1 // SPDX-License-Identifier: GPL-2.0+
3 * This file is part of UBIFS.
5 * Copyright (C) 2006-2008 Nokia Corporation.
7 * Authors: Adrian Hunter
8 * Artem Bityutskiy (Битюцкий Артём)
12 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
15 * At the moment the locking rules of the TNC tree are quite simple and
16 * straightforward. We just have a mutex and lock it when we traverse the
17 * tree. If a znode is not in memory, we read it from flash while still having
23 #include <dm/devres.h>
24 #include <linux/crc32.h>
25 #include <linux/slab.h>
26 #include <u-boot/crc.h>
28 #include <linux/bitops.h>
29 #include <linux/bug.h>
30 #include <linux/compat.h>
31 #include <linux/err.h>
32 #include <linux/stat.h>
37 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
38 * @NAME_LESS: name corresponding to the first argument is less than second
39 * @NAME_MATCHES: names match
40 * @NAME_GREATER: name corresponding to the second argument is greater than
42 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
44 * These constants were introduce to improve readability.
54 * insert_old_idx - record an index node obsoleted since the last commit start.
55 * @c: UBIFS file-system description object
56 * @lnum: LEB number of obsoleted index node
57 * @offs: offset of obsoleted index node
59 * Returns %0 on success, and a negative error code on failure.
61 * For recovery, there must always be a complete intact version of the index on
62 * flash at all times. That is called the "old index". It is the index as at the
63 * time of the last successful commit. Many of the index nodes in the old index
64 * may be dirty, but they must not be erased until the next successful commit
65 * (at which point that index becomes the old index).
67 * That means that the garbage collection and the in-the-gaps method of
68 * committing must be able to determine if an index node is in the old index.
69 * Most of the old index nodes can be found by looking up the TNC using the
70 * 'lookup_znode()' function. However, some of the old index nodes may have
71 * been deleted from the current index or may have been changed so much that
72 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
73 * That is what this function does. The RB-tree is ordered by LEB number and
74 * offset because they uniquely identify the old index node.
76 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
78 struct ubifs_old_idx *old_idx, *o;
79 struct rb_node **p, *parent = NULL;
81 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
82 if (unlikely(!old_idx))
87 p = &c->old_idx.rb_node;
90 o = rb_entry(parent, struct ubifs_old_idx, rb);
93 else if (lnum > o->lnum)
95 else if (offs < o->offs)
97 else if (offs > o->offs)
100 ubifs_err(c, "old idx added twice!");
105 rb_link_node(&old_idx->rb, parent, p);
106 rb_insert_color(&old_idx->rb, &c->old_idx);
111 * insert_old_idx_znode - record a znode obsoleted since last commit start.
112 * @c: UBIFS file-system description object
113 * @znode: znode of obsoleted index node
115 * Returns %0 on success, and a negative error code on failure.
117 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
120 struct ubifs_zbranch *zbr;
122 zbr = &znode->parent->zbranch[znode->iip];
124 return insert_old_idx(c, zbr->lnum, zbr->offs);
127 return insert_old_idx(c, c->zroot.lnum,
133 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
134 * @c: UBIFS file-system description object
135 * @znode: znode of obsoleted index node
137 * Returns %0 on success, and a negative error code on failure.
139 static int ins_clr_old_idx_znode(struct ubifs_info *c,
140 struct ubifs_znode *znode)
145 struct ubifs_zbranch *zbr;
147 zbr = &znode->parent->zbranch[znode->iip];
149 err = insert_old_idx(c, zbr->lnum, zbr->offs);
158 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
169 * destroy_old_idx - destroy the old_idx RB-tree.
170 * @c: UBIFS file-system description object
172 * During start commit, the old_idx RB-tree is used to avoid overwriting index
173 * nodes that were in the index last commit but have since been deleted. This
174 * is necessary for recovery i.e. the old index must be kept intact until the
175 * new index is successfully written. The old-idx RB-tree is used for the
176 * in-the-gaps method of writing index nodes and is destroyed every commit.
178 void destroy_old_idx(struct ubifs_info *c)
180 struct ubifs_old_idx *old_idx, *n;
182 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
185 c->old_idx = RB_ROOT;
189 * copy_znode - copy a dirty znode.
190 * @c: UBIFS file-system description object
191 * @znode: znode to copy
193 * A dirty znode being committed may not be changed, so it is copied.
195 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
196 struct ubifs_znode *znode)
198 struct ubifs_znode *zn;
200 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
202 return ERR_PTR(-ENOMEM);
204 memcpy(zn, znode, c->max_znode_sz);
206 __set_bit(DIRTY_ZNODE, &zn->flags);
207 __clear_bit(COW_ZNODE, &zn->flags);
209 ubifs_assert(!ubifs_zn_obsolete(znode));
210 __set_bit(OBSOLETE_ZNODE, &znode->flags);
212 if (znode->level != 0) {
214 const int n = zn->child_cnt;
216 /* The children now have new parent */
217 for (i = 0; i < n; i++) {
218 struct ubifs_zbranch *zbr = &zn->zbranch[i];
221 zbr->znode->parent = zn;
225 atomic_long_inc(&c->dirty_zn_cnt);
230 * add_idx_dirt - add dirt due to a dirty znode.
231 * @c: UBIFS file-system description object
232 * @lnum: LEB number of index node
233 * @dirt: size of index node
235 * This function updates lprops dirty space and the new size of the index.
237 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
239 c->calc_idx_sz -= ALIGN(dirt, 8);
240 return ubifs_add_dirt(c, lnum, dirt);
244 * dirty_cow_znode - ensure a znode is not being committed.
245 * @c: UBIFS file-system description object
246 * @zbr: branch of znode to check
248 * Returns dirtied znode on success or negative error code on failure.
250 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
251 struct ubifs_zbranch *zbr)
253 struct ubifs_znode *znode = zbr->znode;
254 struct ubifs_znode *zn;
257 if (!ubifs_zn_cow(znode)) {
258 /* znode is not being committed */
259 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
260 atomic_long_inc(&c->dirty_zn_cnt);
261 atomic_long_dec(&c->clean_zn_cnt);
262 atomic_long_dec(&ubifs_clean_zn_cnt);
263 err = add_idx_dirt(c, zbr->lnum, zbr->len);
270 zn = copy_znode(c, znode);
275 err = insert_old_idx(c, zbr->lnum, zbr->offs);
278 err = add_idx_dirt(c, zbr->lnum, zbr->len);
293 * lnc_add - add a leaf node to the leaf node cache.
294 * @c: UBIFS file-system description object
295 * @zbr: zbranch of leaf node
298 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
299 * purpose of the leaf node cache is to save re-reading the same leaf node over
300 * and over again. Most things are cached by VFS, however the file system must
301 * cache directory entries for readdir and for resolving hash collisions. The
302 * present implementation of the leaf node cache is extremely simple, and
303 * allows for error returns that are not used but that may be needed if a more
304 * complex implementation is created.
306 * Note, this function does not add the @node object to LNC directly, but
307 * allocates a copy of the object and adds the copy to LNC. The reason for this
308 * is that @node has been allocated outside of the TNC subsystem and will be
309 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
310 * may be changed at any time, e.g. freed by the shrinker.
312 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
317 const struct ubifs_dent_node *dent = node;
319 ubifs_assert(!zbr->leaf);
320 ubifs_assert(zbr->len != 0);
321 ubifs_assert(is_hash_key(c, &zbr->key));
323 err = ubifs_validate_entry(c, dent);
326 ubifs_dump_node(c, dent);
330 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
332 /* We don't have to have the cache, so no error */
335 zbr->leaf = lnc_node;
340 * lnc_add_directly - add a leaf node to the leaf-node-cache.
341 * @c: UBIFS file-system description object
342 * @zbr: zbranch of leaf node
345 * This function is similar to 'lnc_add()', but it does not create a copy of
346 * @node but inserts @node to TNC directly.
348 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
353 ubifs_assert(!zbr->leaf);
354 ubifs_assert(zbr->len != 0);
356 err = ubifs_validate_entry(c, node);
359 ubifs_dump_node(c, node);
368 * lnc_free - remove a leaf node from the leaf node cache.
369 * @zbr: zbranch of leaf node
372 static void lnc_free(struct ubifs_zbranch *zbr)
381 * tnc_read_node_nm - read a "hashed" leaf node.
382 * @c: UBIFS file-system description object
383 * @zbr: key and position of the node
384 * @node: node is returned here
386 * This function reads a "hashed" node defined by @zbr from the leaf node cache
387 * (in it is there) or from the hash media, in which case the node is also
388 * added to LNC. Returns zero in case of success or a negative negative error
389 * code in case of failure.
