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.
53 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
54 int len, int lnum, int offs);
55 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
56 struct ubifs_zbranch *zbr, void *node);
59 * insert_old_idx - record an index node obsoleted since the last commit start.
60 * @c: UBIFS file-system description object
61 * @lnum: LEB number of obsoleted index node
62 * @offs: offset of obsoleted index node
64 * Returns %0 on success, and a negative error code on failure.
66 * For recovery, there must always be a complete intact version of the index on
67 * flash at all times. That is called the "old index". It is the index as at the
68 * time of the last successful commit. Many of the index nodes in the old index
69 * may be dirty, but they must not be erased until the next successful commit
70 * (at which point that index becomes the old index).
72 * That means that the garbage collection and the in-the-gaps method of
73 * committing must be able to determine if an index node is in the old index.
74 * Most of the old index nodes can be found by looking up the TNC using the
75 * 'lookup_znode()' function. However, some of the old index nodes may have
76 * been deleted from the current index or may have been changed so much that
77 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
78 * That is what this function does. The RB-tree is ordered by LEB number and
79 * offset because they uniquely identify the old index node.
81 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
83 struct ubifs_old_idx *old_idx, *o;
84 struct rb_node **p, *parent = NULL;
86 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
87 if (unlikely(!old_idx))
92 p = &c->old_idx.rb_node;
95 o = rb_entry(parent, struct ubifs_old_idx, rb);
98 else if (lnum > o->lnum)
100 else if (offs < o->offs)
102 else if (offs > o->offs)
105 ubifs_err(c, "old idx added twice!");
110 rb_link_node(&old_idx->rb, parent, p);
111 rb_insert_color(&old_idx->rb, &c->old_idx);
116 * insert_old_idx_znode - record a znode obsoleted since last commit start.
117 * @c: UBIFS file-system description object
118 * @znode: znode of obsoleted index node
120 * Returns %0 on success, and a negative error code on failure.
122 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
125 struct ubifs_zbranch *zbr;
127 zbr = &znode->parent->zbranch[znode->iip];
129 return insert_old_idx(c, zbr->lnum, zbr->offs);
132 return insert_old_idx(c, c->zroot.lnum,
138 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
139 * @c: UBIFS file-system description object
140 * @znode: znode of obsoleted index node
142 * Returns %0 on success, and a negative error code on failure.
144 static int ins_clr_old_idx_znode(struct ubifs_info *c,
145 struct ubifs_znode *znode)
150 struct ubifs_zbranch *zbr;
152 zbr = &znode->parent->zbranch[znode->iip];
154 err = insert_old_idx(c, zbr->lnum, zbr->offs);
163 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
174 * destroy_old_idx - destroy the old_idx RB-tree.
175 * @c: UBIFS file-system description object
177 * During start commit, the old_idx RB-tree is used to avoid overwriting index
178 * nodes that were in the index last commit but have since been deleted. This
179 * is necessary for recovery i.e. the old index must be kept intact until the
180 * new index is successfully written. The old-idx RB-tree is used for the
181 * in-the-gaps method of writing index nodes and is destroyed every commit.
183 void destroy_old_idx(struct ubifs_info *c)
185 struct ubifs_old_idx *old_idx, *n;
187 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
190 c->old_idx = RB_ROOT;
194 * copy_znode - copy a dirty znode.
195 * @c: UBIFS file-system description object
196 * @znode: znode to copy
198 * A dirty znode being committed may not be changed, so it is copied.
200 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
201 struct ubifs_znode *znode)
203 struct ubifs_znode *zn;
205 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
207 return ERR_PTR(-ENOMEM);
209 memcpy(zn, znode, c->max_znode_sz);
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_node_nm - 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_node_nm(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 qstr *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, nm->name, min_t(int, nlen, nm->len));
566 else if (nlen < nm->len)
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 qstr *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 qstr *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, nm->name, min_t(int, nlen, nm->len));
859 else if (nlen < nm->len)
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 qstr *nm, int adding)
900 struct ubifs_znode *o_znode = NULL, *znode = *zn;
901 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
903 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
904 if (unlikely(cmp < 0))
906 if (cmp == NAME_MATCHES)
908 if (cmp == NOT_ON_MEDIA) {
912 * We are unlucky and hit a dangling branch straight away.
913 * Now we do not really know where to go to find the needed
914 * branch - to the left or to the right. Well, let's try left.
918 unsure = 1; /* Remove a dangling branch wherever it is */
920 if (cmp == NAME_GREATER || unsure) {
923 err = tnc_prev(c, zn, n);
924 if (err == -ENOENT) {
925 ubifs_assert(*n == 0);
931 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
932 /* See comments in 'resolve_collision()' */
933 if (*n == (*zn)->child_cnt - 1) {
934 err = tnc_next(c, zn, n);
936 /* Should be impossible */
942 ubifs_assert(*n == 0);
947 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
950 if (err == NAME_MATCHES)
952 if (err == NOT_ON_MEDIA) {
959 if (err == NAME_LESS)
966 if (cmp == NAME_LESS || unsure) {
971 err = tnc_next(c, &znode, &nn);
976 if (keys_cmp(c, &znode->zbranch[nn].key, key))
978 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
981 if (err == NAME_GREATER)
985 if (err == NAME_MATCHES)
987 if (err == NOT_ON_MEDIA) {
994 /* Never match a dangling branch when adding */
995 if (adding || !o_znode)
998 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
999 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1000 o_znode->zbranch[o_n].len);
1007 * matches_position - determine if a zbranch matches a given position.
1008 * @zbr: zbranch of dent
1009 * @lnum: LEB number of dent to match
1010 * @offs: offset of dent to match
1012 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1014 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1016 if (zbr->lnum == lnum && zbr->offs == offs)
1023 * resolve_collision_directly - resolve a collision directly.
1024 * @c: UBIFS file-system description object
1025 * @key: key of directory entry
1026 * @zn: znode is passed and returned here
1027 * @n: zbranch number is passed and returned here
1028 * @lnum: LEB number of dent node to match
1029 * @offs: offset of dent node to match
1031 * This function is used for "hashed" keys to make sure the found directory or
1032 * extended attribute entry node is what was looked for. It is used when the
1033 * flash address of the right node is known (@lnum:@offs) which makes it much
1034 * easier to resolve collisions (no need to read entries and match full
1035 * names). This function returns %1 and sets @zn and @n if the collision is
1036 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1037 * previous directory entry. Otherwise a negative error code is returned.
1039 static int resolve_collision_directly(struct ubifs_info *c,
1040 const union ubifs_key *key,
1041 struct ubifs_znode **zn, int *n,
1044 struct ubifs_znode *znode;
1049 if (matches_position(&znode->zbranch[nn], lnum, offs))
1054 err = tnc_prev(c, &znode, &nn);
1059 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1061 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1072 err = tnc_next(c, &znode, &nn);
1077 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1081 if (matches_position(&znode->zbranch[nn], lnum, offs))
1087 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1088 * @c: UBIFS file-system description object
1089 * @znode: znode to dirty
1091 * If we do not have a unique key that resides in a znode, then we cannot
1092 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1093 * This function records the path back to the last dirty ancestor, and then
1094 * dirties the znodes on that path.
