2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
27 * At the moment the locking rules of the TNC tree are quite simple and
28 * straightforward. We just have a mutex and lock it when we traverse the
29 * tree. If a znode is not in memory, we read it from flash while still having
33 #include <linux/crc32.h>
34 #include <linux/slab.h>
37 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
38 int len, int lnum, int offs);
39 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
40 struct ubifs_zbranch *zbr, void *node);
43 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
44 * @NAME_LESS: name corresponding to the first argument is less than second
45 * @NAME_MATCHES: names match
46 * @NAME_GREATER: name corresponding to the second argument is greater than
48 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
50 * These constants were introduce to improve readability.
60 * insert_old_idx - record an index node obsoleted since the last commit start.
61 * @c: UBIFS file-system description object
62 * @lnum: LEB number of obsoleted index node
63 * @offs: offset of obsoleted index node
65 * Returns %0 on success, and a negative error code on failure.
67 * For recovery, there must always be a complete intact version of the index on
68 * flash at all times. That is called the "old index". It is the index as at the
69 * time of the last successful commit. Many of the index nodes in the old index
70 * may be dirty, but they must not be erased until the next successful commit
71 * (at which point that index becomes the old index).
73 * That means that the garbage collection and the in-the-gaps method of
74 * committing must be able to determine if an index node is in the old index.
75 * Most of the old index nodes can be found by looking up the TNC using the
76 * 'lookup_znode()' function. However, some of the old index nodes may have
77 * been deleted from the current index or may have been changed so much that
78 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
79 * That is what this function does. The RB-tree is ordered by LEB number and
80 * offset because they uniquely identify the old index node.
82 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
84 struct ubifs_old_idx *old_idx, *o;
85 struct rb_node **p, *parent = NULL;
87 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
88 if (unlikely(!old_idx))
93 p = &c->old_idx.rb_node;
96 o = rb_entry(parent, struct ubifs_old_idx, rb);
99 else if (lnum > o->lnum)
101 else if (offs < o->offs)
103 else if (offs > o->offs)
106 ubifs_err("old idx added twice!");
111 rb_link_node(&old_idx->rb, parent, p);
112 rb_insert_color(&old_idx->rb, &c->old_idx);
117 * insert_old_idx_znode - record a znode obsoleted since last commit start.
118 * @c: UBIFS file-system description object
119 * @znode: znode of obsoleted index node
121 * Returns %0 on success, and a negative error code on failure.
123 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
126 struct ubifs_zbranch *zbr;
128 zbr = &znode->parent->zbranch[znode->iip];
130 return insert_old_idx(c, zbr->lnum, zbr->offs);
133 return insert_old_idx(c, c->zroot.lnum,
139 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
140 * @c: UBIFS file-system description object
141 * @znode: znode of obsoleted index node
143 * Returns %0 on success, and a negative error code on failure.
145 static int ins_clr_old_idx_znode(struct ubifs_info *c,
146 struct ubifs_znode *znode)
151 struct ubifs_zbranch *zbr;
153 zbr = &znode->parent->zbranch[znode->iip];
155 err = insert_old_idx(c, zbr->lnum, zbr->offs);
164 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
175 * destroy_old_idx - destroy the old_idx RB-tree.
176 * @c: UBIFS file-system description object
178 * During start commit, the old_idx RB-tree is used to avoid overwriting index
179 * nodes that were in the index last commit but have since been deleted. This
180 * is necessary for recovery i.e. the old index must be kept intact until the
181 * new index is successfully written. The old-idx RB-tree is used for the
182 * in-the-gaps method of writing index nodes and is destroyed every commit.
184 void destroy_old_idx(struct ubifs_info *c)
186 struct rb_node *this = c->old_idx.rb_node;
187 struct ubifs_old_idx *old_idx;
191 this = this->rb_left;
193 } else if (this->rb_right) {
194 this = this->rb_right;
197 old_idx = rb_entry(this, struct ubifs_old_idx, rb);
198 this = rb_parent(this);
200 if (this->rb_left == &old_idx->rb)
201 this->rb_left = NULL;
203 this->rb_right = NULL;
207 c->old_idx = RB_ROOT;
211 * copy_znode - copy a dirty znode.
212 * @c: UBIFS file-system description object
213 * @znode: znode to copy
215 * A dirty znode being committed may not be changed, so it is copied.
217 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
218 struct ubifs_znode *znode)
220 struct ubifs_znode *zn;
222 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
224 return ERR_PTR(-ENOMEM);
226 memcpy(zn, znode, c->max_znode_sz);
228 __set_bit(DIRTY_ZNODE, &zn->flags);
229 __clear_bit(COW_ZNODE, &zn->flags);
231 ubifs_assert(!ubifs_zn_obsolete(znode));
232 __set_bit(OBSOLETE_ZNODE, &znode->flags);
234 if (znode->level != 0) {
236 const int n = zn->child_cnt;
238 /* The children now have new parent */
239 for (i = 0; i < n; i++) {
240 struct ubifs_zbranch *zbr = &zn->zbranch[i];
243 zbr->znode->parent = zn;
247 atomic_long_inc(&c->dirty_zn_cnt);
252 * add_idx_dirt - add dirt due to a dirty znode.
253 * @c: UBIFS file-system description object
254 * @lnum: LEB number of index node
255 * @dirt: size of index node
257 * This function updates lprops dirty space and the new size of the index.
259 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
261 c->calc_idx_sz -= ALIGN(dirt, 8);
262 return ubifs_add_dirt(c, lnum, dirt);
266 * dirty_cow_znode - ensure a znode is not being committed.
267 * @c: UBIFS file-system description object
268 * @zbr: branch of znode to check
270 * Returns dirtied znode on success or negative error code on failure.
272 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
273 struct ubifs_zbranch *zbr)
275 struct ubifs_znode *znode = zbr->znode;
276 struct ubifs_znode *zn;
279 if (!ubifs_zn_cow(znode)) {
280 /* znode is not being committed */
281 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
282 atomic_long_inc(&c->dirty_zn_cnt);
283 atomic_long_dec(&c->clean_zn_cnt);
284 atomic_long_dec(&ubifs_clean_zn_cnt);
285 err = add_idx_dirt(c, zbr->lnum, zbr->len);
292 zn = copy_znode(c, znode);
297 err = insert_old_idx(c, zbr->lnum, zbr->offs);
300 err = add_idx_dirt(c, zbr->lnum, zbr->len);
315 * lnc_add - add a leaf node to the leaf node cache.
316 * @c: UBIFS file-system description object
317 * @zbr: zbranch of leaf node
320 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
321 * purpose of the leaf node cache is to save re-reading the same leaf node over
322 * and over again. Most things are cached by VFS, however the file system must
323 * cache directory entries for readdir and for resolving hash collisions. The
324 * present implementation of the leaf node cache is extremely simple, and
325 * allows for error returns that are not used but that may be needed if a more
326 * complex implementation is created.
328 * Note, this function does not add the @node object to LNC directly, but
329 * allocates a copy of the object and adds the copy to LNC. The reason for this
330 * is that @node has been allocated outside of the TNC subsystem and will be
331 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
332 * may be changed at any time, e.g. freed by the shrinker.
334 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
339 const struct ubifs_dent_node *dent = node;
341 ubifs_assert(!zbr->leaf);
342 ubifs_assert(zbr->len != 0);
343 ubifs_assert(is_hash_key(c, &zbr->key));
345 err = ubifs_validate_entry(c, dent);
348 dbg_dump_node(c, dent);
352 lnc_node = kmalloc(zbr->len, GFP_NOFS);
354 /* We don't have to have the cache, so no error */
357 memcpy(lnc_node, node, zbr->len);
358 zbr->leaf = lnc_node;
363 * lnc_add_directly - add a leaf node to the leaf-node-cache.
364 * @c: UBIFS file-system description object
365 * @zbr: zbranch of leaf node
368 * This function is similar to 'lnc_add()', but it does not create a copy of
369 * @node but inserts @node to TNC directly.
371 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
376 ubifs_assert(!zbr->leaf);
377 ubifs_assert(zbr->len != 0);
379 err = ubifs_validate_entry(c, node);
382 dbg_dump_node(c, node);
391 * lnc_free - remove a leaf node from the leaf node cache.
392 * @zbr: zbranch of leaf node
395 static void lnc_free(struct ubifs_zbranch *zbr)
404 * tnc_read_node_nm - read a "hashed" leaf node.
405 * @c: UBIFS file-system description object
406 * @zbr: key and position of the node
407 * @node: node is returned here
409 * This function reads a "hashed" node defined by @zbr from the leaf node cache
410 * (in it is there) or from the hash media, in which case the node is also
411 * added to LNC. Returns zero in case of success or a negative negative error
412 * code in case of failure.
414 static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
419 ubifs_assert(is_hash_key(c, &zbr->key));
422 /* Read from the leaf node cache */
423 ubifs_assert(zbr->len != 0);
424 memcpy(node, zbr->leaf, zbr->len);
429 err = fallible_read_node(c, &zbr->key, zbr, node);
431 * When the node was not found, return -ENOENT, 0 otherwise.
