;
2521 if (znode->child_cnt > 0)
2525 * This was the last zbranch, we have to delete this znode from the
2530 ubifs_assert(!ubifs_zn_obsolete(znode));
2531 ubifs_assert(ubifs_zn_dirty(znode));
2536 atomic_long_dec(&c->dirty_zn_cnt);
2538 err = insert_old_idx_znode(c, znode);
2543 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2544 atomic_long_inc(&c->clean_zn_cnt);
2545 atomic_long_inc(&ubifs_clean_zn_cnt);
2549 } while (znode->child_cnt == 1); /* while removing last child */
2551 /* Remove from znode, entry n - 1 */
2552 znode->child_cnt -= 1;
2553 ubifs_assert(znode->level != 0);
2554 for (i = n; i < znode->child_cnt; i++) {
2555 znode->zbranch[i] = znode->zbranch[i + 1];
2556 if (znode->zbranch[i].znode)
2557 znode->zbranch[i].znode->iip = i;
2561 * If this is the root and it has only 1 child then
2562 * collapse the tree.
2564 if (!znode->parent) {
2565 while (znode->child_cnt == 1 && znode->level != 0) {
2567 zbr = &znode->zbranch[0];
2568 znode = get_znode(c, znode, 0);
2570 return PTR_ERR(znode);
2571 znode = dirty_cow_znode(c, zbr);
2573 return PTR_ERR(znode);
2574 znode->parent = NULL;
2577 err = insert_old_idx(c, c->zroot.lnum,
2582 c->zroot.lnum = zbr->lnum;
2583 c->zroot.offs = zbr->offs;
2584 c->zroot.len = zbr->len;
2585 c->zroot.znode = znode;
2586 ubifs_assert(!ubifs_zn_obsolete(zp));
2587 ubifs_assert(ubifs_zn_dirty(zp));
2588 atomic_long_dec(&c->dirty_zn_cnt);
2591 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2592 atomic_long_inc(&c->clean_zn_cnt);
2593 atomic_long_inc(&ubifs_clean_zn_cnt);
2603 * ubifs_tnc_remove - remove an index entry of a node.
2604 * @c: UBIFS file-system description object
2607 * Returns %0 on success or negative error code on failure.
2609 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2611 int found, n, err = 0;
2612 struct ubifs_znode *znode;
2614 mutex_lock(&c->tnc_mutex);
2615 dbg_tnck(key, "key ");
2616 found = lookup_level0_dirty(c, key, &znode, &n);
2622 err = tnc_delete(c, znode, n);
2624 err = dbg_check_tnc(c, 0);
2627 mutex_unlock(&c->tnc_mutex);
2632 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2633 * @c: UBIFS file-system description object
2635 * @nm: directory entry name
2637 * Returns %0 on success or negative error code on failure.
2639 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2640 const struct fscrypt_name *nm)
2643 struct ubifs_znode *znode;
2645 mutex_lock(&c->tnc_mutex);
2646 dbg_tnck(key, "key ");
2647 err = lookup_level0_dirty(c, key, &znode, &n);
2653 err = fallible_resolve_collision(c, key, &znode, &n,
2656 err = resolve_collision(c, key, &znode, &n, nm);
2657 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2661 /* Ensure the znode is dirtied */
2662 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2663 znode = dirty_cow_bottom_up(c, znode);
2664 if (IS_ERR(znode)) {
2665 err = PTR_ERR(znode);
2669 err = tnc_delete(c, znode, n);
2675 err = dbg_check_tnc(c, 0);
2676 mutex_unlock(&c->tnc_mutex);
2681 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2682 * @c: UBIFS file-system description object
2684 * @cookie: node cookie for collision resolution
2686 * Returns %0 on success or negative error code on failure.
2688 int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2692 struct ubifs_znode *znode;
2693 struct ubifs_dent_node *dent;
2694 struct ubifs_zbranch *zbr;
2696 if (!c->double_hash)
2699 mutex_lock(&c->tnc_mutex);
2700 err = lookup_level0_dirty(c, key, &znode, &n);
2704 zbr = &znode->zbranch[n];
2705 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2711 err = tnc_read_hashed_node(c, zbr, dent);
2715 /* If the cookie does not match, we're facing a hash collision. */
2716 if (le32_to_cpu(dent->cookie) != cookie) {
2717 union ubifs_key start_key;
2719 lowest_dent_key(c, &start_key, key_inum(c, key));
2721 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2722 if (unlikely(err < 0))
2725 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
2730 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2731 znode = dirty_cow_bottom_up(c, znode);
2732 if (IS_ERR(znode)) {
2733 err = PTR_ERR(znode);
2737 err = tnc_delete(c, znode, n);
2743 err = dbg_check_tnc(c, 0);
2744 mutex_unlock(&c->tnc_mutex);
2749 * key_in_range - determine if a key falls within a range of keys.