1096 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1097 struct ubifs_znode *znode)
1099 struct ubifs_znode *zp;
1100 int *path = c->bottom_up_buf, p = 0;
1102 ubifs_assert(c->zroot.znode);
1103 ubifs_assert(znode);
1104 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1105 kfree(c->bottom_up_buf);
1106 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1108 if (!c->bottom_up_buf)
1109 return ERR_PTR(-ENOMEM);
1110 path = c->bottom_up_buf;
1112 if (c->zroot.znode->level) {
1113 /* Go up until parent is dirty */
1121 ubifs_assert(p < c->zroot.znode->level);
1123 if (!zp->cnext && ubifs_zn_dirty(znode))
1129 /* Come back down, dirtying as we go */
1131 struct ubifs_zbranch *zbr;
1135 ubifs_assert(path[p - 1] >= 0);
1136 ubifs_assert(path[p - 1] < zp->child_cnt);
1137 zbr = &zp->zbranch[path[--p]];
1138 znode = dirty_cow_znode(c, zbr);
1140 ubifs_assert(znode == c->zroot.znode);
1141 znode = dirty_cow_znode(c, &c->zroot);
1143 if (IS_ERR(znode) || !p)
1145 ubifs_assert(path[p - 1] >= 0);
1146 ubifs_assert(path[p - 1] < znode->child_cnt);
1147 znode = znode->zbranch[path[p - 1]].znode;
1154 * ubifs_lookup_level0 - search for zero-level znode.
1155 * @c: UBIFS file-system description object
1156 * @key: key to lookup
1157 * @zn: znode is returned here
1158 * @n: znode branch slot number is returned here
1160 * This function looks up the TNC tree and search for zero-level znode which
1161 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1163 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1164 * is returned and slot number of the matched branch is stored in @n;
1165 * o not exact match, which means that zero-level znode does not contain
1166 * @key, then %0 is returned and slot number of the closest branch is stored
1168 * o @key is so small that it is even less than the lowest key of the
1169 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1171 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1172 * function reads corresponding indexing nodes and inserts them to TNC. In
1173 * case of failure, a negative error code is returned.
1175 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1176 struct ubifs_znode **zn, int *n)
1179 struct ubifs_znode *znode;
1180 unsigned long time = get_seconds();
1182 dbg_tnck(key, "search key ");
1183 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1185 znode = c->zroot.znode;
1186 if (unlikely(!znode)) {
1187 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1189 return PTR_ERR(znode);
1195 struct ubifs_zbranch *zbr;
1197 exact = ubifs_search_zbranch(c, znode, key, n);
1199 if (znode->level == 0)
1204 zbr = &znode->zbranch[*n];
1212 /* znode is not in TNC cache, load it from the media */
1213 znode = ubifs_load_znode(c, zbr, znode, *n);
1215 return PTR_ERR(znode);
1219 if (exact || !is_hash_key(c, key) || *n != -1) {
1220 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1225 * Here is a tricky place. We have not found the key and this is a
1226 * "hashed" key, which may collide. The rest of the code deals with
1227 * situations like this:
1231 * | 3 | 5 | | 6 | 7 | (x)
1233 * Or more a complex example:
1237 * | 1 | 3 | | 5 | 8 |
1239 * | 5 | 5 | | 6 | 7 | (x)
1241 * In the examples, if we are looking for key "5", we may reach nodes
1242 * marked with "(x)". In this case what we have do is to look at the
1243 * left and see if there is "5" key there. If there is, we have to
1246 * Note, this whole situation is possible because we allow to have
1247 * elements which are equivalent to the next key in the parent in the
1248 * children of current znode. For example, this happens if we split a
1249 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1253 * | 3 | 5 | | 5 | 6 | 7 |
1255 * And this becomes what is at the first "picture" after key "5" marked
1256 * with "^" is removed. What could be done is we could prohibit
1257 * splitting in the middle of the colliding sequence. Also, when
1258 * removing the leftmost key, we would have to correct the key of the
1259 * parent node, which would introduce additional complications. Namely,
1260 * if we changed the leftmost key of the parent znode, the garbage
1261 * collector would be unable to find it (GC is doing this when GC'ing
1262 * indexing LEBs). Although we already have an additional RB-tree where
1263 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1264 * after the commit. But anyway, this does not look easy to implement
1265 * so we did not try this.
1267 err = tnc_prev(c, &znode, n);
1268 if (err == -ENOENT) {
1269 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1273 if (unlikely(err < 0))
1275 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1276 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1281 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1287 * lookup_level0_dirty - search for zero-level znode dirtying.
1288 * @c: UBIFS file-system description object
1289 * @key: key to lookup
1290 * @zn: znode is returned here
1291 * @n: znode branch slot number is returned here
1293 * This function looks up the TNC tree and search for zero-level znode which
1294 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1296 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1297 * is returned and slot number of the matched branch is stored in @n;
1298 * o not exact match, which means that zero-level znode does not contain @key
1299 * then %0 is returned and slot number of the closed branch is stored in
1301 * o @key is so small that it is even less than the lowest key of the
1302 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1304 * Additionally all znodes in the path from the root to the located zero-level
1305 * znode are marked as dirty.
1307 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1308 * function reads corresponding indexing nodes and inserts them to TNC. In
1309 * case of failure, a negative error code is returned.
1311 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1312 struct ubifs_znode **zn, int *n)
1315 struct ubifs_znode *znode;
1316 unsigned long time = get_seconds();
1318 dbg_tnck(key, "search and dirty key ");
1320 znode = c->zroot.znode;
1321 if (unlikely(!znode)) {
1322 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1324 return PTR_ERR(znode);
1327 znode = dirty_cow_znode(c, &c->zroot);
1329 return PTR_ERR(znode);
1334 struct ubifs_zbranch *zbr;
1336 exact = ubifs_search_zbranch(c, znode, key, n);
1338 if (znode->level == 0)
1343 zbr = &znode->zbranch[*n];
1347 znode = dirty_cow_znode(c, zbr);
1349 return PTR_ERR(znode);
1353 /* znode is not in TNC cache, load it from the media */
1354 znode = ubifs_load_znode(c, zbr, znode, *n);
1356 return PTR_ERR(znode);
1357 znode = dirty_cow_znode(c, zbr);
1359 return PTR_ERR(znode);
1363 if (exact || !is_hash_key(c, key) || *n != -1) {
1364 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1369 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1372 err = tnc_prev(c, &znode, n);
1373 if (err == -ENOENT) {
1375 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1378 if (unlikely(err < 0))
1380 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1382 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1386 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1387 znode = dirty_cow_bottom_up(c, znode);
1389 return PTR_ERR(znode);
1392 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1398 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1399 * @c: UBIFS file-system description object
1401 * @gc_seq1: garbage collection sequence number
1403 * This function determines if @lnum may have been garbage collected since
1404 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1407 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)
1432 /* No garbage collection in the read-only U-Boot implementation */
1438 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1439 * @c: UBIFS file-system description object
1440 * @key: node key to lookup
1441 * @node: the node is returned here
1442 * @lnum: LEB number is returned here
1443 * @offs: offset is returned here
1445 * This function looks up and reads node with key @key. The caller has to make
1446 * sure the @node buffer is large enough to fit the node. Returns zero in case
1447 * of success, %-ENOENT if the node was not found, and a negative error code in
1448 * case of failure. The node location can be returned in @lnum and @offs.