391 static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
396 ubifs_assert(is_hash_key(c, &zbr->key));
399 /* Read from the leaf node cache */
400 ubifs_assert(zbr->len != 0);
401 memcpy(node, zbr->leaf, zbr->len);
405 err = ubifs_tnc_read_node(c, zbr, node);
409 /* Add the node to the leaf node cache */
410 err = lnc_add(c, zbr, node);
415 * try_read_node - read a node if it is a node.
416 * @c: UBIFS file-system description object
417 * @buf: buffer to read to
419 * @len: node length (not aligned)
420 * @lnum: LEB number of node to read
421 * @offs: offset of node to read
423 * This function tries to read a node of known type and length, checks it and
424 * stores it in @buf. This function returns %1 if a node is present and %0 if
425 * a node is not present. A negative error code is returned for I/O errors.
426 * This function performs that same function as ubifs_read_node except that
427 * it does not require that there is actually a node present and instead
428 * the return code indicates if a node was read.
430 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
431 * is true (it is controlled by corresponding mount option). However, if
432 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
433 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
434 * because during mounting or re-mounting from R/O mode to R/W mode we may read
435 * journal nodes (when replying the journal or doing the recovery) and the
436 * journal nodes may potentially be corrupted, so checking is required.
438 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
439 int len, int lnum, int offs)
442 struct ubifs_ch *ch = buf;
443 uint32_t crc, node_crc;
445 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
447 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
449 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
450 type, lnum, offs, err);
454 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
457 if (ch->node_type != type)
460 node_len = le32_to_cpu(ch->len);
464 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
468 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
469 node_crc = le32_to_cpu(ch->crc);
477 * fallible_read_node - try to read a leaf node.
478 * @c: UBIFS file-system description object
479 * @key: key of node to read
480 * @zbr: position of node
481 * @node: node returned
483 * This function tries to read a node and returns %1 if the node is read, %0
484 * if the node is not present, and a negative error code in the case of error.
486 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
487 struct ubifs_zbranch *zbr, void *node)
491 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
493 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
496 union ubifs_key node_key;
497 struct ubifs_dent_node *dent = node;
499 /* All nodes have key in the same place */
500 key_read(c, &dent->key, &node_key);
501 if (keys_cmp(c, key, &node_key) != 0)
504 if (ret == 0 && c->replaying)
505 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
506 zbr->lnum, zbr->offs, zbr->len);
511 * matches_name - determine if a direntry or xattr entry matches a given name.
512 * @c: UBIFS file-system description object
513 * @zbr: zbranch of dent
516 * This function checks if xentry/direntry referred by zbranch @zbr matches name
517 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
518 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
519 * of failure, a negative error code is returned.
521 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
522 const struct qstr *nm)
524 struct ubifs_dent_node *dent;
527 /* If possible, match against the dent in the leaf node cache */
529 dent = kmalloc(zbr->len, GFP_NOFS);
533 err = ubifs_tnc_read_node(c, zbr, dent);
537 /* Add the node to the leaf node cache */
538 err = lnc_add_directly(c, zbr, dent);
544 nlen = le16_to_cpu(dent->nlen);
545 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
549 else if (nlen < nm->len)
564 * get_znode - get a TNC znode that may not be loaded yet.
565 * @c: UBIFS file-system description object
566 * @znode: parent znode
567 * @n: znode branch slot number
569 * This function returns the znode or a negative error code.
571 static struct ubifs_znode *get_znode(struct ubifs_info *c,
572 struct ubifs_znode *znode, int n)
574 struct ubifs_zbranch *zbr;
576 zbr = &znode->zbranch[n];
580 znode = ubifs_load_znode(c, zbr, znode, n);
585 * tnc_next - find next TNC entry.
586 * @c: UBIFS file-system description object
587 * @zn: znode is passed and returned here
588 * @n: znode branch slot number is passed and returned here
590 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
591 * no next entry, or a negative error code otherwise.
593 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
595 struct ubifs_znode *znode = *zn;
599 if (nn < znode->child_cnt) {
604 struct ubifs_znode *zp;
611 if (nn < znode->child_cnt) {
612 znode = get_znode(c, znode, nn);
614 return PTR_ERR(znode);
615 while (znode->level != 0) {
616 znode = get_znode(c, znode, 0);
618 return PTR_ERR(znode);
630 * tnc_prev - find previous TNC entry.
631 * @c: UBIFS file-system description object
632 * @zn: znode is returned here
633 * @n: znode branch slot number is passed and returned here
635 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
636 * there is no next entry, or a negative error code otherwise.
638 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
640 struct ubifs_znode *znode = *zn;
648 struct ubifs_znode *zp;
656 znode = get_znode(c, znode, nn);
658 return PTR_ERR(znode);
659 while (znode->level != 0) {
660 nn = znode->child_cnt - 1;
661 znode = get_znode(c, znode, nn);
663 return PTR_ERR(znode);
665 nn = znode->child_cnt - 1;
675 * resolve_collision - resolve a collision.
676 * @c: UBIFS file-system description object
677 * @key: key of a directory or extended attribute entry
678 * @zn: znode is returned here
679 * @n: zbranch number is passed and returned here
680 * @nm: name of the entry
682 * This function is called for "hashed" keys to make sure that the found key
683 * really corresponds to the looked up node (directory or extended attribute
684 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
685 * %0 is returned if @nm is not found and @zn and @n are set to the previous
686 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
687 * This means that @n may be set to %-1 if the leftmost key in @zn is the
688 * previous one. A negative error code is returned on failures.
690 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
691 struct ubifs_znode **zn, int *n,
692 const struct qstr *nm)
696 err = matches_name(c, &(*zn)->zbranch[*n], nm);
697 if (unlikely(err < 0))
699 if (err == NAME_MATCHES)
702 if (err == NAME_GREATER) {
705 err = tnc_prev(c, zn, n);
706 if (err == -ENOENT) {
707 ubifs_assert(*n == 0);
713 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
715 * We have found the branch after which we would
716 * like to insert, but inserting in this znode
717 * may still be wrong. Consider the following 3
718 * znodes, in the case where we are resolving a
719 * collision with Key2.
722 * ----------------------
723 * level 1 | Key0 | Key1 |
724 * -----------------------
726 * znode za | | znode zb
727 * ------------ ------------
728 * level 0 | Key0 | | Key2 |
729 * ------------ ------------
731 * The lookup finds Key2 in znode zb. Lets say
732 * there is no match and the name is greater so
733 * we look left. When we find Key0, we end up
734 * here. If we return now, we will insert into
735 * znode za at slot n = 1. But that is invalid
736 * according to the parent's keys. Key2 must
737 * be inserted into znode zb.
739 * Note, this problem is not relevant for the
740 * case when we go right, because
741 * 'tnc_insert()' would correct the parent key.
743 if (*n == (*zn)->child_cnt - 1) {
744 err = tnc_next(c, zn, n);
746 /* Should be impossible */
752 ubifs_assert(*n == 0);
757 err = matches_name(c, &(*zn)->zbranch[*n], nm);
760 if (err == NAME_LESS)
762 if (err == NAME_MATCHES)
764 ubifs_assert(err == NAME_GREATER);
768 struct ubifs_znode *znode = *zn;
772 err = tnc_next(c, &znode, &nn);
777 if (keys_cmp(c, &znode->zbranch[nn].key, key))
779 err = matches_name(c, &znode->zbranch[nn], nm);
782 if (err == NAME_GREATER)
786 if (err == NAME_MATCHES)
788 ubifs_assert(err == NAME_LESS);
794 * fallible_matches_name - determine if a dent matches a given name.
795 * @c: UBIFS file-system description object
796 * @zbr: zbranch of dent
799 * This is a "fallible" version of 'matches_name()' function which does not
800 * panic if the direntry/xentry referred by @zbr does not exist on the media.
802 * This function checks if xentry/direntry referred by zbranch @zbr matches name
803 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
804 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
805 * if xentry/direntry referred by @zbr does not exist on the media. A negative
806 * error code is returned in case of failure.
808 static int fallible_matches_name(struct ubifs_info *c,
809 struct ubifs_zbranch *zbr,
810 const struct qstr *nm)
812 struct ubifs_dent_node *dent;
815 /* If possible, match against the dent in the leaf node cache */
817 dent = kmalloc(zbr->len, GFP_NOFS);
821 err = fallible_read_node(c, &zbr->key, zbr, dent);
825 /* The node was not present */
829 ubifs_assert(err == 1);
831 err = lnc_add_directly(c, zbr, dent);
837 nlen = le16_to_cpu(dent->nlen);
838 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
842 else if (nlen < nm->len)
857 * fallible_resolve_collision - resolve a collision even if nodes are missing.