432 * Negative return codes stay as-is.
439 err = ubifs_tnc_read_node(c, zbr, node);
444 /* Add the node to the leaf node cache */
445 err = lnc_add(c, zbr, node);
450 * try_read_node - read a node if it is a node.
451 * @c: UBIFS file-system description object
452 * @buf: buffer to read to
454 * @len: node length (not aligned)
455 * @lnum: LEB number of node to read
456 * @offs: offset of node to read
458 * This function tries to read a node of known type and length, checks it and
459 * stores it in @buf. This function returns %1 if a node is present and %0 if
460 * a node is not present. A negative error code is returned for I/O errors.
461 * This function performs that same function as ubifs_read_node except that
462 * it does not require that there is actually a node present and instead
463 * the return code indicates if a node was read.
465 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
466 * is true (it is controlled by corresponding mount option). However, if
467 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
468 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
469 * because during mounting or re-mounting from R/O mode to R/W mode we may read
470 * journal nodes (when replying the journal or doing the recovery) and the
471 * journal nodes may potentially be corrupted, so checking is required.
473 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
474 int len, int lnum, int offs)
477 struct ubifs_ch *ch = buf;
478 uint32_t crc, node_crc;
480 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
482 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
484 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
485 type, lnum, offs, err);
489 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
492 if (ch->node_type != type)
495 node_len = le32_to_cpu(ch->len);
499 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
503 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
504 node_crc = le32_to_cpu(ch->crc);
512 * fallible_read_node - try to read a leaf node.
513 * @c: UBIFS file-system description object
514 * @key: key of node to read
515 * @zbr: position of node
516 * @node: node returned
518 * This function tries to read a node and returns %1 if the node is read, %0
519 * if the node is not present, and a negative error code in the case of error.
521 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
522 struct ubifs_zbranch *zbr, void *node)
526 dbg_tnc("LEB %d:%d, key %s", zbr->lnum, zbr->offs, DBGKEY(key));
528 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
531 union ubifs_key node_key;
532 struct ubifs_dent_node *dent = node;
534 /* All nodes have key in the same place */
535 key_read(c, &dent->key, &node_key);
536 if (keys_cmp(c, key, &node_key) != 0)
539 if (ret == 0 && c->replaying)
540 dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
541 zbr->lnum, zbr->offs, zbr->len, DBGKEY(key));
546 * matches_name - determine if a direntry or xattr entry matches a given name.
547 * @c: UBIFS file-system description object
548 * @zbr: zbranch of dent
551 * This function checks if xentry/direntry referred by zbranch @zbr matches name
552 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
553 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
554 * of failure, a negative error code is returned.
556 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
557 const struct qstr *nm)
559 struct ubifs_dent_node *dent;
562 /* If possible, match against the dent in the leaf node cache */
564 dent = kmalloc(zbr->len, GFP_NOFS);
568 err = ubifs_tnc_read_node(c, zbr, dent);
572 /* Add the node to the leaf node cache */
573 err = lnc_add_directly(c, zbr, dent);
579 nlen = le16_to_cpu(dent->nlen);
580 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
584 else if (nlen < nm->len)
599 * get_znode - get a TNC znode that may not be loaded yet.
600 * @c: UBIFS file-system description object
601 * @znode: parent znode
602 * @n: znode branch slot number
604 * This function returns the znode or a negative error code.
606 static struct ubifs_znode *get_znode(struct ubifs_info *c,
607 struct ubifs_znode *znode, int n)
609 struct ubifs_zbranch *zbr;
611 zbr = &znode->zbranch[n];
615 znode = ubifs_load_znode(c, zbr, znode, n);
620 * tnc_next - find next TNC entry.
621 * @c: UBIFS file-system description object
622 * @zn: znode is passed and returned here
623 * @n: znode branch slot number is passed and returned here
625 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
626 * no next entry, or a negative error code otherwise.
628 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
630 struct ubifs_znode *znode = *zn;
634 if (nn < znode->child_cnt) {
639 struct ubifs_znode *zp;
646 if (nn < znode->child_cnt) {
647 znode = get_znode(c, znode, nn);
649 return PTR_ERR(znode);
650 while (znode->level != 0) {
651 znode = get_znode(c, znode, 0);
653 return PTR_ERR(znode);
665 * tnc_prev - find previous TNC entry.
666 * @c: UBIFS file-system description object
667 * @zn: znode is returned here
668 * @n: znode branch slot number is passed and returned here
670 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
671 * there is no next entry, or a negative error code otherwise.
673 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
675 struct ubifs_znode *znode = *zn;
683 struct ubifs_znode *zp;
691 znode = get_znode(c, znode, nn);
693 return PTR_ERR(znode);
694 while (znode->level != 0) {
695 nn = znode->child_cnt - 1;
696 znode = get_znode(c, znode, nn);
698 return PTR_ERR(znode);
700 nn = znode->child_cnt - 1;
710 * resolve_collision - resolve a collision.
711 * @c: UBIFS file-system description object
712 * @key: key of a directory or extended attribute entry
713 * @zn: znode is returned here
714 * @n: zbranch number is passed and returned here
715 * @nm: name of the entry
717 * This function is called for "hashed" keys to make sure that the found key
718 * really corresponds to the looked up node (directory or extended attribute
719 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
720 * %0 is returned if @nm is not found and @zn and @n are set to the previous
721 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
722 * This means that @n may be set to %-1 if the leftmost key in @zn is the
723 * previous one. A negative error code is returned on failures.
725 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
726 struct ubifs_znode **zn, int *n,
727 const struct qstr *nm)
731 err = matches_name(c, &(*zn)->zbranch[*n], nm);
732 if (unlikely(err < 0))
734 if (err == NAME_MATCHES)
737 if (err == NAME_GREATER) {
740 err = tnc_prev(c, zn, n);
741 if (err == -ENOENT) {
742 ubifs_assert(*n == 0);
748 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
750 * We have found the branch after which we would
751 * like to insert, but inserting in this znode
752 * may still be wrong. Consider the following 3
753 * znodes, in the case where we are resolving a
754 * collision with Key2.
757 * ----------------------
758 * level 1 | Key0 | Key1 |
759 * -----------------------
761 * znode za | | znode zb
762 * ------------ ------------
763 * level 0 | Key0 | | Key2 |
764 * ------------ ------------
766 * The lookup finds Key2 in znode zb. Lets say
767 * there is no match and the name is greater so
768 * we look left. When we find Key0, we end up
769 * here. If we return now, we will insert into
770 * znode za at slot n = 1. But that is invalid
771 * according to the parent's keys. Key2 must
772 * be inserted into znode zb.
774 * Note, this problem is not relevant for the
775 * case when we go right, because
776 * 'tnc_insert()' would correct the parent key.
778 if (*n == (*zn)->child_cnt - 1) {
779 err = tnc_next(c, zn, n);
781 /* Should be impossible */
787 ubifs_assert(*n == 0);
792 err = matches_name(c, &(*zn)->zbranch[*n], nm);
795 if (err == NAME_LESS)
797 if (err == NAME_MATCHES)
799 ubifs_assert(err == NAME_GREATER);
803 struct ubifs_znode *znode = *zn;
807 err = tnc_next(c, &znode, &nn);
812 if (keys_cmp(c, &znode->zbranch[nn].key, key))
814 err = matches_name(c, &znode->zbranch[nn], nm);
817 if (err == NAME_GREATER)
821 if (err == NAME_MATCHES)
823 ubifs_assert(err == NAME_LESS);
829 * fallible_matches_name - determine if a dent matches a given name.
830 * @c: UBIFS file-system description object
831 * @zbr: zbranch of dent
834 * This is a "fallible" version of 'matches_name()' function which does not
835 * panic if the direntry/xentry referred by @zbr does not exist on the media.
837 * This function checks if xentry/direntry referred by zbranch @zbr matches name
838 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
839 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
840 * if xentry/direntry referred by @zbr does not exist on the media. A negative
841 * error code is returned in case of failure.
843 static int fallible_matches_name(struct ubifs_info *c,
844 struct ubifs_zbranch *zbr,
845 const struct qstr *nm)
847 struct ubifs_dent_node *dent;
850 /* If possible, match against the dent in the leaf node cache */
852 dent = kmalloc(zbr->len, GFP_NOFS);
856 err = fallible_read_node(c, &zbr->key, zbr, dent);
860 /* The node was not present */
864 ubifs_assert(err == 1);
866 err = lnc_add_directly(c, zbr, dent);
872 nlen = le16_to_cpu(dent->nlen);
873 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
877 else if (nlen < nm->len)
892 * fallible_resolve_collision - resolve a collision even if nodes are missing.