1450 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1451 void *node, int *lnum, int *offs)
1453 int found, n, err, safely = 0, gc_seq1;
1454 struct ubifs_znode *znode;
1455 struct ubifs_zbranch zbr, *zt;
1458 mutex_lock(&c->tnc_mutex);
1459 found = ubifs_lookup_level0(c, key, &znode, &n);
1463 } else if (found < 0) {
1467 zt = &znode->zbranch[n];
1472 if (is_hash_key(c, key)) {
1474 * In this case the leaf node cache gets used, so we pass the
1475 * address of the zbranch and keep the mutex locked
1477 err = tnc_read_node_nm(c, zt, node);
1481 err = ubifs_tnc_read_node(c, zt, node);
1484 /* Drop the TNC mutex prematurely and race with garbage collection */
1485 zbr = znode->zbranch[n];
1486 gc_seq1 = c->gc_seq;
1487 mutex_unlock(&c->tnc_mutex);
1489 if (ubifs_get_wbuf(c, zbr.lnum)) {
1490 /* We do not GC journal heads */
1491 err = ubifs_tnc_read_node(c, &zbr, node);
1495 err = fallible_read_node(c, key, &zbr, node);
1496 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1498 * The node may have been GC'ed out from under us so try again
1499 * while keeping the TNC mutex locked.
1507 mutex_unlock(&c->tnc_mutex);
1512 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1513 * @c: UBIFS file-system description object
1514 * @bu: bulk-read parameters and results
1516 * Lookup consecutive data node keys for the same inode that reside
1517 * consecutively in the same LEB. This function returns zero in case of success
1518 * and a negative error code in case of failure.
1520 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1521 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1522 * maximum possible amount of nodes for bulk-read.
1524 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1526 int n, err = 0, lnum = -1, uninitialized_var(offs);
1527 int uninitialized_var(len);
1528 unsigned int block = key_block(c, &bu->key);
1529 struct ubifs_znode *znode;
1535 mutex_lock(&c->tnc_mutex);
1536 /* Find first key */
1537 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1542 len = znode->zbranch[n].len;
1543 /* The buffer must be big enough for at least 1 node */
1544 if (len > bu->buf_len) {
1549 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1551 lnum = znode->zbranch[n].lnum;
1552 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1555 struct ubifs_zbranch *zbr;
1556 union ubifs_key *key;
1557 unsigned int next_block;
1560 err = tnc_next(c, &znode, &n);
1563 zbr = &znode->zbranch[n];
1565 /* See if there is another data key for this file */
1566 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1567 key_type(c, key) != UBIFS_DATA_KEY) {
1572 /* First key found */
1574 offs = ALIGN(zbr->offs + zbr->len, 8);
1576 if (len > bu->buf_len) {
1582 * The data nodes must be in consecutive positions in
1585 if (zbr->lnum != lnum || zbr->offs != offs)
1587 offs += ALIGN(zbr->len, 8);
1588 len = ALIGN(len, 8) + zbr->len;
1589 /* Must not exceed buffer length */
1590 if (len > bu->buf_len)
1593 /* Allow for holes */
1594 next_block = key_block(c, key);
1595 bu->blk_cnt += (next_block - block - 1);
1596 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1600 bu->zbranch[bu->cnt++] = *zbr;
1602 /* See if we have room for more */
1603 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1605 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1609 if (err == -ENOENT) {
1613 bu->gc_seq = c->gc_seq;
1614 mutex_unlock(&c->tnc_mutex);
1618 * An enormous hole could cause bulk-read to encompass too many
1619 * page cache pages, so limit the number here.
1621 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1622 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1624 * Ensure that bulk-read covers a whole number of page cache
1627 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1628 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1631 /* At the end of file we can round up */
1632 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1635 /* Exclude data nodes that do not make up a whole page cache page */
1636 block = key_block(c, &bu->key) + bu->blk_cnt;
1637 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1639 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1647 * read_wbuf - bulk-read from a LEB with a wbuf.
1648 * @wbuf: wbuf that may overlap the read
1649 * @buf: buffer into which to read
1651 * @lnum: LEB number from which to read
1652 * @offs: offset from which to read
1654 * This functions returns %0 on success or a negative error code on failure.
1656 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1659 const struct ubifs_info *c = wbuf->c;
1662 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1663 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1664 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1665 ubifs_assert(offs + len <= c->leb_size);
1667 spin_lock(&wbuf->lock);
1668 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1670 /* We may safely unlock the write-buffer and read the data */
1671 spin_unlock(&wbuf->lock);
1672 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1675 /* Don't read under wbuf */
1676 rlen = wbuf->offs - offs;
1680 /* Copy the rest from the write-buffer */
1681 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1682 spin_unlock(&wbuf->lock);
1685 /* Read everything that goes before write-buffer */
1686 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1692 * validate_data_node - validate data nodes for bulk-read.
1693 * @c: UBIFS file-system description object
1694 * @buf: buffer containing data node to validate
1695 * @zbr: zbranch of data node to validate
1697 * This functions returns %0 on success or a negative error code on failure.
1699 static int validate_data_node(struct ubifs_info *c, void *buf,
1700 struct ubifs_zbranch *zbr)
1702 union ubifs_key key1;
1703 struct ubifs_ch *ch = buf;
1706 if (ch->node_type != UBIFS_DATA_NODE) {
1707 ubifs_err(c, "bad node type (%d but expected %d)",
1708 ch->node_type, UBIFS_DATA_NODE);
1712 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1714 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1718 len = le32_to_cpu(ch->len);
1719 if (len != zbr->len) {
1720 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1724 /* Make sure the key of the read node is correct */
1725 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1726 if (!keys_eq(c, &zbr->key, &key1)) {
1727 ubifs_err(c, "bad key in node at LEB %d:%d",
1728 zbr->lnum, zbr->offs);
1729 dbg_tnck(&zbr->key, "looked for key ");
1730 dbg_tnck(&key1, "found node's key ");
1739 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1740 ubifs_dump_node(c, buf);
1746 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1747 * @c: UBIFS file-system description object
1748 * @bu: bulk-read parameters and results
1750 * This functions reads and validates the data nodes that were identified by the
1751 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1752 * -EAGAIN to indicate a race with GC, or another negative error code on
1755 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1757 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1758 struct ubifs_wbuf *wbuf;
1761 len = bu->zbranch[bu->cnt - 1].offs;
1762 len += bu->zbranch[bu->cnt - 1].len - offs;
1763 if (len > bu->buf_len) {
1764 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1769 wbuf = ubifs_get_wbuf(c, lnum);
1771 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1773 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1775 /* Check for a race with GC */
1776 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1779 if (err && err != -EBADMSG) {
1780 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1783 dbg_tnck(&bu->key, "key ");
1787 /* Validate the nodes read */
1789 for (i = 0; i < bu->cnt; i++) {
1790 err = validate_data_node(c, buf, &bu->zbranch[i]);
1793 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1800 * do_lookup_nm- look up a "hashed" node.