858 * @c: UBIFS file-system description object
860 * @zn: znode is returned here
861 * @n: branch number is passed and returned here
862 * @nm: name of directory entry
863 * @adding: indicates caller is adding a key to the TNC
865 * This is a "fallible" version of the 'resolve_collision()' function which
866 * does not panic if one of the nodes referred to by TNC does not exist on the
867 * media. This may happen when replaying the journal if a deleted node was
868 * Garbage-collected and the commit was not done. A branch that refers to a node
869 * that is not present is called a dangling branch. The following are the return
870 * codes for this function:
871 * o if @nm was found, %1 is returned and @zn and @n are set to the found
873 * o if we are @adding and @nm was not found, %0 is returned;
874 * o if we are not @adding and @nm was not found, but a dangling branch was
875 * found, then %1 is returned and @zn and @n are set to the dangling branch;
876 * o a negative error code is returned in case of failure.
878 static int fallible_resolve_collision(struct ubifs_info *c,
879 const union ubifs_key *key,
880 struct ubifs_znode **zn, int *n,
881 const struct qstr *nm, int adding)
883 struct ubifs_znode *o_znode = NULL, *znode = *zn;
884 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
886 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
887 if (unlikely(cmp < 0))
889 if (cmp == NAME_MATCHES)
891 if (cmp == NOT_ON_MEDIA) {
895 * We are unlucky and hit a dangling branch straight away.
896 * Now we do not really know where to go to find the needed
897 * branch - to the left or to the right. Well, let's try left.
901 unsure = 1; /* Remove a dangling branch wherever it is */
903 if (cmp == NAME_GREATER || unsure) {
906 err = tnc_prev(c, zn, n);
907 if (err == -ENOENT) {
908 ubifs_assert(*n == 0);
914 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
915 /* See comments in 'resolve_collision()' */
916 if (*n == (*zn)->child_cnt - 1) {
917 err = tnc_next(c, zn, n);
919 /* Should be impossible */
925 ubifs_assert(*n == 0);
930 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
933 if (err == NAME_MATCHES)
935 if (err == NOT_ON_MEDIA) {
942 if (err == NAME_LESS)
949 if (cmp == NAME_LESS || unsure) {
954 err = tnc_next(c, &znode, &nn);
959 if (keys_cmp(c, &znode->zbranch[nn].key, key))
961 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
964 if (err == NAME_GREATER)
968 if (err == NAME_MATCHES)
970 if (err == NOT_ON_MEDIA) {
977 /* Never match a dangling branch when adding */
978 if (adding || !o_znode)
981 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
982 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
983 o_znode->zbranch[o_n].len);
990 * matches_position - determine if a zbranch matches a given position.
991 * @zbr: zbranch of dent
992 * @lnum: LEB number of dent to match
993 * @offs: offset of dent to match
995 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
997 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
999 if (zbr->lnum == lnum && zbr->offs == offs)
1006 * resolve_collision_directly - resolve a collision directly.
1007 * @c: UBIFS file-system description object
1008 * @key: key of directory entry
1009 * @zn: znode is passed and returned here
1010 * @n: zbranch number is passed and returned here
1011 * @lnum: LEB number of dent node to match
1012 * @offs: offset of dent node to match
1014 * This function is used for "hashed" keys to make sure the found directory or
1015 * extended attribute entry node is what was looked for. It is used when the
1016 * flash address of the right node is known (@lnum:@offs) which makes it much
1017 * easier to resolve collisions (no need to read entries and match full
1018 * names). This function returns %1 and sets @zn and @n if the collision is
1019 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1020 * previous directory entry. Otherwise a negative error code is returned.
1022 static int resolve_collision_directly(struct ubifs_info *c,
1023 const union ubifs_key *key,
1024 struct ubifs_znode **zn, int *n,
1027 struct ubifs_znode *znode;
1032 if (matches_position(&znode->zbranch[nn], lnum, offs))
1037 err = tnc_prev(c, &znode, &nn);
1042 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1044 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1055 err = tnc_next(c, &znode, &nn);
1060 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1064 if (matches_position(&znode->zbranch[nn], lnum, offs))
1070 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1071 * @c: UBIFS file-system description object
1072 * @znode: znode to dirty
1074 * If we do not have a unique key that resides in a znode, then we cannot
1075 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1076 * This function records the path back to the last dirty ancestor, and then
1077 * dirties the znodes on that path.
1079 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1080 struct ubifs_znode *znode)
1082 struct ubifs_znode *zp;
1083 int *path = c->bottom_up_buf, p = 0;
1085 ubifs_assert(c->zroot.znode);
1086 ubifs_assert(znode);
1087 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1088 kfree(c->bottom_up_buf);
1089 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1091 if (!c->bottom_up_buf)
1092 return ERR_PTR(-ENOMEM);
1093 path = c->bottom_up_buf;
1095 if (c->zroot.znode->level) {
1096 /* Go up until parent is dirty */
1104 ubifs_assert(p < c->zroot.znode->level);
1106 if (!zp->cnext && ubifs_zn_dirty(znode))
1112 /* Come back down, dirtying as we go */
1114 struct ubifs_zbranch *zbr;
1118 ubifs_assert(path[p - 1] >= 0);
1119 ubifs_assert(path[p - 1] < zp->child_cnt);
1120 zbr = &zp->zbranch[path[--p]];
1121 znode = dirty_cow_znode(c, zbr);
1123 ubifs_assert(znode == c->zroot.znode);
1124 znode = dirty_cow_znode(c, &c->zroot);
1126 if (IS_ERR(znode) || !p)
1128 ubifs_assert(path[p - 1] >= 0);
1129 ubifs_assert(path[p - 1] < znode->child_cnt);
1130 znode = znode->zbranch[path[p - 1]].znode;
1137 * ubifs_lookup_level0 - search for zero-level znode.
1138 * @c: UBIFS file-system description object
1139 * @key: key to lookup
1140 * @zn: znode is returned here
1141 * @n: znode branch slot number is returned here
1143 * This function looks up the TNC tree and search for zero-level znode which
1144 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1146 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1147 * is returned and slot number of the matched branch is stored in @n;
1148 * o not exact match, which means that zero-level znode does not contain
1149 * @key, then %0 is returned and slot number of the closest branch is stored
1151 * o @key is so small that it is even less than the lowest key of the
1152 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1154 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1155 * function reads corresponding indexing nodes and inserts them to TNC. In
1156 * case of failure, a negative error code is returned.
1158 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1159 struct ubifs_znode **zn, int *n)
1162 struct ubifs_znode *znode;
1163 unsigned long time = get_seconds();
1165 dbg_tnck(key, "search key ");
1166 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1168 znode = c->zroot.znode;
1169 if (unlikely(!znode)) {
1170 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1172 return PTR_ERR(znode);
1178 struct ubifs_zbranch *zbr;
1180 exact = ubifs_search_zbranch(c, znode, key, n);
1182 if (znode->level == 0)
1187 zbr = &znode->zbranch[*n];
1195 /* znode is not in TNC cache, load it from the media */
1196 znode = ubifs_load_znode(c, zbr, znode, *n);
1198 return PTR_ERR(znode);
1202 if (exact || !is_hash_key(c, key) || *n != -1) {
1203 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1208 * Here is a tricky place. We have not found the key and this is a
1209 * "hashed" key, which may collide. The rest of the code deals with
1210 * situations like this:
1214 * | 3 | 5 | | 6 | 7 | (x)
1216 * Or more a complex example:
1220 * | 1 | 3 | | 5 | 8 |
1222 * | 5 | 5 | | 6 | 7 | (x)
1224 * In the examples, if we are looking for key "5", we may reach nodes
1225 * marked with "(x)". In this case what we have do is to look at the
1226 * left and see if there is "5" key there. If there is, we have to
1229 * Note, this whole situation is possible because we allow to have
1230 * elements which are equivalent to the next key in the parent in the
1231 * children of current znode. For example, this happens if we split a
1232 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1236 * | 3 | 5 | | 5 | 6 | 7 |
1238 * And this becomes what is at the first "picture" after key "5" marked
1239 * with "^" is removed. What could be done is we could prohibit
1240 * splitting in the middle of the colliding sequence. Also, when
1241 * removing the leftmost key, we would have to correct the key of the
1242 * parent node, which would introduce additional complications. Namely,
1243 * if we changed the leftmost key of the parent znode, the garbage
1244 * collector would be unable to find it (GC is doing this when GC'ing
1245 * indexing LEBs). Although we already have an additional RB-tree where
1246 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1247 * after the commit. But anyway, this does not look easy to implement
1248 * so we did not try this.
1250 err = tnc_prev(c, &znode, n);
1251 if (err == -ENOENT) {
1252 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1256 if (unlikely(err < 0))
1258 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1259 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1264 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1270 * lookup_level0_dirty - search for zero-level znode dirtying.