893 * @c: UBIFS file-system description object
895 * @zn: znode is returned here
896 * @n: branch number is passed and returned here
897 * @nm: name of directory entry
898 * @adding: indicates caller is adding a key to the TNC
900 * This is a "fallible" version of the 'resolve_collision()' function which
901 * does not panic if one of the nodes referred to by TNC does not exist on the
902 * media. This may happen when replaying the journal if a deleted node was
903 * Garbage-collected and the commit was not done. A branch that refers to a node
904 * that is not present is called a dangling branch. The following are the return
905 * codes for this function:
906 * o if @nm was found, %1 is returned and @zn and @n are set to the found
908 * o if we are @adding and @nm was not found, %0 is returned;
909 * o if we are not @adding and @nm was not found, but a dangling branch was
910 * found, then %1 is returned and @zn and @n are set to the dangling branch;
911 * o a negative error code is returned in case of failure.
913 static int fallible_resolve_collision(struct ubifs_info *c,
914 const union ubifs_key *key,
915 struct ubifs_znode **zn, int *n,
916 const struct qstr *nm, int adding)
918 struct ubifs_znode *o_znode = NULL, *znode = *zn;
919 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
921 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
922 if (unlikely(cmp < 0))
924 if (cmp == NAME_MATCHES)
926 if (cmp == NOT_ON_MEDIA) {
930 * We are unlucky and hit a dangling branch straight away.
931 * Now we do not really know where to go to find the needed
932 * branch - to the left or to the right. Well, let's try left.
936 unsure = 1; /* Remove a dangling branch wherever it is */
938 if (cmp == NAME_GREATER || unsure) {
941 err = tnc_prev(c, zn, n);
942 if (err == -ENOENT) {
943 ubifs_assert(*n == 0);
949 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
950 /* See comments in 'resolve_collision()' */
951 if (*n == (*zn)->child_cnt - 1) {
952 err = tnc_next(c, zn, n);
954 /* Should be impossible */
960 ubifs_assert(*n == 0);
965 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
968 if (err == NAME_MATCHES)
970 if (err == NOT_ON_MEDIA) {
977 if (err == NAME_LESS)
984 if (cmp == NAME_LESS || unsure) {
989 err = tnc_next(c, &znode, &nn);
994 if (keys_cmp(c, &znode->zbranch[nn].key, key))
996 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
999 if (err == NAME_GREATER)
1003 if (err == NAME_MATCHES)
1005 if (err == NOT_ON_MEDIA) {
1012 /* Never match a dangling branch when adding */
1013 if (adding || !o_znode)
1016 dbg_mnt("dangling match LEB %d:%d len %d %s",
1017 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1018 o_znode->zbranch[o_n].len, DBGKEY(key));
1025 * matches_position - determine if a zbranch matches a given position.
1026 * @zbr: zbranch of dent
1027 * @lnum: LEB number of dent to match
1028 * @offs: offset of dent to match
1030 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1032 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1034 if (zbr->lnum == lnum && zbr->offs == offs)
1041 * resolve_collision_directly - resolve a collision directly.
1042 * @c: UBIFS file-system description object
1043 * @key: key of directory entry
1044 * @zn: znode is passed and returned here
1045 * @n: zbranch number is passed and returned here
1046 * @lnum: LEB number of dent node to match
1047 * @offs: offset of dent node to match
1049 * This function is used for "hashed" keys to make sure the found directory or
1050 * extended attribute entry node is what was looked for. It is used when the
1051 * flash address of the right node is known (@lnum:@offs) which makes it much
1052 * easier to resolve collisions (no need to read entries and match full
1053 * names). This function returns %1 and sets @zn and @n if the collision is
1054 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1055 * previous directory entry. Otherwise a negative error code is returned.
1057 static int resolve_collision_directly(struct ubifs_info *c,
1058 const union ubifs_key *key,
1059 struct ubifs_znode **zn, int *n,
1062 struct ubifs_znode *znode;
1067 if (matches_position(&znode->zbranch[nn], lnum, offs))
1072 err = tnc_prev(c, &znode, &nn);
1077 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1079 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1090 err = tnc_next(c, &znode, &nn);
1095 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1099 if (matches_position(&znode->zbranch[nn], lnum, offs))
1105 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1106 * @c: UBIFS file-system description object
1107 * @znode: znode to dirty
1109 * If we do not have a unique key that resides in a znode, then we cannot
1110 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1111 * This function records the path back to the last dirty ancestor, and then
1112 * dirties the znodes on that path.
1114 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1115 struct ubifs_znode *znode)
1117 struct ubifs_znode *zp;
1118 int *path = c->bottom_up_buf, p = 0;
1120 ubifs_assert(c->zroot.znode);
1121 ubifs_assert(znode);
1122 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1123 kfree(c->bottom_up_buf);
1124 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1126 if (!c->bottom_up_buf)
1127 return ERR_PTR(-ENOMEM);
1128 path = c->bottom_up_buf;
1130 if (c->zroot.znode->level) {
1131 /* Go up until parent is dirty */
1139 ubifs_assert(p < c->zroot.znode->level);
1141 if (!zp->cnext && ubifs_zn_dirty(znode))
1147 /* Come back down, dirtying as we go */
1149 struct ubifs_zbranch *zbr;
1153 ubifs_assert(path[p - 1] >= 0);
1154 ubifs_assert(path[p - 1] < zp->child_cnt);
1155 zbr = &zp->zbranch[path[--p]];
1156 znode = dirty_cow_znode(c, zbr);
1158 ubifs_assert(znode == c->zroot.znode);
1159 znode = dirty_cow_znode(c, &c->zroot);
1161 if (IS_ERR(znode) || !p)
1163 ubifs_assert(path[p - 1] >= 0);
1164 ubifs_assert(path[p - 1] < znode->child_cnt);
1165 znode = znode->zbranch[path[p - 1]].znode;
1172 * ubifs_lookup_level0 - search for zero-level znode.
1173 * @c: UBIFS file-system description object
1174 * @key: key to lookup
1175 * @zn: znode is returned here
1176 * @n: znode branch slot number is returned here
1178 * This function looks up the TNC tree and search for zero-level znode which
1179 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1181 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1182 * is returned and slot number of the matched branch is stored in @n;
1183 * o not exact match, which means that zero-level znode does not contain
1184 * @key, then %0 is returned and slot number of the closest branch is stored
1186 * o @key is so small that it is even less than the lowest key of the
1187 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1189 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1190 * function reads corresponding indexing nodes and inserts them to TNC. In
1191 * case of failure, a negative error code is returned.
1193 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1194 struct ubifs_znode **zn, int *n)
1197 struct ubifs_znode *znode;
1198 unsigned long time = get_seconds();
1200 dbg_tnc("search key %s", DBGKEY(key));
1201 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1203 znode = c->zroot.znode;
1204 if (unlikely(!znode)) {
1205 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1207 return PTR_ERR(znode);
1213 struct ubifs_zbranch *zbr;
1215 exact = ubifs_search_zbranch(c, znode, key, n);
1217 if (znode->level == 0)
1222 zbr = &znode->zbranch[*n];
1230 /* znode is not in TNC cache, load it from the media */
1231 znode = ubifs_load_znode(c, zbr, znode, *n);
1233 return PTR_ERR(znode);
1237 if (exact || !is_hash_key(c, key) || *n != -1) {
1238 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1243 * Here is a tricky place. We have not found the key and this is a
1244 * "hashed" key, which may collide. The rest of the code deals with
1245 * situations like this:
1249 * | 3 | 5 | | 6 | 7 | (x)
1251 * Or more a complex example:
1255 * | 1 | 3 | | 5 | 8 |
1257 * | 5 | 5 | | 6 | 7 | (x)
1259 * In the examples, if we are looking for key "5", we may reach nodes
1260 * marked with "(x)". In this case what we have do is to look at the
1261 * left and see if there is "5" key there. If there is, we have to
1264 * Note, this whole situation is possible because we allow to have
1265 * elements which are equivalent to the next key in the parent in the
1266 * children of current znode. For example, this happens if we split a
1267 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1271 * | 3 | 5 | | 5 | 6 | 7 |
1273 * And this becomes what is at the first "picture" after key "5" marked
1274 * with "^" is removed. What could be done is we could prohibit
1275 * splitting in the middle of the colliding sequence. Also, when
1276 * removing the leftmost key, we would have to correct the key of the
1277 * parent node, which would introduce additional complications. Namely,
1278 * if we changed the leftmost key of the parent znode, the garbage
1279 * collector would be unable to find it (GC is doing this when GC'ing
1280 * indexing LEBs). Although we already have an additional RB-tree where
1281 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1282 * after the commit. But anyway, this does not look easy to implement
1283 * so we did not try this.
1285 err = tnc_prev(c, &znode, n);
1286 if (err == -ENOENT) {
1287 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1291 if (unlikely(err < 0))
1293 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1294 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1299 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1305 * lookup_level0_dirty - search for zero-level znode dirtying.
1306 * @c: UBIFS file-system description object
1307 * @key: key to lookup
1308 * @zn: znode is returned here
1309 * @n: znode branch slot number is returned here
1311 * This function looks up the TNC tree and search for zero-level znode which
1312 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1314 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1315 * is returned and slot number of the matched branch is stored in @n;
1316 * o not exact match, which means that zero-level znode does not contain @key
1317 * then %0 is returned and slot number of the closed branch is stored in
1319 * o @key is so small that it is even less than the lowest key of the
1320 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1322 * Additionally all znodes in the path from the root to the located zero-level
1323 * znode are marked as dirty.