2750 * @c: UBIFS file-system description object
2751 * @key: key to check
2752 * @from_key: lowest key in range
2753 * @to_key: highest key in range
2755 * This function returns %1 if the key is in range and %0 otherwise.
2757 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2758 union ubifs_key *from_key, union ubifs_key *to_key)
2760 if (keys_cmp(c, key, from_key) < 0)
2762 if (keys_cmp(c, key, to_key) > 0)
2768 * ubifs_tnc_remove_range - remove index entries in range.
2769 * @c: UBIFS file-system description object
2770 * @from_key: lowest key to remove
2771 * @to_key: highest key to remove
2773 * This function removes index entries starting at @from_key and ending at
2774 * @to_key. This function returns zero in case of success and a negative error
2775 * code in case of failure.
2777 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2778 union ubifs_key *to_key)
2780 int i, n, k, err = 0;
2781 struct ubifs_znode *znode;
2782 union ubifs_key *key;
2784 mutex_lock(&c->tnc_mutex);
2786 /* Find first level 0 znode that contains keys to remove */
2787 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2794 err = tnc_next(c, &znode, &n);
2795 if (err == -ENOENT) {
2801 key = &znode->zbranch[n].key;
2802 if (!key_in_range(c, key, from_key, to_key)) {
2808 /* Ensure the znode is dirtied */
2809 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2810 znode = dirty_cow_bottom_up(c, znode);
2811 if (IS_ERR(znode)) {
2812 err = PTR_ERR(znode);
2817 /* Remove all keys in range except the first */
2818 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2819 key = &znode->zbranch[i].key;
2820 if (!key_in_range(c, key, from_key, to_key))
2822 lnc_free(&znode->zbranch[i]);
2823 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2824 znode->zbranch[i].len);
2826 ubifs_dump_znode(c, znode);
2829 dbg_tnck(key, "removing key ");
2832 for (i = n + 1 + k; i < znode->child_cnt; i++)
2833 znode->zbranch[i - k] = znode->zbranch[i];
2834 znode->child_cnt -= k;
2837 /* Now delete the first */
2838 err = tnc_delete(c, znode, n);
2845 err = dbg_check_tnc(c, 0);
2846 mutex_unlock(&c->tnc_mutex);
2851 * ubifs_tnc_remove_ino - remove an inode from TNC.
2852 * @c: UBIFS file-system description object
2853 * @inum: inode number to remove
2855 * This function remove inode @inum and all the extended attributes associated
2856 * with the anode from TNC and returns zero in case of success or a negative
2857 * error code in case of failure.
2859 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2861 union ubifs_key key1, key2;
2862 struct ubifs_dent_node *xent, *pxent = NULL;
2863 struct fscrypt_name nm = {0};
2865 dbg_tnc("ino %lu", (unsigned long)inum);
2868 * Walk all extended attribute entries and remove them together with
2869 * corresponding extended attribute inodes.
2871 lowest_xent_key(c, &key1, inum);
2876 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2878 err = PTR_ERR(xent);
2884 xattr_inum = le64_to_cpu(xent->inum);
2885 dbg_tnc("xent '%s', ino %lu", xent->name,
2886 (unsigned long)xattr_inum);
2888 ubifs_evict_xattr_inode(c, xattr_inum);
2890 fname_name(&nm) = xent->name;
2891 fname_len(&nm) = le16_to_cpu(xent->nlen);
2892 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2898 lowest_ino_key(c, &key1, xattr_inum);
2899 highest_ino_key(c, &key2, xattr_inum);
2900 err = ubifs_tnc_remove_range(c, &key1, &key2);
2908 key_read(c, &xent->key, &key1);
2912 lowest_ino_key(c, &key1, inum);
2913 highest_ino_key(c, &key2, inum);
2915 return ubifs_tnc_remove_range(c, &key1, &key2);
2919 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2920 * @c: UBIFS file-system description object
2921 * @key: key of last entry
2922 * @nm: name of last entry found or %NULL
2924 * This function finds and reads the next directory or extended attribute entry
2925 * after the given key (@key) if there is one. @nm is used to resolve
2928 * If the name of the current entry is not known and only the key is known,
2929 * @nm->name has to be %NULL. In this case the semantics of this function is a
2930 * little bit different and it returns the entry corresponding to this key, not
2931 * the next one. If the key was not found, the closest "right" entry is
2934 * If the fist entry has to be found, @key has to contain the lowest possible
2935 * key value for this inode and @name has to be %NULL.