1271 * @c: UBIFS file-system description object
1272 * @key: key to lookup
1273 * @zn: znode is returned here
1274 * @n: znode branch slot number is returned here
1276 * This function looks up the TNC tree and search for zero-level znode which
1277 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1279 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1280 * is returned and slot number of the matched branch is stored in @n;
1281 * o not exact match, which means that zero-level znode does not contain @key
1282 * then %0 is returned and slot number of the closed branch is stored in
1284 * o @key is so small that it is even less than the lowest key of the
1285 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1287 * Additionally all znodes in the path from the root to the located zero-level
1288 * znode are marked as dirty.
1290 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1291 * function reads corresponding indexing nodes and inserts them to TNC. In
1292 * case of failure, a negative error code is returned.
1294 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1295 struct ubifs_znode **zn, int *n)
1298 struct ubifs_znode *znode;
1299 unsigned long time = get_seconds();
1301 dbg_tnck(key, "search and dirty key ");
1303 znode = c->zroot.znode;
1304 if (unlikely(!znode)) {
1305 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1307 return PTR_ERR(znode);
1310 znode = dirty_cow_znode(c, &c->zroot);
1312 return PTR_ERR(znode);
1317 struct ubifs_zbranch *zbr;
1319 exact = ubifs_search_zbranch(c, znode, key, n);
1321 if (znode->level == 0)
1326 zbr = &znode->zbranch[*n];
1330 znode = dirty_cow_znode(c, zbr);
1332 return PTR_ERR(znode);
1336 /* znode is not in TNC cache, load it from the media */
1337 znode = ubifs_load_znode(c, zbr, znode, *n);
1339 return PTR_ERR(znode);
1340 znode = dirty_cow_znode(c, zbr);
1342 return PTR_ERR(znode);
1346 if (exact || !is_hash_key(c, key) || *n != -1) {
1347 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1352 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1355 err = tnc_prev(c, &znode, n);
1356 if (err == -ENOENT) {
1358 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1361 if (unlikely(err < 0))
1363 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1365 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1369 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1370 znode = dirty_cow_bottom_up(c, znode);
1372 return PTR_ERR(znode);
1375 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1381 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1382 * @c: UBIFS file-system description object
1384 * @gc_seq1: garbage collection sequence number
1386 * This function determines if @lnum may have been garbage collected since
1387 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1390 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1393 int gc_seq2, gced_lnum;
1395 gced_lnum = c->gced_lnum;
1397 gc_seq2 = c->gc_seq;
1398 /* Same seq means no GC */
1399 if (gc_seq1 == gc_seq2)
1401 /* Different by more than 1 means we don't know */
1402 if (gc_seq1 + 1 != gc_seq2)
1405 * We have seen the sequence number has increased by 1. Now we need to
1406 * be sure we read the right LEB number, so read it again.
1409 if (gced_lnum != c->gced_lnum)
1411 /* Finally we can check lnum */
1412 if (gced_lnum == lnum)
1415 /* No garbage collection in the read-only U-Boot implementation */
1421 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1422 * @c: UBIFS file-system description object
1423 * @key: node key to lookup
1424 * @node: the node is returned here
1425 * @lnum: LEB number is returned here
1426 * @offs: offset is returned here
1428 * This function looks up and reads node with key @key. The caller has to make
1429 * sure the @node buffer is large enough to fit the node. Returns zero in case
1430 * of success, %-ENOENT if the node was not found, and a negative error code in
1431 * case of failure. The node location can be returned in @lnum and @offs.
1433 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1434 void *node, int *lnum, int *offs)
1436 int found, n, err, safely = 0, gc_seq1;
1437 struct ubifs_znode *znode;
1438 struct ubifs_zbranch zbr, *zt;
1441 mutex_lock(&c->tnc_mutex);
1442 found = ubifs_lookup_level0(c, key, &znode, &n);
1446 } else if (found < 0) {
1450 zt = &znode->zbranch[n];
1455 if (is_hash_key(c, key)) {
1457 * In this case the leaf node cache gets used, so we pass the
1458 * address of the zbranch and keep the mutex locked
1460 err = tnc_read_node_nm(c, zt, node);
1464 err = ubifs_tnc_read_node(c, zt, node);
1467 /* Drop the TNC mutex prematurely and race with garbage collection */
1468 zbr = znode->zbranch[n];
1469 gc_seq1 = c->gc_seq;
1470 mutex_unlock(&c->tnc_mutex);
1472 if (ubifs_get_wbuf(c, zbr.lnum)) {
1473 /* We do not GC journal heads */
1474 err = ubifs_tnc_read_node(c, &zbr, node);
1478 err = fallible_read_node(c, key, &zbr, node);
1479 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1481 * The node may have been GC'ed out from under us so try again
1482 * while keeping the TNC mutex locked.
1490 mutex_unlock(&c->tnc_mutex);
1495 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1496 * @c: UBIFS file-system description object
1497 * @bu: bulk-read parameters and results
1499 * Lookup consecutive data node keys for the same inode that reside
1500 * consecutively in the same LEB. This function returns zero in case of success
1501 * and a negative error code in case of failure.
1503 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1504 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1505 * maximum possible amount of nodes for bulk-read.
1507 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1509 int n, err = 0, lnum = -1, uninitialized_var(offs);
1510 int uninitialized_var(len);
1511 unsigned int block = key_block(c, &bu->key);
1512 struct ubifs_znode *znode;
1518 mutex_lock(&c->tnc_mutex);
1519 /* Find first key */
1520 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1525 len = znode->zbranch[n].len;
1526 /* The buffer must be big enough for at least 1 node */
1527 if (len > bu->buf_len) {
1532 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1534 lnum = znode->zbranch[n].lnum;
1535 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1538 struct ubifs_zbranch *zbr;
1539 union ubifs_key *key;
1540 unsigned int next_block;
1543 err = tnc_next(c, &znode, &n);
1546 zbr = &znode->zbranch[n];
1548 /* See if there is another data key for this file */
1549 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1550 key_type(c, key) != UBIFS_DATA_KEY) {
1555 /* First key found */
1557 offs = ALIGN(zbr->offs + zbr->len, 8);
1559 if (len > bu->buf_len) {
1565 * The data nodes must be in consecutive positions in
1568 if (zbr->lnum != lnum || zbr->offs != offs)
1570 offs += ALIGN(zbr->len, 8);
1571 len = ALIGN(len, 8) + zbr->len;
1572 /* Must not exceed buffer length */
1573 if (len > bu->buf_len)
1576 /* Allow for holes */
1577 next_block = key_block(c, key);
1578 bu->blk_cnt += (next_block - block - 1);
1579 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1583 bu->zbranch[bu->cnt++] = *zbr;
1585 /* See if we have room for more */
1586 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1588 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1592 if (err == -ENOENT) {
1596 bu->gc_seq = c->gc_seq;
1597 mutex_unlock(&c->tnc_mutex);
1601 * An enormous hole could cause bulk-read to encompass too many
1602 * page cache pages, so limit the number here.
1604 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1605 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1607 * Ensure that bulk-read covers a whole number of page cache
1610 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1611 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1614 /* At the end of file we can round up */
1615 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1618 /* Exclude data nodes that do not make up a whole page cache page */
1619 block = key_block(c, &bu->key) + bu->blk_cnt;
1620 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1622 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1630 * read_wbuf - bulk-read from a LEB with a wbuf.
1631 * @wbuf: wbuf that may overlap the read
1632 * @buf: buffer into which to read
1634 * @lnum: LEB number from which to read
1635 * @offs: offset from which to read
1637 * This functions returns %0 on success or a negative error code on failure.
1639 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1642 const struct ubifs_info *c = wbuf->c;
1645 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1646 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1647 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1648 ubifs_assert(offs + len <= c->leb_size);
1650 spin_lock(&wbuf->lock);
1651 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1653 /* We may safely unlock the write-buffer and read the data */
1654 spin_unlock(&wbuf->lock);
1655 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1658 /* Don't read under wbuf */
1659 rlen = wbuf->offs - offs;
1663 /* Copy the rest from the write-buffer */
1664 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1665 spin_unlock(&wbuf->lock);
1668 /* Read everything that goes before write-buffer */
1669 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1675 * validate_data_node - validate data nodes for bulk-read.
1676 * @c: UBIFS file-system description object
1677 * @buf: buffer containing data node to validate
1678 * @zbr: zbranch of data node to validate
1680 * This functions returns %0 on success or a negative error code on failure.