1325 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1326 * function reads corresponding indexing nodes and inserts them to TNC. In
1327 * case of failure, a negative error code is returned.
1329 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1330 struct ubifs_znode **zn, int *n)
1333 struct ubifs_znode *znode;
1334 unsigned long time = get_seconds();
1336 dbg_tnc("search and dirty key %s", DBGKEY(key));
1338 znode = c->zroot.znode;
1339 if (unlikely(!znode)) {
1340 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1342 return PTR_ERR(znode);
1345 znode = dirty_cow_znode(c, &c->zroot);
1347 return PTR_ERR(znode);
1352 struct ubifs_zbranch *zbr;
1354 exact = ubifs_search_zbranch(c, znode, key, n);
1356 if (znode->level == 0)
1361 zbr = &znode->zbranch[*n];
1365 znode = dirty_cow_znode(c, zbr);
1367 return PTR_ERR(znode);
1371 /* znode is not in TNC cache, load it from the media */
1372 znode = ubifs_load_znode(c, zbr, znode, *n);
1374 return PTR_ERR(znode);
1375 znode = dirty_cow_znode(c, zbr);
1377 return PTR_ERR(znode);
1381 if (exact || !is_hash_key(c, key) || *n != -1) {
1382 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1387 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1390 err = tnc_prev(c, &znode, n);
1391 if (err == -ENOENT) {
1393 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1396 if (unlikely(err < 0))
1398 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1400 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1404 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1405 znode = dirty_cow_bottom_up(c, znode);
1407 return PTR_ERR(znode);
1410 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1416 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1417 * @c: UBIFS file-system description object
1419 * @gc_seq1: garbage collection sequence number
1421 * This function determines if @lnum may have been garbage collected since
1422 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1425 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1427 int gc_seq2, gced_lnum;
1429 gced_lnum = c->gced_lnum;
1431 gc_seq2 = c->gc_seq;
1432 /* Same seq means no GC */
1433 if (gc_seq1 == gc_seq2)
1435 /* Different by more than 1 means we don't know */
1436 if (gc_seq1 + 1 != gc_seq2)
1439 * We have seen the sequence number has increased by 1. Now we need to
1440 * be sure we read the right LEB number, so read it again.
1443 if (gced_lnum != c->gced_lnum)
1445 /* Finally we can check lnum */
1446 if (gced_lnum == lnum)
1452 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1453 * @c: UBIFS file-system description object
1454 * @key: node key to lookup
1455 * @node: the node is returned here
1456 * @lnum: LEB number is returned here
1457 * @offs: offset is returned here
1459 * This function looks up and reads node with key @key. The caller has to make
1460 * sure the @node buffer is large enough to fit the node. Returns zero in case
1461 * of success, %-ENOENT if the node was not found, and a negative error code in
1462 * case of failure. The node location can be returned in @lnum and @offs.
1464 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1465 void *node, int *lnum, int *offs)
1467 int found, n, err, safely = 0, gc_seq1;
1468 struct ubifs_znode *znode;
1469 struct ubifs_zbranch zbr, *zt;
1472 mutex_lock(&c->tnc_mutex);
1473 found = ubifs_lookup_level0(c, key, &znode, &n);
1477 } else if (found < 0) {
1481 zt = &znode->zbranch[n];
1486 if (is_hash_key(c, key)) {
1488 * In this case the leaf node cache gets used, so we pass the
1489 * address of the zbranch and keep the mutex locked
1491 err = tnc_read_node_nm(c, zt, node);
1495 err = ubifs_tnc_read_node(c, zt, node);
1498 /* Drop the TNC mutex prematurely and race with garbage collection */
1499 zbr = znode->zbranch[n];
1500 gc_seq1 = c->gc_seq;
1501 mutex_unlock(&c->tnc_mutex);
1503 if (ubifs_get_wbuf(c, zbr.lnum)) {
1504 /* We do not GC journal heads */
1505 err = ubifs_tnc_read_node(c, &zbr, node);
1509 err = fallible_read_node(c, key, &zbr, node);
1510 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1512 * The node may have been GC'ed out from under us so try again
1513 * while keeping the TNC mutex locked.
1521 mutex_unlock(&c->tnc_mutex);
1526 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1527 * @c: UBIFS file-system description object
1528 * @bu: bulk-read parameters and results
1530 * Lookup consecutive data node keys for the same inode that reside
1531 * consecutively in the same LEB. This function returns zero in case of success
1532 * and a negative error code in case of failure.
1534 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1535 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1536 * maximum possible amount of nodes for bulk-read.
1538 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1540 int n, err = 0, lnum = -1, uninitialized_var(offs);
1541 int uninitialized_var(len);
1542 unsigned int block = key_block(c, &bu->key);
1543 struct ubifs_znode *znode;
1549 mutex_lock(&c->tnc_mutex);
1550 /* Find first key */
1551 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1556 len = znode->zbranch[n].len;
1557 /* The buffer must be big enough for at least 1 node */
1558 if (len > bu->buf_len) {
1563 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1565 lnum = znode->zbranch[n].lnum;
1566 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1569 struct ubifs_zbranch *zbr;
1570 union ubifs_key *key;
1571 unsigned int next_block;
1574 err = tnc_next(c, &znode, &n);
1577 zbr = &znode->zbranch[n];
1579 /* See if there is another data key for this file */
1580 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1581 key_type(c, key) != UBIFS_DATA_KEY) {
1586 /* First key found */
1588 offs = ALIGN(zbr->offs + zbr->len, 8);
1590 if (len > bu->buf_len) {
1596 * The data nodes must be in consecutive positions in
1599 if (zbr->lnum != lnum || zbr->offs != offs)
1601 offs += ALIGN(zbr->len, 8);
1602 len = ALIGN(len, 8) + zbr->len;
1603 /* Must not exceed buffer length */
1604 if (len > bu->buf_len)
1607 /* Allow for holes */
1608 next_block = key_block(c, key);
1609 bu->blk_cnt += (next_block - block - 1);
1610 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1614 bu->zbranch[bu->cnt++] = *zbr;
1616 /* See if we have room for more */
1617 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1619 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1623 if (err == -ENOENT) {
1627 bu->gc_seq = c->gc_seq;
1628 mutex_unlock(&c->tnc_mutex);
1632 * An enormous hole could cause bulk-read to encompass too many
1633 * page cache pages, so limit the number here.
1635 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1636 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1638 * Ensure that bulk-read covers a whole number of page cache
1641 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1642 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1645 /* At the end of file we can round up */
1646 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1649 /* Exclude data nodes that do not make up a whole page cache page */
1650 block = key_block(c, &bu->key) + bu->blk_cnt;
1651 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1653 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1661 * read_wbuf - bulk-read from a LEB with a wbuf.
1662 * @wbuf: wbuf that may overlap the read
1663 * @buf: buffer into which to read
1665 * @lnum: LEB number from which to read
1666 * @offs: offset from which to read
1668 * This functions returns %0 on success or a negative error code on failure.
1670 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1673 const struct ubifs_info *c = wbuf->c;
1676 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1677 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1678 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1679 ubifs_assert(offs + len <= c->leb_size);
1681 spin_lock(&wbuf->lock);
1682 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1684 /* We may safely unlock the write-buffer and read the data */
1685 spin_unlock(&wbuf->lock);
1686 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1689 /* Don't read under wbuf */
1690 rlen = wbuf->offs - offs;
1694 /* Copy the rest from the write-buffer */
1695 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1696 spin_unlock(&wbuf->lock);
1699 /* Read everything that goes before write-buffer */
1700 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1706 * validate_data_node - validate data nodes for bulk-read.
1707 * @c: UBIFS file-system description object
1708 * @buf: buffer containing data node to validate
1709 * @zbr: zbranch of data node to validate
1711 * This functions returns %0 on success or a negative error code on failure.