2937 * This function returns the found directory or extended attribute entry node
2938 * in case of success, %-ENOENT is returned if no entry was found, and a
2939 * negative error code is returned in case of failure.
2941 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2942 union ubifs_key *key,
2943 const struct fscrypt_name *nm)
2945 int n, err, type = key_type(c, key);
2946 struct ubifs_znode *znode;
2947 struct ubifs_dent_node *dent;
2948 struct ubifs_zbranch *zbr;
2949 union ubifs_key *dkey;
2951 dbg_tnck(key, "key ");
2952 ubifs_assert(is_hash_key(c, key));
2954 mutex_lock(&c->tnc_mutex);
2955 err = ubifs_lookup_level0(c, key, &znode, &n);
2956 if (unlikely(err < 0))
2959 if (fname_len(nm) > 0) {
2961 /* Handle collisions */
2963 err = fallible_resolve_collision(c, key, &znode, &n,
2966 err = resolve_collision(c, key, &znode, &n, nm);
2967 dbg_tnc("rc returned %d, znode %p, n %d",
2969 if (unlikely(err < 0))
2973 /* Now find next entry */
2974 err = tnc_next(c, &znode, &n);
2979 * The full name of the entry was not given, in which case the
2980 * behavior of this function is a little different and it
2981 * returns current entry, not the next one.
2985 * However, the given key does not exist in the TNC
2986 * tree and @znode/@n variables contain the closest
2987 * "preceding" element. Switch to the next one.
2989 err = tnc_next(c, &znode, &n);
2995 zbr = &znode->zbranch[n];
2996 dent = kmalloc(zbr->len, GFP_NOFS);
2997 if (unlikely(!dent)) {
3003 * The above 'tnc_next()' call could lead us to the next inode, check
3007 if (key_inum(c, dkey) != key_inum(c, key) ||
3008 key_type(c, dkey) != type) {
3013 err = tnc_read_hashed_node(c, zbr, dent);
3017 mutex_unlock(&c->tnc_mutex);
3023 mutex_unlock(&c->tnc_mutex);
3024 return ERR_PTR(err);
3028 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3029 * @c: UBIFS file-system description object
3031 * Destroy left-over obsolete znodes from a failed commit.
3033 static void tnc_destroy_cnext(struct ubifs_info *c)
3035 struct ubifs_znode *cnext;
3039 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
3042 struct ubifs_znode *znode = cnext;
3044 cnext = cnext->cnext;
3045 if (ubifs_zn_obsolete(znode))
3047 } while (cnext && cnext != c->cnext);
3051 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3052 * @c: UBIFS file-system description object
3054 void ubifs_tnc_close(struct ubifs_info *c)
3056 tnc_destroy_cnext(c);
3057 if (c->zroot.znode) {
3060 n = atomic_long_read(&c->clean_zn_cnt);
3061 freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
3062 ubifs_assert(freed == n);
3063 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3071 * left_znode - get the znode to the left.
3072 * @c: UBIFS file-system description object
3075 * This function returns a pointer to the znode to the left of @znode or NULL if
3076 * there is not one. A negative error code is returned on failure.
3078 static struct ubifs_znode *left_znode(struct ubifs_info *c,
3079 struct ubifs_znode *znode)
3081 int level = znode->level;
3084 int n = znode->iip - 1;
3086 /* Go up until we can go left */
3087 znode = znode->parent;
3091 /* Now go down the rightmost branch to 'level' */
3092 znode = get_znode(c, znode, n);
3095 while (znode->level != level) {
3096 n = znode->child_cnt - 1;
3097 znode = get_znode(c, znode, n);
3108 * right_znode - get the znode to the right.