1682 static int validate_data_node(struct ubifs_info *c, void *buf,
1683 struct ubifs_zbranch *zbr)
1685 union ubifs_key key1;
1686 struct ubifs_ch *ch = buf;
1689 if (ch->node_type != UBIFS_DATA_NODE) {
1690 ubifs_err(c, "bad node type (%d but expected %d)",
1691 ch->node_type, UBIFS_DATA_NODE);
1695 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1697 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1701 len = le32_to_cpu(ch->len);
1702 if (len != zbr->len) {
1703 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1707 /* Make sure the key of the read node is correct */
1708 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1709 if (!keys_eq(c, &zbr->key, &key1)) {
1710 ubifs_err(c, "bad key in node at LEB %d:%d",
1711 zbr->lnum, zbr->offs);
1712 dbg_tnck(&zbr->key, "looked for key ");
1713 dbg_tnck(&key1, "found node's key ");
1722 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1723 ubifs_dump_node(c, buf);
1729 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1730 * @c: UBIFS file-system description object
1731 * @bu: bulk-read parameters and results
1733 * This functions reads and validates the data nodes that were identified by the
1734 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1735 * -EAGAIN to indicate a race with GC, or another negative error code on
1738 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1740 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1741 struct ubifs_wbuf *wbuf;
1744 len = bu->zbranch[bu->cnt - 1].offs;
1745 len += bu->zbranch[bu->cnt - 1].len - offs;
1746 if (len > bu->buf_len) {
1747 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1752 wbuf = ubifs_get_wbuf(c, lnum);
1754 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1756 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1758 /* Check for a race with GC */
1759 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1762 if (err && err != -EBADMSG) {
1763 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1766 dbg_tnck(&bu->key, "key ");
1770 /* Validate the nodes read */
1772 for (i = 0; i < bu->cnt; i++) {
1773 err = validate_data_node(c, buf, &bu->zbranch[i]);
1776 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1783 * do_lookup_nm- look up a "hashed" node.
1784 * @c: UBIFS file-system description object
1785 * @key: node key to lookup
1786 * @node: the node is returned here
1789 * This function look up and reads a node which contains name hash in the key.
1790 * Since the hash may have collisions, there may be many nodes with the same
1791 * key, so we have to sequentially look to all of them until the needed one is
1792 * found. This function returns zero in case of success, %-ENOENT if the node
1793 * was not found, and a negative error code in case of failure.
1795 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1796 void *node, const struct qstr *nm)
1799 struct ubifs_znode *znode;
1801 dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name);
1802 mutex_lock(&c->tnc_mutex);
1803 found = ubifs_lookup_level0(c, key, &znode, &n);
1807 } else if (found < 0) {
1812 ubifs_assert(n >= 0);
1814 err = resolve_collision(c, key, &znode, &n, nm);
1815 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1816 if (unlikely(err < 0))
1823 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1826 mutex_unlock(&c->tnc_mutex);
1831 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1832 * @c: UBIFS file-system description object
1833 * @key: node key to lookup
1834 * @node: the node is returned here
1837 * This function look up and reads a node which contains name hash in the key.
1838 * Since the hash may have collisions, there may be many nodes with the same
1839 * key, so we have to sequentially look to all of them until the needed one is
1840 * found. This function returns zero in case of success, %-ENOENT if the node
1841 * was not found, and a negative error code in case of failure.
1843 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1844 void *node, const struct qstr *nm)
1847 const struct ubifs_dent_node *dent = node;
1850 * We assume that in most of the cases there are no name collisions and
1851 * 'ubifs_tnc_lookup()' returns us the right direntry.
1853 err = ubifs_tnc_lookup(c, key, node);
1857 len = le16_to_cpu(dent->nlen);
1858 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1862 * Unluckily, there are hash collisions and we have to iterate over
1863 * them look at each direntry with colliding name hash sequentially.
1865 return do_lookup_nm(c, key, node, nm);
1869 * correct_parent_keys - correct parent znodes' keys.
1870 * @c: UBIFS file-system description object
1871 * @znode: znode to correct parent znodes for
1873 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1874 * zbranch changes, keys of parent znodes have to be corrected. This helper
1875 * function is called in such situations and corrects the keys if needed.
1877 static void correct_parent_keys(const struct ubifs_info *c,
1878 struct ubifs_znode *znode)
1880 union ubifs_key *key, *key1;
1882 ubifs_assert(znode->parent);
1883 ubifs_assert(znode->iip == 0);
1885 key = &znode->zbranch[0].key;
1886 key1 = &znode->parent->zbranch[0].key;
1888 while (keys_cmp(c, key, key1) < 0) {
1889 key_copy(c, key, key1);
1890 znode = znode->parent;
1892 if (!znode->parent || znode->iip)
1894 key1 = &znode->parent->zbranch[0].key;
1899 * insert_zbranch - insert a zbranch into a znode.
1900 * @znode: znode into which to insert
1901 * @zbr: zbranch to insert
1902 * @n: slot number to insert to
1904 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1905 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1906 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1907 * slot, zbranches starting from @n have to be moved right.
1909 static void insert_zbranch(struct ubifs_znode *znode,
1910 const struct ubifs_zbranch *zbr, int n)
1914 ubifs_assert(ubifs_zn_dirty(znode));
1917 for (i = znode->child_cnt; i > n; i--) {
1918 znode->zbranch[i] = znode->zbranch[i - 1];
1919 if (znode->zbranch[i].znode)
1920 znode->zbranch[i].znode->iip = i;
1923 zbr->znode->iip = n;
1925 for (i = znode->child_cnt; i > n; i--)
1926 znode->zbranch[i] = znode->zbranch[i - 1];
1928 znode->zbranch[n] = *zbr;
1929 znode->child_cnt += 1;
1932 * After inserting at slot zero, the lower bound of the key range of
1933 * this znode may have changed. If this znode is subsequently split
1934 * then the upper bound of the key range may change, and furthermore
1935 * it could change to be lower than the original lower bound. If that
1936 * happens, then it will no longer be possible to find this znode in the
1937 * TNC using the key from the index node on flash. That is bad because
1938 * if it is not found, we will assume it is obsolete and may overwrite
1939 * it. Then if there is an unclean unmount, we will start using the
1940 * old index which will be broken.
1942 * So we first mark znodes that have insertions at slot zero, and then
1943 * if they are split we add their lnum/offs to the old_idx tree.
1950 * tnc_insert - insert a node into TNC.
1951 * @c: UBIFS file-system description object
1952 * @znode: znode to insert into
1953 * @zbr: branch to insert
1954 * @n: slot number to insert new zbranch to
1956 * This function inserts a new node described by @zbr into znode @znode. If
1957 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1958 * are splat as well if needed. Returns zero in case of success or a negative
1959 * error code in case of failure.
1961 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1962 struct ubifs_zbranch *zbr, int n)
1964 struct ubifs_znode *zn, *zi, *zp;
1965 int i, keep, move, appending = 0;
1966 union ubifs_key *key = &zbr->key, *key1;
1968 ubifs_assert(n >= 0 && n <= c->fanout);
1970 /* Implement naive insert for now */
1973 if (znode->child_cnt < c->fanout) {
1974 ubifs_assert(n != c->fanout);
1975 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
1977 insert_zbranch(znode, zbr, n);
1979 /* Ensure parent's key is correct */
1980 if (n == 0 && zp && znode->iip == 0)
1981 correct_parent_keys(c, znode);
1987 * Unfortunately, @znode does not have more empty slots and we have to
1990 dbg_tnck(key, "splitting level %d, key ", znode->level);
1994 * We can no longer be sure of finding this znode by key, so we
1995 * record it in the old_idx tree.
1997 ins_clr_old_idx_znode(c, znode);
1999 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2003 zn->level = znode->level;
2005 /* Decide where to split */
2006 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2007 /* Try not to split consecutive data keys */
2008 if (n == c->fanout) {
2009 key1 = &znode->zbranch[n - 1].key;
2010 if (key_inum(c, key1) == key_inum(c, key) &&
2011 key_type(c, key1) == UBIFS_DATA_KEY)
2015 } else if (appending && n != c->fanout) {
2016 /* Try not to split consecutive data keys */
2019 if (n >= (c->fanout + 1) / 2) {
2020 key1 = &znode->zbranch[0].key;
2021 if (key_inum(c, key1) == key_inum(c, key) &&
2022 key_type(c, key1) == UBIFS_DATA_KEY) {
2023 key1 = &znode->zbranch[n].key;
2024 if (key_inum(c, key1) != key_inum(c, key) ||
2025 key_type(c, key1) != UBIFS_DATA_KEY) {
2027 move = c->fanout - keep;
2039 keep = (c->fanout + 1) / 2;
2040 move = c->fanout - keep;
2044 * Although we don't at present, we could look at the neighbors and see
2045 * if we can move some zbranches there.