1713 static int validate_data_node(struct ubifs_info *c, void *buf,
1714 struct ubifs_zbranch *zbr)
1716 union ubifs_key key1;
1717 struct ubifs_ch *ch = buf;
1720 if (ch->node_type != UBIFS_DATA_NODE) {
1721 ubifs_err("bad node type (%d but expected %d)",
1722 ch->node_type, UBIFS_DATA_NODE);
1726 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1728 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1732 len = le32_to_cpu(ch->len);
1733 if (len != zbr->len) {
1734 ubifs_err("bad node length %d, expected %d", len, zbr->len);
1738 /* Make sure the key of the read node is correct */
1739 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1740 if (!keys_eq(c, &zbr->key, &key1)) {
1741 ubifs_err("bad key in node at LEB %d:%d",
1742 zbr->lnum, zbr->offs);
1743 dbg_tnc("looked for key %s found node's key %s",
1744 DBGKEY(&zbr->key), DBGKEY1(&key1));
1753 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1754 dbg_dump_node(c, buf);
1760 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1761 * @c: UBIFS file-system description object
1762 * @bu: bulk-read parameters and results
1764 * This functions reads and validates the data nodes that were identified by the
1765 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1766 * -EAGAIN to indicate a race with GC, or another negative error code on
1769 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1771 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1772 struct ubifs_wbuf *wbuf;
1775 len = bu->zbranch[bu->cnt - 1].offs;
1776 len += bu->zbranch[bu->cnt - 1].len - offs;
1777 if (len > bu->buf_len) {
1778 ubifs_err("buffer too small %d vs %d", bu->buf_len, len);
1783 wbuf = ubifs_get_wbuf(c, lnum);
1785 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1787 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1789 /* Check for a race with GC */
1790 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1793 if (err && err != -EBADMSG) {
1794 ubifs_err("failed to read from LEB %d:%d, error %d",
1797 dbg_tnc("key %s", DBGKEY(&bu->key));
1801 /* Validate the nodes read */
1803 for (i = 0; i < bu->cnt; i++) {
1804 err = validate_data_node(c, buf, &bu->zbranch[i]);
1807 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1814 * do_lookup_nm- look up a "hashed" node.
1815 * @c: UBIFS file-system description object
1816 * @key: node key to lookup
1817 * @node: the node is returned here
1820 * This function look up and reads a node which contains name hash in the key.
1821 * Since the hash may have collisions, there may be many nodes with the same
1822 * key, so we have to sequentially look to all of them until the needed one is
1823 * found. This function returns zero in case of success, %-ENOENT if the node
1824 * was not found, and a negative error code in case of failure.
1826 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1827 void *node, const struct qstr *nm)
1830 struct ubifs_znode *znode;
1832 dbg_tnc("name '%.*s' key %s", nm->len, nm->name, DBGKEY(key));
1833 mutex_lock(&c->tnc_mutex);
1834 found = ubifs_lookup_level0(c, key, &znode, &n);
1838 } else if (found < 0) {
1843 ubifs_assert(n >= 0);
1845 err = resolve_collision(c, key, &znode, &n, nm);
1846 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1847 if (unlikely(err < 0))
1854 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
1857 mutex_unlock(&c->tnc_mutex);
1862 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1863 * @c: UBIFS file-system description object
1864 * @key: node key to lookup
1865 * @node: the node is returned here
1868 * This function look up and reads a node which contains name hash in the key.
1869 * Since the hash may have collisions, there may be many nodes with the same
1870 * key, so we have to sequentially look to all of them until the needed one is
1871 * found. This function returns zero in case of success, %-ENOENT if the node
1872 * was not found, and a negative error code in case of failure.
1874 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1875 void *node, const struct qstr *nm)
1878 const struct ubifs_dent_node *dent = node;
1881 * We assume that in most of the cases there are no name collisions and
1882 * 'ubifs_tnc_lookup()' returns us the right direntry.
1884 err = ubifs_tnc_lookup(c, key, node);
1888 len = le16_to_cpu(dent->nlen);
1889 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1893 * Unluckily, there are hash collisions and we have to iterate over
1894 * them look at each direntry with colliding name hash sequentially.
1896 return do_lookup_nm(c, key, node, nm);
1900 * correct_parent_keys - correct parent znodes' keys.
1901 * @c: UBIFS file-system description object
1902 * @znode: znode to correct parent znodes for
1904 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1905 * zbranch changes, keys of parent znodes have to be corrected. This helper
1906 * function is called in such situations and corrects the keys if needed.
1908 static void correct_parent_keys(const struct ubifs_info *c,
1909 struct ubifs_znode *znode)
1911 union ubifs_key *key, *key1;
1913 ubifs_assert(znode->parent);
1914 ubifs_assert(znode->iip == 0);
1916 key = &znode->zbranch[0].key;
1917 key1 = &znode->parent->zbranch[0].key;
1919 while (keys_cmp(c, key, key1) < 0) {
1920 key_copy(c, key, key1);
1921 znode = znode->parent;
1923 if (!znode->parent || znode->iip)
1925 key1 = &znode->parent->zbranch[0].key;
1930 * insert_zbranch - insert a zbranch into a znode.
1931 * @znode: znode into which to insert
1932 * @zbr: zbranch to insert
1933 * @n: slot number to insert to
1935 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1936 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1937 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1938 * slot, zbranches starting from @n have to be moved right.
1940 static void insert_zbranch(struct ubifs_znode *znode,
1941 const struct ubifs_zbranch *zbr, int n)
1945 ubifs_assert(ubifs_zn_dirty(znode));
1948 for (i = znode->child_cnt; i > n; i--) {
1949 znode->zbranch[i] = znode->zbranch[i - 1];
1950 if (znode->zbranch[i].znode)
1951 znode->zbranch[i].znode->iip = i;
1954 zbr->znode->iip = n;
1956 for (i = znode->child_cnt; i > n; i--)
1957 znode->zbranch[i] = znode->zbranch[i - 1];
1959 znode->zbranch[n] = *zbr;
1960 znode->child_cnt += 1;
1963 * After inserting at slot zero, the lower bound of the key range of
1964 * this znode may have changed. If this znode is subsequently split
1965 * then the upper bound of the key range may change, and furthermore
1966 * it could change to be lower than the original lower bound. If that
1967 * happens, then it will no longer be possible to find this znode in the
1968 * TNC using the key from the index node on flash. That is bad because
1969 * if it is not found, we will assume it is obsolete and may overwrite
1970 * it. Then if there is an unclean unmount, we will start using the
1971 * old index which will be broken.
1973 * So we first mark znodes that have insertions at slot zero, and then
1974 * if they are split we add their lnum/offs to the old_idx tree.
1981 * tnc_insert - insert a node into TNC.
1982 * @c: UBIFS file-system description object
1983 * @znode: znode to insert into
1984 * @zbr: branch to insert
1985 * @n: slot number to insert new zbranch to
1987 * This function inserts a new node described by @zbr into znode @znode. If
1988 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1989 * are splat as well if needed. Returns zero in case of success or a negative
1990 * error code in case of failure.
1992 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1993 struct ubifs_zbranch *zbr, int n)
1995 struct ubifs_znode *zn, *zi, *zp;
1996 int i, keep, move, appending = 0;
1997 union ubifs_key *key = &zbr->key, *key1;
1999 ubifs_assert(n >= 0 && n <= c->fanout);
2001 /* Implement naive insert for now */
2004 if (znode->child_cnt < c->fanout) {
2005 ubifs_assert(n != c->fanout);
2006 dbg_tnc("inserted at %d level %d, key %s", n, znode->level,
2009 insert_zbranch(znode, zbr, n);
2011 /* Ensure parent's key is correct */
2012 if (n == 0 && zp && znode->iip == 0)
2013 correct_parent_keys(c, znode);
2019 * Unfortunately, @znode does not have more empty slots and we have to
2022 dbg_tnc("splitting level %d, key %s", znode->level, DBGKEY(key));
2026 * We can no longer be sure of finding this znode by key, so we
2027 * record it in the old_idx tree.
2029 ins_clr_old_idx_znode(c, znode);
2031 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2035 zn->level = znode->level;
2037 /* Decide where to split */
2038 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2039 /* Try not to split consecutive data keys */
2040 if (n == c->fanout) {
2041 key1 = &znode->zbranch[n - 1].key;
2042 if (key_inum(c, key1) == key_inum(c, key) &&
2043 key_type(c, key1) == UBIFS_DATA_KEY)
2047 } else if (appending && n != c->fanout) {
2048 /* Try not to split consecutive data keys */
2051 if (n >= (c->fanout + 1) / 2) {
2052 key1 = &znode->zbranch[0].key;
2053 if (key_inum(c, key1) == key_inum(c, key) &&
2054 key_type(c, key1) == UBIFS_DATA_KEY) {
2055 key1 = &znode->zbranch[n].key;
2056 if (key_inum(c, key1) != key_inum(c, key) ||
2057 key_type(c, key1) != UBIFS_DATA_KEY) {
2059 move = c->fanout - keep;
2071 keep = (c->fanout + 1) / 2;
2072 move = c->fanout - keep;
2076 * Although we don't at present, we could look at the neighbors and see
2077 * if we can move some zbranches there.