1801 * @c: UBIFS file-system description object
1802 * @key: node key to lookup
1803 * @node: the node is returned here
1806 * This function look up and reads a node which contains name hash in the key.
1807 * Since the hash may have collisions, there may be many nodes with the same
1808 * key, so we have to sequentially look to all of them until the needed one is
1809 * found. This function returns zero in case of success, %-ENOENT if the node
1810 * was not found, and a negative error code in case of failure.
1812 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1813 void *node, const struct qstr *nm)
1816 struct ubifs_znode *znode;
1818 dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name);
1819 mutex_lock(&c->tnc_mutex);
1820 found = ubifs_lookup_level0(c, key, &znode, &n);
1824 } else if (found < 0) {
1829 ubifs_assert(n >= 0);
1831 err = resolve_collision(c, key, &znode, &n, nm);
1832 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1833 if (unlikely(err < 0))
1840 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1843 mutex_unlock(&c->tnc_mutex);
1848 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1849 * @c: UBIFS file-system description object
1850 * @key: node key to lookup
1851 * @node: the node is returned here
1854 * This function look up and reads a node which contains name hash in the key.
1855 * Since the hash may have collisions, there may be many nodes with the same
1856 * key, so we have to sequentially look to all of them until the needed one is
1857 * found. This function returns zero in case of success, %-ENOENT if the node
1858 * was not found, and a negative error code in case of failure.
1860 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1861 void *node, const struct qstr *nm)
1864 const struct ubifs_dent_node *dent = node;
1867 * We assume that in most of the cases there are no name collisions and
1868 * 'ubifs_tnc_lookup()' returns us the right direntry.
1870 err = ubifs_tnc_lookup(c, key, node);
1874 len = le16_to_cpu(dent->nlen);
1875 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1879 * Unluckily, there are hash collisions and we have to iterate over
1880 * them look at each direntry with colliding name hash sequentially.
1882 return do_lookup_nm(c, key, node, nm);
1886 * correct_parent_keys - correct parent znodes' keys.
1887 * @c: UBIFS file-system description object
1888 * @znode: znode to correct parent znodes for
1890 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1891 * zbranch changes, keys of parent znodes have to be corrected. This helper
1892 * function is called in such situations and corrects the keys if needed.
1894 static void correct_parent_keys(const struct ubifs_info *c,
1895 struct ubifs_znode *znode)
1897 union ubifs_key *key, *key1;
1899 ubifs_assert(znode->parent);
1900 ubifs_assert(znode->iip == 0);
1902 key = &znode->zbranch[0].key;
1903 key1 = &znode->parent->zbranch[0].key;
1905 while (keys_cmp(c, key, key1) < 0) {
1906 key_copy(c, key, key1);
1907 znode = znode->parent;
1909 if (!znode->parent || znode->iip)
1911 key1 = &znode->parent->zbranch[0].key;
1916 * insert_zbranch - insert a zbranch into a znode.
1917 * @znode: znode into which to insert
1918 * @zbr: zbranch to insert
1919 * @n: slot number to insert to
1921 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1922 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1923 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1924 * slot, zbranches starting from @n have to be moved right.
1926 static void insert_zbranch(struct ubifs_znode *znode,
1927 const struct ubifs_zbranch *zbr, int n)
1931 ubifs_assert(ubifs_zn_dirty(znode));
1934 for (i = znode->child_cnt; i > n; i--) {
1935 znode->zbranch[i] = znode->zbranch[i - 1];
1936 if (znode->zbranch[i].znode)
1937 znode->zbranch[i].znode->iip = i;
1940 zbr->znode->iip = n;
1942 for (i = znode->child_cnt; i > n; i--)
1943 znode->zbranch[i] = znode->zbranch[i - 1];
1945 znode->zbranch[n] = *zbr;
1946 znode->child_cnt += 1;
1949 * After inserting at slot zero, the lower bound of the key range of
1950 * this znode may have changed. If this znode is subsequently split
1951 * then the upper bound of the key range may change, and furthermore
1952 * it could change to be lower than the original lower bound. If that
1953 * happens, then it will no longer be possible to find this znode in the
1954 * TNC using the key from the index node on flash. That is bad because
1955 * if it is not found, we will assume it is obsolete and may overwrite
1956 * it. Then if there is an unclean unmount, we will start using the
1957 * old index which will be broken.
1959 * So we first mark znodes that have insertions at slot zero, and then
1960 * if they are split we add their lnum/offs to the old_idx tree.
1967 * tnc_insert - insert a node into TNC.
1968 * @c: UBIFS file-system description object
1969 * @znode: znode to insert into
1970 * @zbr: branch to insert
1971 * @n: slot number to insert new zbranch to
1973 * This function inserts a new node described by @zbr into znode @znode. If
1974 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1975 * are splat as well if needed. Returns zero in case of success or a negative
1976 * error code in case of failure.
1978 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1979 struct ubifs_zbranch *zbr, int n)
1981 struct ubifs_znode *zn, *zi, *zp;
1982 int i, keep, move, appending = 0;
1983 union ubifs_key *key = &zbr->key, *key1;
1985 ubifs_assert(n >= 0 && n <= c->fanout);
1987 /* Implement naive insert for now */
1990 if (znode->child_cnt < c->fanout) {
1991 ubifs_assert(n != c->fanout);
1992 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
1994 insert_zbranch(znode, zbr, n);
1996 /* Ensure parent's key is correct */
1997 if (n == 0 && zp && znode->iip == 0)
1998 correct_parent_keys(c, znode);
2004 * Unfortunately, @znode does not have more empty slots and we have to
2007 dbg_tnck(key, "splitting level %d, key ", znode->level);
2011 * We can no longer be sure of finding this znode by key, so we
2012 * record it in the old_idx tree.