2049 /* Insert into existing znode */
2054 /* Insert into new znode */
2059 zbr->znode->parent = zn;
2064 __set_bit(DIRTY_ZNODE, &zn->flags);
2065 atomic_long_inc(&c->dirty_zn_cnt);
2067 zn->child_cnt = move;
2068 znode->child_cnt = keep;
2070 dbg_tnc("moving %d, keeping %d", move, keep);
2073 for (i = 0; i < move; i++) {
2074 zn->zbranch[i] = znode->zbranch[keep + i];
2077 if (zn->zbranch[i].znode) {
2078 zn->zbranch[i].znode->parent = zn;
2079 zn->zbranch[i].znode->iip = i;
2083 /* Insert new key and branch */
2084 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2086 insert_zbranch(zi, zbr, n);
2088 /* Insert new znode (produced by spitting) into the parent */
2090 if (n == 0 && zi == znode && znode->iip == 0)
2091 correct_parent_keys(c, znode);
2093 /* Locate insertion point */
2096 /* Tail recursion */
2097 zbr->key = zn->zbranch[0].key;
2107 /* We have to split root znode */
2108 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2110 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2115 zi->level = znode->level + 1;
2117 __set_bit(DIRTY_ZNODE, &zi->flags);
2118 atomic_long_inc(&c->dirty_zn_cnt);
2120 zi->zbranch[0].key = znode->zbranch[0].key;
2121 zi->zbranch[0].znode = znode;
2122 zi->zbranch[0].lnum = c->zroot.lnum;
2123 zi->zbranch[0].offs = c->zroot.offs;
2124 zi->zbranch[0].len = c->zroot.len;
2125 zi->zbranch[1].key = zn->zbranch[0].key;
2126 zi->zbranch[1].znode = zn;
2131 c->zroot.znode = zi;
2142 * ubifs_tnc_add - add a node to TNC.
2143 * @c: UBIFS file-system description object
2145 * @lnum: LEB number of node
2146 * @offs: node offset
2149 * This function adds a node with key @key to TNC. The node may be new or it may
2150 * obsolete some existing one. Returns %0 on success or negative error code on
2153 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2156 int found, n, err = 0;
2157 struct ubifs_znode *znode;
2159 mutex_lock(&c->tnc_mutex);
2160 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2161 found = lookup_level0_dirty(c, key, &znode, &n);
2163 struct ubifs_zbranch zbr;
2169 key_copy(c, key, &zbr.key);
2170 err = tnc_insert(c, znode, &zbr, n + 1);
2171 } else if (found == 1) {
2172 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2175 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2182 err = dbg_check_tnc(c, 0);
2183 mutex_unlock(&c->tnc_mutex);
2189 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2190 * @c: UBIFS file-system description object
2192 * @old_lnum: LEB number of old node
2193 * @old_offs: old node offset
2194 * @lnum: LEB number of node
2195 * @offs: node offset
2198 * This function replaces a node with key @key in the TNC only if the old node
2199 * is found. This function is called by garbage collection when node are moved.
2200 * Returns %0 on success or negative error code on failure.
2202 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2203 int old_lnum, int old_offs, int lnum, int offs, int len)
2205 int found, n, err = 0;
2206 struct ubifs_znode *znode;
2208 mutex_lock(&c->tnc_mutex);
2209 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2210 old_offs, lnum, offs, len);
2211 found = lookup_level0_dirty(c, key, &znode, &n);
2218 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2221 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2223 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2230 } else if (is_hash_key(c, key)) {
2231 found = resolve_collision_directly(c, key, &znode, &n,
2232 old_lnum, old_offs);
2233 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2234 found, znode, n, old_lnum, old_offs);
2241 /* Ensure the znode is dirtied */
2242 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2243 znode = dirty_cow_bottom_up(c, znode);
2244 if (IS_ERR(znode)) {
2245 err = PTR_ERR(znode);
2249 zbr = &znode->zbranch[n];
2251 err = ubifs_add_dirt(c, zbr->lnum,
2263 err = ubifs_add_dirt(c, lnum, len);
2266 err = dbg_check_tnc(c, 0);
2269 mutex_unlock(&c->tnc_mutex);
2274 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2275 * @c: UBIFS file-system description object
2277 * @lnum: LEB number of node
2278 * @offs: node offset
2282 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2283 * may have collisions, like directory entry keys.
2285 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2286 int lnum, int offs, int len, const struct qstr *nm)
2288 int found, n, err = 0;
2289 struct ubifs_znode *znode;
2291 mutex_lock(&c->tnc_mutex);
2292 dbg_tnck(key, "LEB %d:%d, name '%.*s', key ",
2293 lnum, offs, nm->len, nm->name);
2294 found = lookup_level0_dirty(c, key, &znode, &n);
2302 found = fallible_resolve_collision(c, key, &znode, &n,
2305 found = resolve_collision(c, key, &znode, &n, nm);
2306 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2312 /* Ensure the znode is dirtied */
2313 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2314 znode = dirty_cow_bottom_up(c, znode);
2315 if (IS_ERR(znode)) {
2316 err = PTR_ERR(znode);
2322 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2325 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2334 struct ubifs_zbranch zbr;
2340 key_copy(c, key, &zbr.key);
2341 err = tnc_insert(c, znode, &zbr, n + 1);
2346 * We did not find it in the index so there may be a
2347 * dangling branch still in the index. So we remove it
2348 * by passing 'ubifs_tnc_remove_nm()' the same key but
2349 * an unmatchable name.
2351 struct qstr noname = { .name = "" };
2353 err = dbg_check_tnc(c, 0);
2354 mutex_unlock(&c->tnc_mutex);
2357 return ubifs_tnc_remove_nm(c, key, &noname);
2363 err = dbg_check_tnc(c, 0);
2364 mutex_unlock(&c->tnc_mutex);
2369 * tnc_delete - delete a znode form TNC.
2370 * @c: UBIFS file-system description object
2371 * @znode: znode to delete from
2372 * @n: zbranch slot number to delete
2374 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2375 * case of success and a negative error code in case of failure.
2377 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2379 struct ubifs_zbranch *zbr;
2380 struct ubifs_znode *zp;
2383 /* Delete without merge for now */
2384 ubifs_assert(znode->level == 0);
2385 ubifs_assert(n >= 0 && n < c->fanout);
2386 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2388 zbr = &znode->zbranch[n];
2391 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2393 ubifs_dump_znode(c, znode);
2397 /* We do not "gap" zbranch slots */
2398 for (i = n; i < znode->child_cnt - 1; i++)
2399 znode->zbranch[i] = znode->zbranch[i + 1];
2400 znode->child_cnt -= 1;
2402 if (znode->child_cnt > 0)
2406 * This was the last zbranch, we have to delete this znode from the
2411 ubifs_assert(!ubifs_zn_obsolete(znode));
2412 ubifs_assert(ubifs_zn_dirty(znode));
2417 atomic_long_dec(&c->dirty_zn_cnt);
2419 err = insert_old_idx_znode(c, znode);
2424 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2425 atomic_long_inc(&c->clean_zn_cnt);
2426 atomic_long_inc(&ubifs_clean_zn_cnt);
2430 } while (znode->child_cnt == 1); /* while removing last child */
2432 /* Remove from znode, entry n - 1 */
2433 znode->child_cnt -= 1;
2434 ubifs_assert(znode->level != 0);
2435 for (i = n; i < znode->child_cnt; i++) {
2436 znode->zbranch[i] = znode->zbranch[i + 1];
2437 if (znode->zbranch[i].znode)
2438 znode->zbranch[i].znode->iip = i;
2442 * If this is the root and it has only 1 child then
2443 * collapse the tree.
2445 if (!znode->parent) {
2446 while (znode->child_cnt == 1 && znode->level != 0) {
2448 zbr = &znode->zbranch[0];
2449 znode = get_znode(c, znode, 0);
2451 return PTR_ERR(znode);
2452 znode = dirty_cow_znode(c, zbr);
2454 return PTR_ERR(znode);
2455 znode->parent = NULL;
2458 err = insert_old_idx(c, c->zroot.lnum,
2463 c->zroot.lnum = zbr->lnum;
2464 c->zroot.offs = zbr->offs;
2465 c->zroot.len = zbr->len;
2466 c->zroot.znode = znode;
2467 ubifs_assert(!ubifs_zn_obsolete(zp));
2468 ubifs_assert(ubifs_zn_dirty(zp));
2469 atomic_long_dec(&c->dirty_zn_cnt);
2472 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2473 atomic_long_inc(&c->clean_zn_cnt);
2474 atomic_long_inc(&ubifs_clean_zn_cnt);
2484 * ubifs_tnc_remove - remove an index entry of a node.
2485 * @c: UBIFS file-system description object
2488 * Returns %0 on success or negative error code on failure.
2490 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2492 int found, n, err = 0;
2493 struct ubifs_znode *znode;
2495 mutex_lock(&c->tnc_mutex);
2496 dbg_tnck(key, "key ");
2497 found = lookup_level0_dirty(c, key, &znode, &n);
2503 err = tnc_delete(c, znode, n);
2505 err = dbg_check_tnc(c, 0);
2508 mutex_unlock(&c->tnc_mutex);
2513 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2514 * @c: UBIFS file-system description object
2516 * @nm: directory entry name
2518 * Returns %0 on success or negative error code on failure.