2081 /* Insert into existing znode */
2086 /* Insert into new znode */
2091 zbr->znode->parent = zn;
2096 __set_bit(DIRTY_ZNODE, &zn->flags);
2097 atomic_long_inc(&c->dirty_zn_cnt);
2099 zn->child_cnt = move;
2100 znode->child_cnt = keep;
2102 dbg_tnc("moving %d, keeping %d", move, keep);
2105 for (i = 0; i < move; i++) {
2106 zn->zbranch[i] = znode->zbranch[keep + i];
2109 if (zn->zbranch[i].znode) {
2110 zn->zbranch[i].znode->parent = zn;
2111 zn->zbranch[i].znode->iip = i;
2115 /* Insert new key and branch */
2116 dbg_tnc("inserting at %d level %d, key %s", n, zn->level, DBGKEY(key));
2118 insert_zbranch(zi, zbr, n);
2120 /* Insert new znode (produced by spitting) into the parent */
2122 if (n == 0 && zi == znode && znode->iip == 0)
2123 correct_parent_keys(c, znode);
2125 /* Locate insertion point */
2128 /* Tail recursion */
2129 zbr->key = zn->zbranch[0].key;
2139 /* We have to split root znode */
2140 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2142 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2147 zi->level = znode->level + 1;
2149 __set_bit(DIRTY_ZNODE, &zi->flags);
2150 atomic_long_inc(&c->dirty_zn_cnt);
2152 zi->zbranch[0].key = znode->zbranch[0].key;
2153 zi->zbranch[0].znode = znode;
2154 zi->zbranch[0].lnum = c->zroot.lnum;
2155 zi->zbranch[0].offs = c->zroot.offs;
2156 zi->zbranch[0].len = c->zroot.len;
2157 zi->zbranch[1].key = zn->zbranch[0].key;
2158 zi->zbranch[1].znode = zn;
2163 c->zroot.znode = zi;
2174 * ubifs_tnc_add - add a node to TNC.
2175 * @c: UBIFS file-system description object
2177 * @lnum: LEB number of node
2178 * @offs: node offset
2181 * This function adds a node with key @key to TNC. The node may be new or it may
2182 * obsolete some existing one. Returns %0 on success or negative error code on
2185 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2188 int found, n, err = 0;
2189 struct ubifs_znode *znode;
2191 mutex_lock(&c->tnc_mutex);
2192 dbg_tnc("%d:%d, len %d, key %s", lnum, offs, len, DBGKEY(key));
2193 found = lookup_level0_dirty(c, key, &znode, &n);
2195 struct ubifs_zbranch zbr;
2201 key_copy(c, key, &zbr.key);
2202 err = tnc_insert(c, znode, &zbr, n + 1);
2203 } else if (found == 1) {
2204 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2207 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2214 err = dbg_check_tnc(c, 0);
2215 mutex_unlock(&c->tnc_mutex);
2221 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2222 * @c: UBIFS file-system description object
2224 * @old_lnum: LEB number of old node
2225 * @old_offs: old node offset
2226 * @lnum: LEB number of node
2227 * @offs: node offset
2230 * This function replaces a node with key @key in the TNC only if the old node
2231 * is found. This function is called by garbage collection when node are moved.
2232 * Returns %0 on success or negative error code on failure.
2234 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2235 int old_lnum, int old_offs, int lnum, int offs, int len)
2237 int found, n, err = 0;
2238 struct ubifs_znode *znode;
2240 mutex_lock(&c->tnc_mutex);
2241 dbg_tnc("old LEB %d:%d, new LEB %d:%d, len %d, key %s", old_lnum,
2242 old_offs, lnum, offs, len, DBGKEY(key));
2243 found = lookup_level0_dirty(c, key, &znode, &n);
2250 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2253 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2255 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2262 } else if (is_hash_key(c, key)) {
2263 found = resolve_collision_directly(c, key, &znode, &n,
2264 old_lnum, old_offs);
2265 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2266 found, znode, n, old_lnum, old_offs);
2273 /* Ensure the znode is dirtied */
2274 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2275 znode = dirty_cow_bottom_up(c, znode);
2276 if (IS_ERR(znode)) {
2277 err = PTR_ERR(znode);
2281 zbr = &znode->zbranch[n];
2283 err = ubifs_add_dirt(c, zbr->lnum,
2295 err = ubifs_add_dirt(c, lnum, len);
2298 err = dbg_check_tnc(c, 0);
2301 mutex_unlock(&c->tnc_mutex);
2306 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2307 * @c: UBIFS file-system description object
2309 * @lnum: LEB number of node
2310 * @offs: node offset
2314 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2315 * may have collisions, like directory entry keys.
2317 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2318 int lnum, int offs, int len, const struct qstr *nm)
2320 int found, n, err = 0;
2321 struct ubifs_znode *znode;
2323 mutex_lock(&c->tnc_mutex);
2324 dbg_tnc("LEB %d:%d, name '%.*s', key %s", lnum, offs, nm->len, nm->name,
2326 found = lookup_level0_dirty(c, key, &znode, &n);
2334 found = fallible_resolve_collision(c, key, &znode, &n,
2337 found = resolve_collision(c, key, &znode, &n, nm);
2338 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2344 /* Ensure the znode is dirtied */
2345 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2346 znode = dirty_cow_bottom_up(c, znode);
2347 if (IS_ERR(znode)) {
2348 err = PTR_ERR(znode);
2354 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2357 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2366 struct ubifs_zbranch zbr;
2372 key_copy(c, key, &zbr.key);
2373 err = tnc_insert(c, znode, &zbr, n + 1);
2378 * We did not find it in the index so there may be a
2379 * dangling branch still in the index. So we remove it
2380 * by passing 'ubifs_tnc_remove_nm()' the same key but
2381 * an unmatchable name.
2383 struct qstr noname = { .len = 0, .name = "" };
2385 err = dbg_check_tnc(c, 0);
2386 mutex_unlock(&c->tnc_mutex);
2389 return ubifs_tnc_remove_nm(c, key, &noname);
2395 err = dbg_check_tnc(c, 0);
2396 mutex_unlock(&c->tnc_mutex);
2401 * tnc_delete - delete a znode form TNC.
2402 * @c: UBIFS file-system description object
2403 * @znode: znode to delete from
2404 * @n: zbranch slot number to delete
2406 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2407 * case of success and a negative error code in case of failure.
2409 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2411 struct ubifs_zbranch *zbr;
2412 struct ubifs_znode *zp;
2415 /* Delete without merge for now */
2416 ubifs_assert(znode->level == 0);
2417 ubifs_assert(n >= 0 && n < c->fanout);
2418 dbg_tnc("deleting %s", DBGKEY(&znode->zbranch[n].key));
2420 zbr = &znode->zbranch[n];
2423 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2425 dbg_dump_znode(c, znode);
2429 /* We do not "gap" zbranch slots */
2430 for (i = n; i < znode->child_cnt - 1; i++)
2431 znode->zbranch[i] = znode->zbranch[i + 1];
2432 znode->child_cnt -= 1;
2434 if (znode->child_cnt > 0)
2438 * This was the last zbranch, we have to delete this znode from the
2443 ubifs_assert(!ubifs_zn_obsolete(znode));
2444 ubifs_assert(ubifs_zn_dirty(znode));
2449 atomic_long_dec(&c->dirty_zn_cnt);
2451 err = insert_old_idx_znode(c, znode);
2456 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2457 atomic_long_inc(&c->clean_zn_cnt);
2458 atomic_long_inc(&ubifs_clean_zn_cnt);
2462 } while (znode->child_cnt == 1); /* while removing last child */
2464 /* Remove from znode, entry n - 1 */
2465 znode->child_cnt -= 1;
2466 ubifs_assert(znode->level != 0);
2467 for (i = n; i < znode->child_cnt; i++) {
2468 znode->zbranch[i] = znode->zbranch[i + 1];
2469 if (znode->zbranch[i].znode)
2470 znode->zbranch[i].znode->iip = i;
2474 * If this is the root and it has only 1 child then
2475 * collapse the tree.
2477 if (!znode->parent) {
2478 while (znode->child_cnt == 1 && znode->level != 0) {
2480 zbr = &znode->zbranch[0];
2481 znode = get_znode(c, znode, 0);
2483 return PTR_ERR(znode);
2484 znode = dirty_cow_znode(c, zbr);
2486 return PTR_ERR(znode);
2487 znode->parent = NULL;
2490 err = insert_old_idx(c, c->zroot.lnum,
2495 c->zroot.lnum = zbr->lnum;
2496 c->zroot.offs = zbr->offs;
2497 c->zroot.len = zbr->len;
2498 c->zroot.znode = znode;
2499 ubifs_assert(!ubifs_zn_obsolete(zp));
2500 ubifs_assert(ubifs_zn_dirty(zp));
2501 atomic_long_dec(&c->dirty_zn_cnt);
2504 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2505 atomic_long_inc(&c->clean_zn_cnt);
2506 atomic_long_inc(&ubifs_clean_zn_cnt);
2516 * ubifs_tnc_remove - remove an index entry of a node.
2517 * @c: UBIFS file-system description object
2520 * Returns %0 on success or negative error code on failure.
2522 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2524 int found, n, err = 0;
2525 struct ubifs_znode *znode;
2527 mutex_lock(&c->tnc_mutex);
2528 dbg_tnc("key %s", DBGKEY(key));
2529 found = lookup_level0_dirty(c, key, &znode, &n);
2535 err = tnc_delete(c, znode, n);
2537 err = dbg_check_tnc(c, 0);
2540 mutex_unlock(&c->tnc_mutex);
2545 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2546 * @c: UBIFS file-system description object
2548 * @nm: directory entry name
2550 * Returns %0 on success or negative error code on failure.