2014 ins_clr_old_idx_znode(c, znode);
2016 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2020 zn->level = znode->level;
2022 /* Decide where to split */
2023 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2024 /* Try not to split consecutive data keys */
2025 if (n == c->fanout) {
2026 key1 = &znode->zbranch[n - 1].key;
2027 if (key_inum(c, key1) == key_inum(c, key) &&
2028 key_type(c, key1) == UBIFS_DATA_KEY)
2032 } else if (appending && n != c->fanout) {
2033 /* Try not to split consecutive data keys */
2036 if (n >= (c->fanout + 1) / 2) {
2037 key1 = &znode->zbranch[0].key;
2038 if (key_inum(c, key1) == key_inum(c, key) &&
2039 key_type(c, key1) == UBIFS_DATA_KEY) {
2040 key1 = &znode->zbranch[n].key;
2041 if (key_inum(c, key1) != key_inum(c, key) ||
2042 key_type(c, key1) != UBIFS_DATA_KEY) {
2044 move = c->fanout - keep;
2056 keep = (c->fanout + 1) / 2;
2057 move = c->fanout - keep;
2061 * Although we don't at present, we could look at the neighbors and see
2062 * if we can move some zbranches there.
2066 /* Insert into existing znode */
2071 /* Insert into new znode */
2076 zbr->znode->parent = zn;
2081 __set_bit(DIRTY_ZNODE, &zn->flags);
2082 atomic_long_inc(&c->dirty_zn_cnt);
2084 zn->child_cnt = move;
2085 znode->child_cnt = keep;
2087 dbg_tnc("moving %d, keeping %d", move, keep);
2090 for (i = 0; i < move; i++) {
2091 zn->zbranch[i] = znode->zbranch[keep + i];
2094 if (zn->zbranch[i].znode) {
2095 zn->zbranch[i].znode->parent = zn;
2096 zn->zbranch[i].znode->iip = i;
2100 /* Insert new key and branch */
2101 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2103 insert_zbranch(zi, zbr, n);
2105 /* Insert new znode (produced by spitting) into the parent */
2107 if (n == 0 && zi == znode && znode->iip == 0)
2108 correct_parent_keys(c, znode);
2110 /* Locate insertion point */
2113 /* Tail recursion */
2114 zbr->key = zn->zbranch[0].key;
2124 /* We have to split root znode */
2125 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2127 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2132 zi->level = znode->level + 1;
2134 __set_bit(DIRTY_ZNODE, &zi->flags);
2135 atomic_long_inc(&c->dirty_zn_cnt);
2137 zi->zbranch[0].key = znode->zbranch[0].key;
2138 zi->zbranch[0].znode = znode;
2139 zi->zbranch[0].lnum = c->zroot.lnum;
2140 zi->zbranch[0].offs = c->zroot.offs;
2141 zi->zbranch[0].len = c->zroot.len;
2142 zi->zbranch[1].key = zn->zbranch[0].key;
2143 zi->zbranch[1].znode = zn;
2148 c->zroot.znode = zi;
2159 * ubifs_tnc_add - add a node to TNC.
2160 * @c: UBIFS file-system description object
2162 * @lnum: LEB number of node
2163 * @offs: node offset
2166 * This function adds a node with key @key to TNC. The node may be new or it may
2167 * obsolete some existing one. Returns %0 on success or negative error code on
2170 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2173 int found, n, err = 0;
2174 struct ubifs_znode *znode;
2176 mutex_lock(&c->tnc_mutex);
2177 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2178 found = lookup_level0_dirty(c, key, &znode, &n);
2180 struct ubifs_zbranch zbr;
2186 key_copy(c, key, &zbr.key);
2187 err = tnc_insert(c, znode, &zbr, n + 1);
2188 } else if (found == 1) {
2189 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2192 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2199 err = dbg_check_tnc(c, 0);
2200 mutex_unlock(&c->tnc_mutex);
2206 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2207 * @c: UBIFS file-system description object
2209 * @old_lnum: LEB number of old node
2210 * @old_offs: old node offset
2211 * @lnum: LEB number of node
2212 * @offs: node offset
2215 * This function replaces a node with key @key in the TNC only if the old node
2216 * is found. This function is called by garbage collection when node are moved.
2217 * Returns %0 on success or negative error code on failure.
2219 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2220 int old_lnum, int old_offs, int lnum, int offs, int len)
2222 int found, n, err = 0;
2223 struct ubifs_znode *znode;
2225 mutex_lock(&c->tnc_mutex);
2226 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2227 old_offs, lnum, offs, len);
2228 found = lookup_level0_dirty(c, key, &znode, &n);
2235 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2238 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2240 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2247 } else if (is_hash_key(c, key)) {
2248 found = resolve_collision_directly(c, key, &znode, &n,
2249 old_lnum, old_offs);
2250 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2251 found, znode, n, old_lnum, old_offs);
2258 /* Ensure the znode is dirtied */
2259 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2260 znode = dirty_cow_bottom_up(c, znode);
2261 if (IS_ERR(znode)) {
2262 err = PTR_ERR(znode);
2266 zbr = &znode->zbranch[n];
2268 err = ubifs_add_dirt(c, zbr->lnum,
2280 err = ubifs_add_dirt(c, lnum, len);
2283 err = dbg_check_tnc(c, 0);
2286 mutex_unlock(&c->tnc_mutex);
2291 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2292 * @c: UBIFS file-system description object
2294 * @lnum: LEB number of node
2295 * @offs: node offset
2299 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2300 * may have collisions, like directory entry keys.
2302 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2303 int lnum, int offs, int len, const struct qstr *nm)
2305 int found, n, err = 0;
2306 struct ubifs_znode *znode;
2308 mutex_lock(&c->tnc_mutex);
2309 dbg_tnck(key, "LEB %d:%d, name '%.*s', key ",
2310 lnum, offs, nm->len, nm->name);
2311 found = lookup_level0_dirty(c, key, &znode, &n);
2319 found = fallible_resolve_collision(c, key, &znode, &n,
2322 found = resolve_collision(c, key, &znode, &n, nm);
2323 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2329 /* Ensure the znode is dirtied */
2330 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2331 znode = dirty_cow_bottom_up(c, znode);
2332 if (IS_ERR(znode)) {
2333 err = PTR_ERR(znode);
2339 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2342 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2351 struct ubifs_zbranch zbr;
2357 key_copy(c, key, &zbr.key);
2358 err = tnc_insert(c, znode, &zbr, n + 1);
2363 * We did not find it in the index so there may be a
2364 * dangling branch still in the index. So we remove it
2365 * by passing 'ubifs_tnc_remove_nm()' the same key but
2366 * an unmatchable name.
2368 struct qstr noname = { .name = "" };
2370 err = dbg_check_tnc(c, 0);
2371 mutex_unlock(&c->tnc_mutex);
2374 return ubifs_tnc_remove_nm(c, key, &noname);
2380 err = dbg_check_tnc(c, 0);
2381 mutex_unlock(&c->tnc_mutex);
2386 * tnc_delete - delete a znode form TNC.
2387 * @c: UBIFS file-system description object
2388 * @znode: znode to delete from
2389 * @n: zbranch slot number to delete
2391 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2392 * case of success and a negative error code in case of failure.