3109 * @c: UBIFS file-system description object
3112 * This function returns a pointer to the znode to the right of @znode or NULL
3113 * if there is not one. A negative error code is returned on failure.
3115 static struct ubifs_znode *right_znode(struct ubifs_info *c,
3116 struct ubifs_znode *znode)
3118 int level = znode->level;
3121 int n = znode->iip + 1;
3123 /* Go up until we can go right */
3124 znode = znode->parent;
3127 if (n < znode->child_cnt) {
3128 /* Now go down the leftmost branch to 'level' */
3129 znode = get_znode(c, znode, n);
3132 while (znode->level != level) {
3133 znode = get_znode(c, znode, 0);
3144 * lookup_znode - find a particular indexing node from TNC.
3145 * @c: UBIFS file-system description object
3146 * @key: index node key to lookup
3147 * @level: index node level
3148 * @lnum: index node LEB number
3149 * @offs: index node offset
3151 * This function searches an indexing node by its first key @key and its
3152 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3153 * nodes it traverses to TNC. This function is called for indexing nodes which
3154 * were found on the media by scanning, for example when garbage-collecting or
3155 * when doing in-the-gaps commit. This means that the indexing node which is
3156 * looked for does not have to have exactly the same leftmost key @key, because
3157 * the leftmost key may have been changed, in which case TNC will contain a
3158 * dirty znode which still refers the same @lnum:@offs. This function is clever
3159 * enough to recognize such indexing nodes.
3161 * Note, if a znode was deleted or changed too much, then this function will
3162 * not find it. For situations like this UBIFS has the old index RB-tree
3163 * (indexed by @lnum:@offs).
3165 * This function returns a pointer to the znode found or %NULL if it is not
3166 * found. A negative error code is returned on failure.
3168 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3169 union ubifs_key *key, int level,
3172 struct ubifs_znode *znode, *zn;
3175 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
3178 * The arguments have probably been read off flash, so don't assume
3182 return ERR_PTR(-EINVAL);
3184 /* Get the root znode */
3185 znode = c->zroot.znode;
3187 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3191 /* Check if it is the one we are looking for */
3192 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3194 /* Descend to the parent level i.e. (level + 1) */
3195 if (level >= znode->level)
3198 ubifs_search_zbranch(c, znode, key, &n);
3201 * We reached a znode where the leftmost key is greater
3202 * than the key we are searching for. This is the same
3203 * situation as the one described in a huge comment at
3204 * the end of the 'ubifs_lookup_level0()' function. And
3205 * for exactly the same reasons we have to try to look
3206 * left before giving up.
3208 znode = left_znode(c, znode);
3213 ubifs_search_zbranch(c, znode, key, &n);
3214 ubifs_assert(n >= 0);
3216 if (znode->level == level + 1)
3218 znode = get_znode(c, znode, n);
3222 /* Check if the child is the one we are looking for */
3223 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3224 return get_znode(c, znode, n);
3225 /* If the key is unique, there is nowhere else to look */
3226 if (!is_hash_key(c, key))
3229 * The key is not unique and so may be also in the znodes to either
3236 /* Move one branch to the left */
3240 znode = left_znode(c, znode);
3245 n = znode->child_cnt - 1;
3248 if (znode->zbranch[n].lnum == lnum &&
3249 znode->zbranch[n].offs == offs)
3250 return get_znode(c, znode, n);
3251 /* Stop if the key is less than the one we are looking for */
3252 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3255 /* Back to the middle */
3260 /* Move one branch to the right */
3261 if (++n >= znode->child_cnt) {
3262 znode = right_znode(c, znode);
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 greater than the one we are looking for */
3274 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3281 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3282 * @c: UBIFS file-system description object
3283 * @key: key of index node
3284 * @level: index node level
3285 * @lnum: LEB number of index node
3286 * @offs: offset of index node
3288 * This function returns %0 if the index node is not referred to in the TNC, %1
3289 * if the index node is referred to in the TNC and the corresponding znode is
3290 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3291 * znode is clean, and a negative error code in case of failure.
3293 * Note, the @key argument has to be the key of the first child. Also note,
3294 * this function relies on the fact that 0:0 is never a valid LEB number and
3295 * offset for a main-area node.