2552 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2553 const struct qstr *nm)
2556 struct ubifs_znode *znode;
2558 mutex_lock(&c->tnc_mutex);
2559 dbg_tnc("%.*s, key %s", nm->len, nm->name, DBGKEY(key));
2560 err = lookup_level0_dirty(c, key, &znode, &n);
2566 err = fallible_resolve_collision(c, key, &znode, &n,
2569 err = resolve_collision(c, key, &znode, &n, nm);
2570 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2574 /* Ensure the znode is dirtied */
2575 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2576 znode = dirty_cow_bottom_up(c, znode);
2577 if (IS_ERR(znode)) {
2578 err = PTR_ERR(znode);
2582 err = tnc_delete(c, znode, n);
2588 err = dbg_check_tnc(c, 0);
2589 mutex_unlock(&c->tnc_mutex);
2594 * key_in_range - determine if a key falls within a range of keys.
2595 * @c: UBIFS file-system description object
2596 * @key: key to check
2597 * @from_key: lowest key in range
2598 * @to_key: highest key in range
2600 * This function returns %1 if the key is in range and %0 otherwise.
2602 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2603 union ubifs_key *from_key, union ubifs_key *to_key)
2605 if (keys_cmp(c, key, from_key) < 0)
2607 if (keys_cmp(c, key, to_key) > 0)
2613 * ubifs_tnc_remove_range - remove index entries in range.
2614 * @c: UBIFS file-system description object
2615 * @from_key: lowest key to remove
2616 * @to_key: highest key to remove
2618 * This function removes index entries starting at @from_key and ending at
2619 * @to_key. This function returns zero in case of success and a negative error
2620 * code in case of failure.
2622 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2623 union ubifs_key *to_key)
2625 int i, n, k, err = 0;
2626 struct ubifs_znode *znode;
2627 union ubifs_key *key;
2629 mutex_lock(&c->tnc_mutex);
2631 /* Find first level 0 znode that contains keys to remove */
2632 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2639 err = tnc_next(c, &znode, &n);
2640 if (err == -ENOENT) {
2646 key = &znode->zbranch[n].key;
2647 if (!key_in_range(c, key, from_key, to_key)) {
2653 /* Ensure the znode is dirtied */
2654 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2655 znode = dirty_cow_bottom_up(c, znode);
2656 if (IS_ERR(znode)) {
2657 err = PTR_ERR(znode);
2662 /* Remove all keys in range except the first */
2663 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2664 key = &znode->zbranch[i].key;
2665 if (!key_in_range(c, key, from_key, to_key))
2667 lnc_free(&znode->zbranch[i]);
2668 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2669 znode->zbranch[i].len);
2671 dbg_dump_znode(c, znode);
2674 dbg_tnc("removing %s", DBGKEY(key));
2677 for (i = n + 1 + k; i < znode->child_cnt; i++)
2678 znode->zbranch[i - k] = znode->zbranch[i];
2679 znode->child_cnt -= k;
2682 /* Now delete the first */
2683 err = tnc_delete(c, znode, n);
2690 err = dbg_check_tnc(c, 0);
2691 mutex_unlock(&c->tnc_mutex);
2696 * ubifs_tnc_remove_ino - remove an inode from TNC.
2697 * @c: UBIFS file-system description object
2698 * @inum: inode number to remove
2700 * This function remove inode @inum and all the extended attributes associated
2701 * with the anode from TNC and returns zero in case of success or a negative
2702 * error code in case of failure.
2704 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2706 union ubifs_key key1, key2;
2707 struct ubifs_dent_node *xent, *pxent = NULL;
2708 struct qstr nm = { .name = NULL };
2710 dbg_tnc("ino %lu", (unsigned long)inum);
2713 * Walk all extended attribute entries and remove them together with
2714 * corresponding extended attribute inodes.
2716 lowest_xent_key(c, &key1, inum);
2721 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2723 err = PTR_ERR(xent);
2729 xattr_inum = le64_to_cpu(xent->inum);
2730 dbg_tnc("xent '%s', ino %lu", xent->name,
2731 (unsigned long)xattr_inum);
2733 #ifdef CONFIG_UBIFS_FS_XATTR
2734 ubifs_evict_xattr_inode(c, xattr_inum);
2737 nm.name = xent->name;
2738 nm.len = le16_to_cpu(xent->nlen);
2739 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2745 lowest_ino_key(c, &key1, xattr_inum);
2746 highest_ino_key(c, &key2, xattr_inum);
2747 err = ubifs_tnc_remove_range(c, &key1, &key2);
2755 key_read(c, &xent->key, &key1);
2759 lowest_ino_key(c, &key1, inum);
2760 highest_ino_key(c, &key2, inum);
2762 return ubifs_tnc_remove_range(c, &key1, &key2);
2766 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2767 * @c: UBIFS file-system description object
2768 * @key: key of last entry
2769 * @nm: name of last entry found or %NULL
2771 * This function finds and reads the next directory or extended attribute entry
2772 * after the given key (@key) if there is one. @nm is used to resolve
2775 * If the name of the current entry is not known and only the key is known,
2776 * @nm->name has to be %NULL. In this case the semantics of this function is a
2777 * little bit different and it returns the entry corresponding to this key, not
2778 * the next one. If the key was not found, the closest "right" entry is
2781 * If the fist entry has to be found, @key has to contain the lowest possible
2782 * key value for this inode and @name has to be %NULL.
2784 * This function returns the found directory or extended attribute entry node
2785 * in case of success, %-ENOENT is returned if no entry was found, and a
2786 * negative error code is returned in case of failure.
2788 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2789 union ubifs_key *key,
2790 const struct qstr *nm)
2792 int n, err, type = key_type(c, key);
2793 struct ubifs_znode *znode;
2794 struct ubifs_dent_node *dent;
2795 struct ubifs_zbranch *zbr;
2796 union ubifs_key *dkey;
2798 dbg_tnc("%s %s", nm->name ? (char *)nm->name : "(lowest)", DBGKEY(key));
2799 ubifs_assert(is_hash_key(c, key));
2801 mutex_lock(&c->tnc_mutex);
2802 err = ubifs_lookup_level0(c, key, &znode, &n);
2803 if (unlikely(err < 0))
2808 /* Handle collisions */
2810 err = fallible_resolve_collision(c, key, &znode, &n,
2813 err = resolve_collision(c, key, &znode, &n, nm);
2814 dbg_tnc("rc returned %d, znode %p, n %d",
2816 if (unlikely(err < 0))
2820 /* Now find next entry */
2821 err = tnc_next(c, &znode, &n);
2826 * The full name of the entry was not given, in which case the
2827 * behavior of this function is a little different and it
2828 * returns current entry, not the next one.
2832 * However, the given key does not exist in the TNC
2833 * tree and @znode/@n variables contain the closest
2834 * "preceding" element. Switch to the next one.
2836 err = tnc_next(c, &znode, &n);
2842 zbr = &znode->zbranch[n];
2843 dent = kmalloc(zbr->len, GFP_NOFS);
2844 if (unlikely(!dent)) {
2850 * The above 'tnc_next()' call could lead us to the next inode, check
2854 if (key_inum(c, dkey) != key_inum(c, key) ||
2855 key_type(c, dkey) != type) {
2860 err = tnc_read_node_nm(c, zbr, dent);
2864 mutex_unlock(&c->tnc_mutex);
2870 mutex_unlock(&c->tnc_mutex);
2871 return ERR_PTR(err);
2875 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2876 * @c: UBIFS file-system description object
2878 * Destroy left-over obsolete znodes from a failed commit.
2880 static void tnc_destroy_cnext(struct ubifs_info *c)
2882 struct ubifs_znode *cnext;
2886 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2889 struct ubifs_znode *znode = cnext;
2891 cnext = cnext->cnext;
2892 if (ubifs_zn_obsolete(znode))
2894 } while (cnext && cnext != c->cnext);
2898 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2899 * @c: UBIFS file-system description object
2901 void ubifs_tnc_close(struct ubifs_info *c)
2903 tnc_destroy_cnext(c);
2904 if (c->zroot.znode) {
2907 ubifs_destroy_tnc_subtree(c->zroot.znode);
2908 n = atomic_long_read(&c->clean_zn_cnt);
2909 atomic_long_sub(n, &ubifs_clean_zn_cnt);
2917 * left_znode - get the znode to the left.
2918 * @c: UBIFS file-system description object
2921 * This function returns a pointer to the znode to the left of @znode or NULL if
2922 * there is not one. A negative error code is returned on failure.
2924 static struct ubifs_znode *left_znode(struct ubifs_info *c,
2925 struct ubifs_znode *znode)
2927 int level = znode->level;
2930 int n = znode->iip - 1;
2932 /* Go up until we can go left */
2933 znode = znode->parent;
2937 /* Now go down the rightmost branch to 'level' */
2938 znode = get_znode(c, znode, n);
2941 while (znode->level != level) {
2942 n = znode->child_cnt - 1;
2943 znode = get_znode(c, znode, n);
2954 * right_znode - get the znode to the right.