2394 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2396 struct ubifs_zbranch *zbr;
2397 struct ubifs_znode *zp;
2400 /* Delete without merge for now */
2401 ubifs_assert(znode->level == 0);
2402 ubifs_assert(n >= 0 && n < c->fanout);
2403 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2405 zbr = &znode->zbranch[n];
2408 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2410 ubifs_dump_znode(c, znode);
2414 /* We do not "gap" zbranch slots */
2415 for (i = n; i < znode->child_cnt - 1; i++)
2416 znode->zbranch[i] = znode->zbranch[i + 1];
2417 znode->child_cnt -= 1;
2419 if (znode->child_cnt > 0)
2423 * This was the last zbranch, we have to delete this znode from the
2428 ubifs_assert(!ubifs_zn_obsolete(znode));
2429 ubifs_assert(ubifs_zn_dirty(znode));
2434 atomic_long_dec(&c->dirty_zn_cnt);
2436 err = insert_old_idx_znode(c, znode);
2441 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2442 atomic_long_inc(&c->clean_zn_cnt);
2443 atomic_long_inc(&ubifs_clean_zn_cnt);
2447 } while (znode->child_cnt == 1); /* while removing last child */
2449 /* Remove from znode, entry n - 1 */
2450 znode->child_cnt -= 1;
2451 ubifs_assert(znode->level != 0);
2452 for (i = n; i < znode->child_cnt; i++) {
2453 znode->zbranch[i] = znode->zbranch[i + 1];
2454 if (znode->zbranch[i].znode)
2455 znode->zbranch[i].znode->iip = i;
2459 * If this is the root and it has only 1 child then
2460 * collapse the tree.
2462 if (!znode->parent) {
2463 while (znode->child_cnt == 1 && znode->level != 0) {
2465 zbr = &znode->zbranch[0];
2466 znode = get_znode(c, znode, 0);
2468 return PTR_ERR(znode);
2469 znode = dirty_cow_znode(c, zbr);
2471 return PTR_ERR(znode);
2472 znode->parent = NULL;
2475 err = insert_old_idx(c, c->zroot.lnum,
2480 c->zroot.lnum = zbr->lnum;
2481 c->zroot.offs = zbr->offs;
2482 c->zroot.len = zbr->len;
2483 c->zroot.znode = znode;
2484 ubifs_assert(!ubifs_zn_obsolete(zp));
2485 ubifs_assert(ubifs_zn_dirty(zp));
2486 atomic_long_dec(&c->dirty_zn_cnt);
2489 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2490 atomic_long_inc(&c->clean_zn_cnt);
2491 atomic_long_inc(&ubifs_clean_zn_cnt);
2501 * ubifs_tnc_remove - remove an index entry of a node.
2502 * @c: UBIFS file-system description object
2505 * Returns %0 on success or negative error code on failure.
2507 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2509 int found, n, err = 0;
2510 struct ubifs_znode *znode;
2512 mutex_lock(&c->tnc_mutex);
2513 dbg_tnck(key, "key ");
2514 found = lookup_level0_dirty(c, key, &znode, &n);
2520 err = tnc_delete(c, znode, n);
2522 err = dbg_check_tnc(c, 0);
2525 mutex_unlock(&c->tnc_mutex);
2530 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2531 * @c: UBIFS file-system description object
2533 * @nm: directory entry name
2535 * Returns %0 on success or negative error code on failure.
2537 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2538 const struct qstr *nm)
2541 struct ubifs_znode *znode;
2543 mutex_lock(&c->tnc_mutex);
2544 dbg_tnck(key, "%.*s, key ", nm->len, nm->name);
2545 err = lookup_level0_dirty(c, key, &znode, &n);
2551 err = fallible_resolve_collision(c, key, &znode, &n,
2554 err = resolve_collision(c, key, &znode, &n, nm);
2555 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2559 /* Ensure the znode is dirtied */
2560 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2561 znode = dirty_cow_bottom_up(c, znode);
2562 if (IS_ERR(znode)) {
2563 err = PTR_ERR(znode);
2567 err = tnc_delete(c, znode, n);
2573 err = dbg_check_tnc(c, 0);
2574 mutex_unlock(&c->tnc_mutex);
2579 * key_in_range - determine if a key falls within a range of keys.
2580 * @c: UBIFS file-system description object
2581 * @key: key to check
2582 * @from_key: lowest key in range
2583 * @to_key: highest key in range
2585 * This function returns %1 if the key is in range and %0 otherwise.
2587 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2588 union ubifs_key *from_key, union ubifs_key *to_key)
2590 if (keys_cmp(c, key, from_key) < 0)
2592 if (keys_cmp(c, key, to_key) > 0)
2598 * ubifs_tnc_remove_range - remove index entries in range.
2599 * @c: UBIFS file-system description object
2600 * @from_key: lowest key to remove
2601 * @to_key: highest key to remove
2603 * This function removes index entries starting at @from_key and ending at
2604 * @to_key. This function returns zero in case of success and a negative error
2605 * code in case of failure.
2607 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2608 union ubifs_key *to_key)
2610 int i, n, k, err = 0;
2611 struct ubifs_znode *znode;
2612 union ubifs_key *key;
2614 mutex_lock(&c->tnc_mutex);
2616 /* Find first level 0 znode that contains keys to remove */
2617 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2624 err = tnc_next(c, &znode, &n);
2625 if (err == -ENOENT) {
2631 key = &znode->zbranch[n].key;
2632 if (!key_in_range(c, key, from_key, to_key)) {
2638 /* Ensure the znode is dirtied */
2639 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2640 znode = dirty_cow_bottom_up(c, znode);
2641 if (IS_ERR(znode)) {
2642 err = PTR_ERR(znode);
2647 /* Remove all keys in range except the first */
2648 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2649 key = &znode->zbranch[i].key;
2650 if (!key_in_range(c, key, from_key, to_key))
2652 lnc_free(&znode->zbranch[i]);
2653 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2654 znode->zbranch[i].len);
2656 ubifs_dump_znode(c, znode);
2659 dbg_tnck(key, "removing key ");
2662 for (i = n + 1 + k; i < znode->child_cnt; i++)
2663 znode->zbranch[i - k] = znode->zbranch[i];
2664 znode->child_cnt -= k;
2667 /* Now delete the first */
2668 err = tnc_delete(c, znode, n);
2675 err = dbg_check_tnc(c, 0);
2676 mutex_unlock(&c->tnc_mutex);
2681 * ubifs_tnc_remove_ino - remove an inode from TNC.
2682 * @c: UBIFS file-system description object
2683 * @inum: inode number to remove
2685 * This function remove inode @inum and all the extended attributes associated
2686 * with the anode from TNC and returns zero in case of success or a negative
2687 * error code in case of failure.
2689 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2691 union ubifs_key key1, key2;
2692 struct ubifs_dent_node *xent, *pxent = NULL;
2693 struct qstr nm = { .name = NULL };
2695 dbg_tnc("ino %lu", (unsigned long)inum);
2698 * Walk all extended attribute entries and remove them together with
2699 * corresponding extended attribute inodes.