2520 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2521 const struct qstr *nm)
2524 struct ubifs_znode *znode;
2526 mutex_lock(&c->tnc_mutex);
2527 dbg_tnck(key, "%.*s, key ", nm->len, nm->name);
2528 err = lookup_level0_dirty(c, key, &znode, &n);
2534 err = fallible_resolve_collision(c, key, &znode, &n,
2537 err = resolve_collision(c, key, &znode, &n, nm);
2538 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2542 /* Ensure the znode is dirtied */
2543 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2544 znode = dirty_cow_bottom_up(c, znode);
2545 if (IS_ERR(znode)) {
2546 err = PTR_ERR(znode);
2550 err = tnc_delete(c, znode, n);
2556 err = dbg_check_tnc(c, 0);
2557 mutex_unlock(&c->tnc_mutex);
2562 * key_in_range - determine if a key falls within a range of keys.
2563 * @c: UBIFS file-system description object
2564 * @key: key to check
2565 * @from_key: lowest key in range
2566 * @to_key: highest key in range
2568 * This function returns %1 if the key is in range and %0 otherwise.
2570 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2571 union ubifs_key *from_key, union ubifs_key *to_key)
2573 if (keys_cmp(c, key, from_key) < 0)
2575 if (keys_cmp(c, key, to_key) > 0)
2581 * ubifs_tnc_remove_range - remove index entries in range.
2582 * @c: UBIFS file-system description object
2583 * @from_key: lowest key to remove
2584 * @to_key: highest key to remove
2586 * This function removes index entries starting at @from_key and ending at
2587 * @to_key. This function returns zero in case of success and a negative error
2588 * code in case of failure.
2590 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2591 union ubifs_key *to_key)
2593 int i, n, k, err = 0;
2594 struct ubifs_znode *znode;
2595 union ubifs_key *key;
2597 mutex_lock(&c->tnc_mutex);
2599 /* Find first level 0 znode that contains keys to remove */
2600 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2607 err = tnc_next(c, &znode, &n);
2608 if (err == -ENOENT) {
2614 key = &znode->zbranch[n].key;
2615 if (!key_in_range(c, key, from_key, to_key)) {
2621 /* Ensure the znode is dirtied */
2622 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2623 znode = dirty_cow_bottom_up(c, znode);
2624 if (IS_ERR(znode)) {
2625 err = PTR_ERR(znode);
2630 /* Remove all keys in range except the first */
2631 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2632 key = &znode->zbranch[i].key;
2633 if (!key_in_range(c, key, from_key, to_key))
2635 lnc_free(&znode->zbranch[i]);
2636 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2637 znode->zbranch[i].len);
2639 ubifs_dump_znode(c, znode);
2642 dbg_tnck(key, "removing key ");
2645 for (i = n + 1 + k; i < znode->child_cnt; i++)
2646 znode->zbranch[i - k] = znode->zbranch[i];
2647 znode->child_cnt -= k;
2650 /* Now delete the first */
2651 err = tnc_delete(c, znode, n);
2658 err = dbg_check_tnc(c, 0);
2659 mutex_unlock(&c->tnc_mutex);
2664 * ubifs_tnc_remove_ino - remove an inode from TNC.
2665 * @c: UBIFS file-system description object
2666 * @inum: inode number to remove
2668 * This function remove inode @inum and all the extended attributes associated
2669 * with the anode from TNC and returns zero in case of success or a negative
2670 * error code in case of failure.
2672 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2674 union ubifs_key key1, key2;
2675 struct ubifs_dent_node *xent, *pxent = NULL;
2676 struct qstr nm = { .name = NULL };
2678 dbg_tnc("ino %lu", (unsigned long)inum);
2681 * Walk all extended attribute entries and remove them together with
2682 * corresponding extended attribute inodes.
2684 lowest_xent_key(c, &key1, inum);
2689 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2691 err = PTR_ERR(xent);
2697 xattr_inum = le64_to_cpu(xent->inum);
2698 dbg_tnc("xent '%s', ino %lu", xent->name,
2699 (unsigned long)xattr_inum);
2701 nm.name = xent->name;
2702 nm.len = le16_to_cpu(xent->nlen);
2703 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2709 lowest_ino_key(c, &key1, xattr_inum);
2710 highest_ino_key(c, &key2, xattr_inum);
2711 err = ubifs_tnc_remove_range(c, &key1, &key2);
2719 key_read(c, &xent->key, &key1);
2723 lowest_ino_key(c, &key1, inum);
2724 highest_ino_key(c, &key2, inum);
2726 return ubifs_tnc_remove_range(c, &key1, &key2);
2730 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2731 * @c: UBIFS file-system description object
2732 * @key: key of last entry
2733 * @nm: name of last entry found or %NULL
2735 * This function finds and reads the next directory or extended attribute entry
2736 * after the given key (@key) if there is one. @nm is used to resolve
2739 * If the name of the current entry is not known and only the key is known,
2740 * @nm->name has to be %NULL. In this case the semantics of this function is a
2741 * little bit different and it returns the entry corresponding to this key, not
2742 * the next one. If the key was not found, the closest "right" entry is
2745 * If the fist entry has to be found, @key has to contain the lowest possible
2746 * key value for this inode and @name has to be %NULL.
2748 * This function returns the found directory or extended attribute entry node
2749 * in case of success, %-ENOENT is returned if no entry was found, and a
2750 * negative error code is returned in case of failure.
2752 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2753 union ubifs_key *key,
2754 const struct qstr *nm)
2756 int n, err, type = key_type(c, key);
2757 struct ubifs_znode *znode;
2758 struct ubifs_dent_node *dent;
2759 struct ubifs_zbranch *zbr;
2760 union ubifs_key *dkey;
2762 dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)");
2763 ubifs_assert(is_hash_key(c, key));
2765 mutex_lock(&c->tnc_mutex);
2766 err = ubifs_lookup_level0(c, key, &znode, &n);
2767 if (unlikely(err < 0))
2772 /* Handle collisions */
2773 err = resolve_collision(c, key, &znode, &n, nm);
2774 dbg_tnc("rc returned %d, znode %p, n %d",
2776 if (unlikely(err < 0))
2780 /* Now find next entry */
2781 err = tnc_next(c, &znode, &n);
2786 * The full name of the entry was not given, in which case the
2787 * behavior of this function is a little different and it
2788 * returns current entry, not the next one.
2792 * However, the given key does not exist in the TNC
2793 * tree and @znode/@n variables contain the closest
2794 * "preceding" element. Switch to the next one.
2796 err = tnc_next(c, &znode, &n);
2802 zbr = &znode->zbranch[n];
2803 dent = kmalloc(zbr->len, GFP_NOFS);
2804 if (unlikely(!dent)) {
2810 * The above 'tnc_next()' call could lead us to the next inode, check
2814 if (key_inum(c, dkey) != key_inum(c, key) ||
2815 key_type(c, dkey) != type) {
2820 err = tnc_read_node_nm(c, zbr, dent);
2824 mutex_unlock(&c->tnc_mutex);
2830 mutex_unlock(&c->tnc_mutex);
2831 return ERR_PTR(err);
2835 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2836 * @c: UBIFS file-system description object
2838 * Destroy left-over obsolete znodes from a failed commit.
2840 static void tnc_destroy_cnext(struct ubifs_info *c)
2842 struct ubifs_znode *cnext;
2846 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2849 struct ubifs_znode *znode = cnext;
2851 cnext = cnext->cnext;
2852 if (ubifs_zn_obsolete(znode))
2854 } while (cnext && cnext != c->cnext);
2858 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2859 * @c: UBIFS file-system description object
2861 void ubifs_tnc_close(struct ubifs_info *c)
2863 tnc_destroy_cnext(c);
2864 if (c->zroot.znode) {
2867 n = atomic_long_read(&c->clean_zn_cnt);
2868 freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
2869 ubifs_assert(freed == n);
2870 atomic_long_sub(n, &ubifs_clean_zn_cnt);
2878 * left_znode - get the znode to the left.
2879 * @c: UBIFS file-system description object
2882 * This function returns a pointer to the znode to the left of @znode or NULL if
2883 * there is not one. A negative error code is returned on failure.
2885 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2886 struct ubifs_znode *znode)
2888 int level = znode->level;
2891 int n = znode->iip - 1;
2893 /* Go up until we can go left */
2894 znode = znode->parent;
2898 /* Now go down the rightmost branch to 'level' */
2899 znode = get_znode(c, znode, n);
2902 while (znode->level != level) {
2903 n = znode->child_cnt - 1;
2904 znode = get_znode(c, znode, n);
2915 * right_znode - get the znode to the right.