2955 * @c: UBIFS file-system description object
2958 * This function returns a pointer to the znode to the right of @znode or NULL
2959 * if there is not one. A negative error code is returned on failure.
2961 static struct ubifs_znode *right_znode(struct ubifs_info *c,
2962 struct ubifs_znode *znode)
2964 int level = znode->level;
2967 int n = znode->iip + 1;
2969 /* Go up until we can go right */
2970 znode = znode->parent;
2973 if (n < znode->child_cnt) {
2974 /* Now go down the leftmost branch to 'level' */
2975 znode = get_znode(c, znode, n);
2978 while (znode->level != level) {
2979 znode = get_znode(c, znode, 0);
2990 * lookup_znode - find a particular indexing node from TNC.
2991 * @c: UBIFS file-system description object
2992 * @key: index node key to lookup
2993 * @level: index node level
2994 * @lnum: index node LEB number
2995 * @offs: index node offset
2997 * This function searches an indexing node by its first key @key and its
2998 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2999 * nodes it traverses to TNC. This function is called for indexing nodes which
3000 * were found on the media by scanning, for example when garbage-collecting or
3001 * when doing in-the-gaps commit. This means that the indexing node which is
3002 * looked for does not have to have exactly the same leftmost key @key, because
3003 * the leftmost key may have been changed, in which case TNC will contain a
3004 * dirty znode which still refers the same @lnum:@offs. This function is clever
3005 * enough to recognize such indexing nodes.
3007 * Note, if a znode was deleted or changed too much, then this function will
3008 * not find it. For situations like this UBIFS has the old index RB-tree
3009 * (indexed by @lnum:@offs).
3011 * This function returns a pointer to the znode found or %NULL if it is not
3012 * found. A negative error code is returned on failure.
3014 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3015 union ubifs_key *key, int level,
3018 struct ubifs_znode *znode, *zn;
3021 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
3024 * The arguments have probably been read off flash, so don't assume
3028 return ERR_PTR(-EINVAL);
3030 /* Get the root znode */
3031 znode = c->zroot.znode;
3033 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3037 /* Check if it is the one we are looking for */
3038 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3040 /* Descend to the parent level i.e. (level + 1) */
3041 if (level >= znode->level)
3044 ubifs_search_zbranch(c, znode, key, &n);
3047 * We reached a znode where the leftmost key is greater
3048 * than the key we are searching for. This is the same
3049 * situation as the one described in a huge comment at
3050 * the end of the 'ubifs_lookup_level0()' function. And
3051 * for exactly the same reasons we have to try to look
3052 * left before giving up.
3054 znode = left_znode(c, znode);
3059 ubifs_search_zbranch(c, znode, key, &n);
3060 ubifs_assert(n >= 0);
3062 if (znode->level == level + 1)
3064 znode = get_znode(c, znode, n);
3068 /* Check if the child is the one we are looking for */
3069 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3070 return get_znode(c, znode, n);
3071 /* If the key is unique, there is nowhere else to look */
3072 if (!is_hash_key(c, key))
3075 * The key is not unique and so may be also in the znodes to either
3082 /* Move one branch to the left */
3086 znode = left_znode(c, znode);
3091 n = znode->child_cnt - 1;
3094 if (znode->zbranch[n].lnum == lnum &&
3095 znode->zbranch[n].offs == offs)
3096 return get_znode(c, znode, n);
3097 /* Stop if the key is less than the one we are looking for */
3098 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3101 /* Back to the middle */
3106 /* Move one branch to the right */
3107 if (++n >= znode->child_cnt) {
3108 znode = right_znode(c, znode);
3116 if (znode->zbranch[n].lnum == lnum &&
3117 znode->zbranch[n].offs == offs)
3118 return get_znode(c, znode, n);
3119 /* Stop if the key is greater than the one we are looking for */
3120 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3127 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3128 * @c: UBIFS file-system description object
3129 * @key: key of index node
3130 * @level: index node level
3131 * @lnum: LEB number of index node
3132 * @offs: offset of index node
3134 * This function returns %0 if the index node is not referred to in the TNC, %1
3135 * if the index node is referred to in the TNC and the corresponding znode is
3136 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3137 * znode is clean, and a negative error code in case of failure.
3139 * Note, the @key argument has to be the key of the first child. Also note,
3140 * this function relies on the fact that 0:0 is never a valid LEB number and
3141 * offset for a main-area node.
3143 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3146 struct ubifs_znode *znode;
3148 znode = lookup_znode(c, key, level, lnum, offs);
3152 return PTR_ERR(znode);
3154 return ubifs_zn_dirty(znode) ? 1 : 2;
3158 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3159 * @c: UBIFS file-system description object
3161 * @lnum: node LEB number
3162 * @offs: node offset
3164 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3165 * not, and a negative error code in case of failure.
3167 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3168 * and offset for a main-area node.
3170 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3173 struct ubifs_zbranch *zbr;
3174 struct ubifs_znode *znode, *zn;
3175 int n, found, err, nn;
3176 const int unique = !is_hash_key(c, key);
3178 found = ubifs_lookup_level0(c, key, &znode, &n);
3180 return found; /* Error code */
3183 zbr = &znode->zbranch[n];
3184 if (lnum == zbr->lnum && offs == zbr->offs)
3185 return 1; /* Found it */
3189 * Because the key is not unique, we have to look left
3196 err = tnc_prev(c, &znode, &n);
3201 if (keys_cmp(c, key, &znode->zbranch[n].key))
3203 zbr = &znode->zbranch[n];
3204 if (lnum == zbr->lnum && offs == zbr->offs)
3205 return 1; /* Found it */
3211 err = tnc_next(c, &znode, &n);
3217 if (keys_cmp(c, key, &znode->zbranch[n].key))
3219 zbr = &znode->zbranch[n];
3220 if (lnum == zbr->lnum && offs == zbr->offs)
3221 return 1; /* Found it */
3227 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3228 * @c: UBIFS file-system description object
3230 * @level: index node level (if it is an index node)
3231 * @lnum: node LEB number
3232 * @offs: node offset
3233 * @is_idx: non-zero if the node is an index node
3235 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3236 * negative error code in case of failure. For index nodes, @key has to be the
3237 * key of the first child. An index node is considered to be in the TNC only if
3238 * the corresponding znode is clean or has not been loaded.
3240 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3241 int lnum, int offs, int is_idx)
3245 mutex_lock(&c->tnc_mutex);
3247 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3251 /* The index node was found but it was dirty */
3254 /* The index node was found and it was clean */
3259 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3262 mutex_unlock(&c->tnc_mutex);
3267 * ubifs_dirty_idx_node - dirty an index node.
3268 * @c: UBIFS file-system description object
3269 * @key: index node key
3270 * @level: index node level
3271 * @lnum: index node LEB number
3272 * @offs: index node offset
3274 * This function loads and dirties an index node so that it can be garbage
3275 * collected. The @key argument has to be the key of the first child. This
3276 * function relies on the fact that 0:0 is never a valid LEB number and offset
3277 * for a main-area node. Returns %0 on success and a negative error code on
3280 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3283 struct ubifs_znode *znode;
3286 mutex_lock(&c->tnc_mutex);
3287 znode = lookup_znode(c, key, level, lnum, offs);
3290 if (IS_ERR(znode)) {
3291 err = PTR_ERR(znode);
3294 znode = dirty_cow_bottom_up(c, znode);
3295 if (IS_ERR(znode)) {
3296 err = PTR_ERR(znode);
3301 mutex_unlock(&c->tnc_mutex);
3305 #ifdef CONFIG_UBIFS_FS_DEBUG
3308 * dbg_check_inode_size - check if inode size is correct.
3309 * @c: UBIFS file-system description object
3310 * @inum: inode number
3313 * This function makes sure that the inode size (@size) is correct and it does
3314 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3315 * if it has a data page beyond @size, and other negative error code in case of
3318 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3322 union ubifs_key from_key, to_key, *key;
3323 struct ubifs_znode *znode;
3326 if (!S_ISREG(inode->i_mode))
3328 if (!dbg_is_chk_gen(c))
3331 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3332 data_key_init(c, &from_key, inode->i_ino, block);
3333 highest_data_key(c, &to_key, inode->i_ino);
3335 mutex_lock(&c->tnc_mutex);
3336 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3346 err = tnc_next(c, &znode, &n);
3347 if (err == -ENOENT) {
3354 ubifs_assert(err == 0);
3355 key = &znode->zbranch[n].key;
3356 if (!key_in_range(c, key, &from_key, &to_key))
3360 block = key_block(c, key);
3361 ubifs_err("inode %lu has size %lld, but there are data at offset %lld "
3362 "(data key %s)", (unsigned long)inode->i_ino, size,
3363 ((loff_t)block) << UBIFS_BLOCK_SHIFT, DBGKEY(key));
3364 mutex_unlock(&c->tnc_mutex);
3365 dbg_dump_inode(c, inode);
3370 mutex_unlock(&c->tnc_mutex);
3374 #endif /* CONFIG_UBIFS_FS_DEBUG */