2701 lowest_xent_key(c, &key1, inum);
2706 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2708 err = PTR_ERR(xent);
2714 xattr_inum = le64_to_cpu(xent->inum);
2715 dbg_tnc("xent '%s', ino %lu", xent->name,
2716 (unsigned long)xattr_inum);
2718 nm.name = xent->name;
2719 nm.len = le16_to_cpu(xent->nlen);
2720 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2726 lowest_ino_key(c, &key1, xattr_inum);
2727 highest_ino_key(c, &key2, xattr_inum);
2728 err = ubifs_tnc_remove_range(c, &key1, &key2);
2736 key_read(c, &xent->key, &key1);
2740 lowest_ino_key(c, &key1, inum);
2741 highest_ino_key(c, &key2, inum);
2743 return ubifs_tnc_remove_range(c, &key1, &key2);
2747 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2748 * @c: UBIFS file-system description object
2749 * @key: key of last entry
2750 * @nm: name of last entry found or %NULL
2752 * This function finds and reads the next directory or extended attribute entry
2753 * after the given key (@key) if there is one. @nm is used to resolve
2756 * If the name of the current entry is not known and only the key is known,
2757 * @nm->name has to be %NULL. In this case the semantics of this function is a
2758 * little bit different and it returns the entry corresponding to this key, not
2759 * the next one. If the key was not found, the closest "right" entry is
2762 * If the fist entry has to be found, @key has to contain the lowest possible
2763 * key value for this inode and @name has to be %NULL.
2765 * This function returns the found directory or extended attribute entry node
2766 * in case of success, %-ENOENT is returned if no entry was found, and a
2767 * negative error code is returned in case of failure.
2769 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2770 union ubifs_key *key,
2771 const struct qstr *nm)
2773 int n, err, type = key_type(c, key);
2774 struct ubifs_znode *znode;
2775 struct ubifs_dent_node *dent;
2776 struct ubifs_zbranch *zbr;
2777 union ubifs_key *dkey;
2779 dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)");
2780 ubifs_assert(is_hash_key(c, key));
2782 mutex_lock(&c->tnc_mutex);
2783 err = ubifs_lookup_level0(c, key, &znode, &n);
2784 if (unlikely(err < 0))
2789 /* Handle collisions */
2791 err = fallible_resolve_collision(c, key, &znode, &n,
2794 err = resolve_collision(c, key, &znode, &n, nm);
2795 dbg_tnc("rc returned %d, znode %p, n %d",
2797 if (unlikely(err < 0))
2801 /* Now find next entry */
2802 err = tnc_next(c, &znode, &n);
2807 * The full name of the entry was not given, in which case the
2808 * behavior of this function is a little different and it
2809 * returns current entry, not the next one.
2813 * However, the given key does not exist in the TNC
2814 * tree and @znode/@n variables contain the closest
2815 * "preceding" element. Switch to the next one.
2817 err = tnc_next(c, &znode, &n);
2823 zbr = &znode->zbranch[n];
2824 dent = kmalloc(zbr->len, GFP_NOFS);
2825 if (unlikely(!dent)) {
2831 * The above 'tnc_next()' call could lead us to the next inode, check
2835 if (key_inum(c, dkey) != key_inum(c, key) ||
2836 key_type(c, dkey) != type) {
2841 err = tnc_read_node_nm(c, zbr, dent);
2845 mutex_unlock(&c->tnc_mutex);
2851 mutex_unlock(&c->tnc_mutex);
2852 return ERR_PTR(err);
2856 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2857 * @c: UBIFS file-system description object
2859 * Destroy left-over obsolete znodes from a failed commit.
2861 static void tnc_destroy_cnext(struct ubifs_info *c)
2863 struct ubifs_znode *cnext;
2867 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2870 struct ubifs_znode *znode = cnext;
2872 cnext = cnext->cnext;
2873 if (ubifs_zn_obsolete(znode))
2875 } while (cnext && cnext != c->cnext);
2879 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2880 * @c: UBIFS file-system description object
2882 void ubifs_tnc_close(struct ubifs_info *c)
2884 tnc_destroy_cnext(c);
2885 if (c->zroot.znode) {
2888 n = atomic_long_read(&c->clean_zn_cnt);
2889 freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
2890 ubifs_assert(freed == n);
2891 atomic_long_sub(n, &ubifs_clean_zn_cnt);
2899 * left_znode - get the znode to the left.
2900 * @c: UBIFS file-system description object
2903 * This function returns a pointer to the znode to the left of @znode or NULL if
2904 * there is not one. A negative error code is returned on failure.
2906 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2907 struct ubifs_znode *znode)
2909 int level = znode->level;
2912 int n = znode->iip - 1;
2914 /* Go up until we can go left */
2915 znode = znode->parent;
2919 /* Now go down the rightmost branch to 'level' */
2920 znode = get_znode(c, znode, n);
2923 while (znode->level != level) {
2924 n = znode->child_cnt - 1;
2925 znode = get_znode(c, znode, n);
2936 * right_znode - get the znode to the right.
2937 * @c: UBIFS file-system description object
2940 * This function returns a pointer to the znode to the right of @znode or NULL
2941 * if there is not one. A negative error code is returned on failure.
2943 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2944 struct ubifs_znode *znode)
2946 int level = znode->level;
2949 int n = znode->iip + 1;
2951 /* Go up until we can go right */
2952 znode = znode->parent;
2955 if (n < znode->child_cnt) {
2956 /* Now go down the leftmost branch to 'level' */
2957 znode = get_znode(c, znode, n);
2960 while (znode->level != level) {
2961 znode = get_znode(c, znode, 0);
2972 * lookup_znode - find a particular indexing node from TNC.
2973 * @c: UBIFS file-system description object
2974 * @key: index node key to lookup
2975 * @level: index node level
2976 * @lnum: index node LEB number
2977 * @offs: index node offset
2979 * This function searches an indexing node by its first key @key and its
2980 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2981 * nodes it traverses to TNC. This function is called for indexing nodes which
2982 * were found on the media by scanning, for example when garbage-collecting or
2983 * when doing in-the-gaps commit. This means that the indexing node which is
2984 * looked for does not have to have exactly the same leftmost key @key, because
2985 * the leftmost key may have been changed, in which case TNC will contain a
2986 * dirty znode which still refers the same @lnum:@offs. This function is clever
2987 * enough to recognize such indexing nodes.
2989 * Note, if a znode was deleted or changed too much, then this function will
2990 * not find it. For situations like this UBIFS has the old index RB-tree
2991 * (indexed by @lnum:@offs).
2993 * This function returns a pointer to the znode found or %NULL if it is not
2994 * found. A negative error code is returned on failure.