2916 * @c: UBIFS file-system description object
2919 * This function returns a pointer to the znode to the right of @znode or NULL
2920 * if there is not one. A negative error code is returned on failure.
2922 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2923 struct ubifs_znode *znode)
2925 int level = znode->level;
2928 int n = znode->iip + 1;
2930 /* Go up until we can go right */
2931 znode = znode->parent;
2934 if (n < znode->child_cnt) {
2935 /* Now go down the leftmost branch to 'level' */
2936 znode = get_znode(c, znode, n);
2939 while (znode->level != level) {
2940 znode = get_znode(c, znode, 0);
2951 * lookup_znode - find a particular indexing node from TNC.
2952 * @c: UBIFS file-system description object
2953 * @key: index node key to lookup
2954 * @level: index node level
2955 * @lnum: index node LEB number
2956 * @offs: index node offset
2958 * This function searches an indexing node by its first key @key and its
2959 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2960 * nodes it traverses to TNC. This function is called for indexing nodes which
2961 * were found on the media by scanning, for example when garbage-collecting or
2962 * when doing in-the-gaps commit. This means that the indexing node which is
2963 * looked for does not have to have exactly the same leftmost key @key, because
2964 * the leftmost key may have been changed, in which case TNC will contain a
2965 * dirty znode which still refers the same @lnum:@offs. This function is clever
2966 * enough to recognize such indexing nodes.
2968 * Note, if a znode was deleted or changed too much, then this function will
2969 * not find it. For situations like this UBIFS has the old index RB-tree
2970 * (indexed by @lnum:@offs).
2972 * This function returns a pointer to the znode found or %NULL if it is not
2973 * found. A negative error code is returned on failure.
2975 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2976 union ubifs_key *key, int level,
2979 struct ubifs_znode *znode, *zn;
2982 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
2985 * The arguments have probably been read off flash, so don't assume
2989 return ERR_PTR(-EINVAL);
2991 /* Get the root znode */
2992 znode = c->zroot.znode;
2994 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
2998 /* Check if it is the one we are looking for */
2999 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3001 /* Descend to the parent level i.e. (level + 1) */
3002 if (level >= znode->level)
3005 ubifs_search_zbranch(c, znode, key, &n);
3008 * We reached a znode where the leftmost key is greater
3009 * than the key we are searching for. This is the same
3010 * situation as the one described in a huge comment at
3011 * the end of the 'ubifs_lookup_level0()' function. And
3012 * for exactly the same reasons we have to try to look
3013 * left before giving up.
3015 znode = left_znode(c, znode);
3020 ubifs_search_zbranch(c, znode, key, &n);
3021 ubifs_assert(n >= 0);
3023 if (znode->level == level + 1)
3025 znode = get_znode(c, znode, n);
3029 /* Check if the child is the one we are looking for */
3030 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3031 return get_znode(c, znode, n);
3032 /* If the key is unique, there is nowhere else to look */
3033 if (!is_hash_key(c, key))
3036 * The key is not unique and so may be also in the znodes to either
3043 /* Move one branch to the left */
3047 znode = left_znode(c, znode);
3052 n = znode->child_cnt - 1;
3055 if (znode->zbranch[n].lnum == lnum &&
3056 znode->zbranch[n].offs == offs)
3057 return get_znode(c, znode, n);
3058 /* Stop if the key is less than the one we are looking for */
3059 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3062 /* Back to the middle */
3067 /* Move one branch to the right */
3068 if (++n >= znode->child_cnt) {
3069 znode = right_znode(c, znode);
3077 if (znode->zbranch[n].lnum == lnum &&
3078 znode->zbranch[n].offs == offs)
3079 return get_znode(c, znode, n);
3080 /* Stop if the key is greater than the one we are looking for */
3081 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3088 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3089 * @c: UBIFS file-system description object
3090 * @key: key of index node
3091 * @level: index node level
3092 * @lnum: LEB number of index node
3093 * @offs: offset of index node
3095 * This function returns %0 if the index node is not referred to in the TNC, %1
3096 * if the index node is referred to in the TNC and the corresponding znode is
3097 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3098 * znode is clean, and a negative error code in case of failure.
3100 * Note, the @key argument has to be the key of the first child. Also note,
3101 * this function relies on the fact that 0:0 is never a valid LEB number and
3102 * offset for a main-area node.
3104 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3107 struct ubifs_znode *znode;
3109 znode = lookup_znode(c, key, level, lnum, offs);
3113 return PTR_ERR(znode);
3115 return ubifs_zn_dirty(znode) ? 1 : 2;
3119 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3120 * @c: UBIFS file-system description object
3122 * @lnum: node LEB number
3123 * @offs: node offset
3125 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3126 * not, and a negative error code in case of failure.
3128 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3129 * and offset for a main-area node.
3131 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3134 struct ubifs_zbranch *zbr;
3135 struct ubifs_znode *znode, *zn;
3136 int n, found, err, nn;
3137 const int unique = !is_hash_key(c, key);
3139 found = ubifs_lookup_level0(c, key, &znode, &n);
3141 return found; /* Error code */
3144 zbr = &znode->zbranch[n];
3145 if (lnum == zbr->lnum && offs == zbr->offs)
3146 return 1; /* Found it */
3150 * Because the key is not unique, we have to look left
3157 err = tnc_prev(c, &znode, &n);
3162 if (keys_cmp(c, key, &znode->zbranch[n].key))
3164 zbr = &znode->zbranch[n];
3165 if (lnum == zbr->lnum && offs == zbr->offs)
3166 return 1; /* Found it */
3172 err = tnc_next(c, &znode, &n);
3178 if (keys_cmp(c, key, &znode->zbranch[n].key))
3180 zbr = &znode->zbranch[n];
3181 if (lnum == zbr->lnum && offs == zbr->offs)
3182 return 1; /* Found it */
3188 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3189 * @c: UBIFS file-system description object
3191 * @level: index node level (if it is an index node)
3192 * @lnum: node LEB number
3193 * @offs: node offset
3194 * @is_idx: non-zero if the node is an index node
3196 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3197 * negative error code in case of failure. For index nodes, @key has to be the
3198 * key of the first child. An index node is considered to be in the TNC only if
3199 * the corresponding znode is clean or has not been loaded.
3201 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3202 int lnum, int offs, int is_idx)
3206 mutex_lock(&c->tnc_mutex);
3208 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3212 /* The index node was found but it was dirty */
3215 /* The index node was found and it was clean */
3220 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3223 mutex_unlock(&c->tnc_mutex);
3228 * ubifs_dirty_idx_node - dirty an index node.
3229 * @c: UBIFS file-system description object
3230 * @key: index node key
3231 * @level: index node level
3232 * @lnum: index node LEB number
3233 * @offs: index node offset
3235 * This function loads and dirties an index node so that it can be garbage
3236 * collected. The @key argument has to be the key of the first child. This
3237 * function relies on the fact that 0:0 is never a valid LEB number and offset
3238 * for a main-area node. Returns %0 on success and a negative error code on
3241 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3244 struct ubifs_znode *znode;
3247 mutex_lock(&c->tnc_mutex);
3248 znode = lookup_znode(c, key, level, lnum, offs);
3251 if (IS_ERR(znode)) {
3252 err = PTR_ERR(znode);
3255 znode = dirty_cow_bottom_up(c, znode);
3256 if (IS_ERR(znode)) {
3257 err = PTR_ERR(znode);
3262 mutex_unlock(&c->tnc_mutex);
3267 * dbg_check_inode_size - check if inode size is correct.
3268 * @c: UBIFS file-system description object
3269 * @inum: inode number
3272 * This function makes sure that the inode size (@size) is correct and it does
3273 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3274 * if it has a data page beyond @size, and other negative error code in case of
3277 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3281 union ubifs_key from_key, to_key, *key;
3282 struct ubifs_znode *znode;
3285 if (!S_ISREG(inode->i_mode))
3287 if (!dbg_is_chk_gen(c))
3290 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3291 data_key_init(c, &from_key, inode->i_ino, block);
3292 highest_data_key(c, &to_key, inode->i_ino);
3294 mutex_lock(&c->tnc_mutex);
3295 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3304 err = tnc_next(c, &znode, &n);
3305 if (err == -ENOENT) {
3312 ubifs_assert(err == 0);
3313 key = &znode->zbranch[n].key;
3314 if (!key_in_range(c, key, &from_key, &to_key))
3318 block = key_block(c, key);
3319 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3320 (unsigned long)inode->i_ino, size,
3321 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3322 mutex_unlock(&c->tnc_mutex);
3323 ubifs_dump_inode(c, inode);
3328 mutex_unlock(&c->tnc_mutex);