3297 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3300 struct ubifs_znode *znode;
3302 znode = lookup_znode(c, key, level, lnum, offs);
3306 return PTR_ERR(znode);
3308 return ubifs_zn_dirty(znode) ? 1 : 2;
3312 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3313 * @c: UBIFS file-system description object
3315 * @lnum: node LEB number
3316 * @offs: node offset
3318 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3319 * not, and a negative error code in case of failure.
3321 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3322 * and offset for a main-area node.
3324 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3327 struct ubifs_zbranch *zbr;
3328 struct ubifs_znode *znode, *zn;
3329 int n, found, err, nn;
3330 const int unique = !is_hash_key(c, key);
3332 found = ubifs_lookup_level0(c, key, &znode, &n);
3334 return found; /* Error code */
3337 zbr = &znode->zbranch[n];
3338 if (lnum == zbr->lnum && offs == zbr->offs)
3339 return 1; /* Found it */
3343 * Because the key is not unique, we have to look left
3350 err = tnc_prev(c, &znode, &n);
3355 if (keys_cmp(c, key, &znode->zbranch[n].key))
3357 zbr = &znode->zbranch[n];
3358 if (lnum == zbr->lnum && offs == zbr->offs)
3359 return 1; /* Found it */
3365 err = tnc_next(c, &znode, &n);
3371 if (keys_cmp(c, key, &znode->zbranch[n].key))
3373 zbr = &znode->zbranch[n];
3374 if (lnum == zbr->lnum && offs == zbr->offs)
3375 return 1; /* Found it */
3381 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3382 * @c: UBIFS file-system description object
3384 * @level: index node level (if it is an index node)
3385 * @lnum: node LEB number
3386 * @offs: node offset
3387 * @is_idx: non-zero if the node is an index node
3389 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3390 * negative error code in case of failure. For index nodes, @key has to be the
3391 * key of the first child. An index node is considered to be in the TNC only if
3392 * the corresponding znode is clean or has not been loaded.
3394 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3395 int lnum, int offs, int is_idx)
3399 mutex_lock(&c->tnc_mutex);
3401 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3405 /* The index node was found but it was dirty */
3408 /* The index node was found and it was clean */
3413 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3416 mutex_unlock(&c->tnc_mutex);
3421 * ubifs_dirty_idx_node - dirty an index node.
3422 * @c: UBIFS file-system description object
3423 * @key: index node key
3424 * @level: index node level
3425 * @lnum: index node LEB number
3426 * @offs: index node offset
3428 * This function loads and dirties an index node so that it can be garbage
3429 * collected. The @key argument has to be the key of the first child. This
3430 * function relies on the fact that 0:0 is never a valid LEB number and offset
3431 * for a main-area node. Returns %0 on success and a negative error code on
3434 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3437 struct ubifs_znode *znode;
3440 mutex_lock(&c->tnc_mutex);
3441 znode = lookup_znode(c, key, level, lnum, offs);
3444 if (IS_ERR(znode)) {
3445 err = PTR_ERR(znode);
3448 znode = dirty_cow_bottom_up(c, znode);
3449 if (IS_ERR(znode)) {
3450 err = PTR_ERR(znode);
3455 mutex_unlock(&c->tnc_mutex);
3460 * dbg_check_inode_size - check if inode size is correct.
3461 * @c: UBIFS file-system description object
3462 * @inum: inode number
3465 * This function makes sure that the inode size (@size) is correct and it does
3466 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3467 * if it has a data page beyond @size, and other negative error code in case of
3470 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3474 union ubifs_key from_key, to_key, *key;
3475 struct ubifs_znode *znode;
3478 if (!S_ISREG(inode->i_mode))
3480 if (!dbg_is_chk_gen(c))
3483 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3484 data_key_init(c, &from_key, inode->i_ino, block);
3485 highest_data_key(c, &to_key, inode->i_ino);
3487 mutex_lock(&c->tnc_mutex);
3488 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3497 err = tnc_next(c, &znode, &n);
3498 if (err == -ENOENT) {
3505 ubifs_assert(err == 0);
3506 key = &znode->zbranch[n].key;
3507 if (!key_in_range(c, key, &from_key, &to_key))
3511 block = key_block(c, key);
3512 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3513 (unsigned long)inode->i_ino, size,
3514 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3515 mutex_unlock(&c->tnc_mutex);
3516 ubifs_dump_inode(c, inode);
3521 mutex_unlock(&c->tnc_mutex);