2996 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2997 union ubifs_key *key, int level,
3000 struct ubifs_znode *znode, *zn;
3003 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
3006 * The arguments have probably been read off flash, so don't assume
3010 return ERR_PTR(-EINVAL);
3012 /* Get the root znode */
3013 znode = c->zroot.znode;
3015 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3019 /* Check if it is the one we are looking for */
3020 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3022 /* Descend to the parent level i.e. (level + 1) */
3023 if (level >= znode->level)
3026 ubifs_search_zbranch(c, znode, key, &n);
3029 * We reached a znode where the leftmost key is greater
3030 * than the key we are searching for. This is the same
3031 * situation as the one described in a huge comment at
3032 * the end of the 'ubifs_lookup_level0()' function. And
3033 * for exactly the same reasons we have to try to look
3034 * left before giving up.
3036 znode = left_znode(c, znode);
3041 ubifs_search_zbranch(c, znode, key, &n);
3042 ubifs_assert(n >= 0);
3044 if (znode->level == level + 1)
3046 znode = get_znode(c, znode, n);
3050 /* Check if the child is the one we are looking for */
3051 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3052 return get_znode(c, znode, n);
3053 /* If the key is unique, there is nowhere else to look */
3054 if (!is_hash_key(c, key))
3057 * The key is not unique and so may be also in the znodes to either
3064 /* Move one branch to the left */
3068 znode = left_znode(c, znode);
3073 n = znode->child_cnt - 1;
3076 if (znode->zbranch[n].lnum == lnum &&
3077 znode->zbranch[n].offs == offs)
3078 return get_znode(c, znode, n);
3079 /* Stop if the key is less than the one we are looking for */
3080 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3083 /* Back to the middle */
3088 /* Move one branch to the right */
3089 if (++n >= znode->child_cnt) {
3090 znode = right_znode(c, znode);
3098 if (znode->zbranch[n].lnum == lnum &&
3099 znode->zbranch[n].offs == offs)
3100 return get_znode(c, znode, n);
3101 /* Stop if the key is greater than the one we are looking for */
3102 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3109 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3110 * @c: UBIFS file-system description object
3111 * @key: key of index node
3112 * @level: index node level
3113 * @lnum: LEB number of index node
3114 * @offs: offset of index node
3116 * This function returns %0 if the index node is not referred to in the TNC, %1
3117 * if the index node is referred to in the TNC and the corresponding znode is
3118 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3119 * znode is clean, and a negative error code in case of failure.
3121 * Note, the @key argument has to be the key of the first child. Also note,
3122 * this function relies on the fact that 0:0 is never a valid LEB number and
3123 * offset for a main-area node.
3125 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3128 struct ubifs_znode *znode;
3130 znode = lookup_znode(c, key, level, lnum, offs);
3134 return PTR_ERR(znode);
3136 return ubifs_zn_dirty(znode) ? 1 : 2;
3140 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3141 * @c: UBIFS file-system description object
3143 * @lnum: node LEB number
3144 * @offs: node offset
3146 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3147 * not, and a negative error code in case of failure.
3149 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3150 * and offset for a main-area node.
3152 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3155 struct ubifs_zbranch *zbr;
3156 struct ubifs_znode *znode, *zn;
3157 int n, found, err, nn;
3158 const int unique = !is_hash_key(c, key);
3160 found = ubifs_lookup_level0(c, key, &znode, &n);
3162 return found; /* Error code */
3165 zbr = &znode->zbranch[n];
3166 if (lnum == zbr->lnum && offs == zbr->offs)
3167 return 1; /* Found it */
3171 * Because the key is not unique, we have to look left
3178 err = tnc_prev(c, &znode, &n);
3183 if (keys_cmp(c, key, &znode->zbranch[n].key))
3185 zbr = &znode->zbranch[n];
3186 if (lnum == zbr->lnum && offs == zbr->offs)
3187 return 1; /* Found it */
3193 err = tnc_next(c, &znode, &n);
3199 if (keys_cmp(c, key, &znode->zbranch[n].key))
3201 zbr = &znode->zbranch[n];
3202 if (lnum == zbr->lnum && offs == zbr->offs)
3203 return 1; /* Found it */
3209 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3210 * @c: UBIFS file-system description object
3212 * @level: index node level (if it is an index node)
3213 * @lnum: node LEB number
3214 * @offs: node offset
3215 * @is_idx: non-zero if the node is an index node
3217 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3218 * negative error code in case of failure. For index nodes, @key has to be the
3219 * key of the first child. An index node is considered to be in the TNC only if
3220 * the corresponding znode is clean or has not been loaded.
3222 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3223 int lnum, int offs, int is_idx)
3227 mutex_lock(&c->tnc_mutex);
3229 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3233 /* The index node was found but it was dirty */
3236 /* The index node was found and it was clean */
3241 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3244 mutex_unlock(&c->tnc_mutex);
3249 * ubifs_dirty_idx_node - dirty an index node.
3250 * @c: UBIFS file-system description object
3251 * @key: index node key
3252 * @level: index node level
3253 * @lnum: index node LEB number
3254 * @offs: index node offset
3256 * This function loads and dirties an index node so that it can be garbage
3257 * collected. The @key argument has to be the key of the first child. This
3258 * function relies on the fact that 0:0 is never a valid LEB number and offset
3259 * for a main-area node. Returns %0 on success and a negative error code on
3262 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3265 struct ubifs_znode *znode;
3268 mutex_lock(&c->tnc_mutex);
3269 znode = lookup_znode(c, key, level, lnum, offs);
3272 if (IS_ERR(znode)) {
3273 err = PTR_ERR(znode);
3276 znode = dirty_cow_bottom_up(c, znode);
3277 if (IS_ERR(znode)) {
3278 err = PTR_ERR(znode);
3283 mutex_unlock(&c->tnc_mutex);
3288 * dbg_check_inode_size - check if inode size is correct.
3289 * @c: UBIFS file-system description object
3290 * @inum: inode number
3293 * This function makes sure that the inode size (@size) is correct and it does
3294 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3295 * if it has a data page beyond @size, and other negative error code in case of
3298 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3302 union ubifs_key from_key, to_key, *key;
3303 struct ubifs_znode *znode;
3306 if (!S_ISREG(inode->i_mode))
3308 if (!dbg_is_chk_gen(c))
3311 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3312 data_key_init(c, &from_key, inode->i_ino, block);
3313 highest_data_key(c, &to_key, inode->i_ino);
3315 mutex_lock(&c->tnc_mutex);
3316 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3325 err = tnc_next(c, &znode, &n);
3326 if (err == -ENOENT) {
3333 ubifs_assert(err == 0);
3334 key = &znode->zbranch[n].key;
3335 if (!key_in_range(c, key, &from_key, &to_key))
3339 block = key_block(c, key);
3340 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3341 (unsigned long)inode->i_ino, size,
3342 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3343 mutex_unlock(&c->tnc_mutex);
3344 ubifs_dump_inode(c, inode);
3349 mutex_unlock(&c->tnc_mutex);