1 // SPDX-License-Identifier: GPL-2.0-only
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
21 #include <linux/crc32.h>
22 #include <linux/slab.h>
25 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
26 struct ubifs_zbranch *zbr);
27 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
28 struct ubifs_zbranch *zbr, void *node);
31 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
32 * @NAME_LESS: name corresponding to the first argument is less than second
33 * @NAME_MATCHES: names match
34 * @NAME_GREATER: name corresponding to the second argument is greater than
36 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
38 * These constants were introduce to improve readability.
48 * insert_old_idx - record an index node obsoleted since the last commit start.
49 * @c: UBIFS file-system description object
50 * @lnum: LEB number of obsoleted index node
51 * @offs: offset of obsoleted index node
53 * Returns %0 on success, and a negative error code on failure.
55 * For recovery, there must always be a complete intact version of the index on
56 * flash at all times. That is called the "old index". It is the index as at the
57 * time of the last successful commit. Many of the index nodes in the old index
58 * may be dirty, but they must not be erased until the next successful commit
59 * (at which point that index becomes the old index).
61 * That means that the garbage collection and the in-the-gaps method of
62 * committing must be able to determine if an index node is in the old index.
63 * Most of the old index nodes can be found by looking up the TNC using the
64 * 'lookup_znode()' function. However, some of the old index nodes may have
65 * been deleted from the current index or may have been changed so much that
66 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
67 * That is what this function does. The RB-tree is ordered by LEB number and
68 * offset because they uniquely identify the old index node.
70 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
72 struct ubifs_old_idx *old_idx, *o;
73 struct rb_node **p, *parent = NULL;
75 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
76 if (unlikely(!old_idx))
81 p = &c->old_idx.rb_node;
84 o = rb_entry(parent, struct ubifs_old_idx, rb);
87 else if (lnum > o->lnum)
89 else if (offs < o->offs)
91 else if (offs > o->offs)
94 ubifs_err(c, "old idx added twice!");
99 rb_link_node(&old_idx->rb, parent, p);
100 rb_insert_color(&old_idx->rb, &c->old_idx);
105 * insert_old_idx_znode - record a znode obsoleted since last commit start.
106 * @c: UBIFS file-system description object
107 * @znode: znode of obsoleted index node
109 * Returns %0 on success, and a negative error code on failure.
111 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
114 struct ubifs_zbranch *zbr;
116 zbr = &znode->parent->zbranch[znode->iip];
118 return insert_old_idx(c, zbr->lnum, zbr->offs);
121 return insert_old_idx(c, c->zroot.lnum,
127 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
128 * @c: UBIFS file-system description object
129 * @znode: znode of obsoleted index node
131 * Returns %0 on success, and a negative error code on failure.
133 static int ins_clr_old_idx_znode(struct ubifs_info *c,
134 struct ubifs_znode *znode)
139 struct ubifs_zbranch *zbr;
141 zbr = &znode->parent->zbranch[znode->iip];
143 err = insert_old_idx(c, zbr->lnum, zbr->offs);
152 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
163 * destroy_old_idx - destroy the old_idx RB-tree.
164 * @c: UBIFS file-system description object
166 * During start commit, the old_idx RB-tree is used to avoid overwriting index
167 * nodes that were in the index last commit but have since been deleted. This
168 * is necessary for recovery i.e. the old index must be kept intact until the
169 * new index is successfully written. The old-idx RB-tree is used for the
170 * in-the-gaps method of writing index nodes and is destroyed every commit.
172 void destroy_old_idx(struct ubifs_info *c)
174 struct ubifs_old_idx *old_idx, *n;
176 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
179 c->old_idx = RB_ROOT;
183 * copy_znode - copy a dirty znode.
184 * @c: UBIFS file-system description object
185 * @znode: znode to copy
187 * A dirty znode being committed may not be changed, so it is copied.
189 static struct ubifs_znode *copy_znode(struct ubifs_info *c,
190 struct ubifs_znode *znode)
192 struct ubifs_znode *zn;
194 zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS);
196 return ERR_PTR(-ENOMEM);
199 __set_bit(DIRTY_ZNODE, &zn->flags);
200 __clear_bit(COW_ZNODE, &zn->flags);
202 ubifs_assert(c, !ubifs_zn_obsolete(znode));
203 __set_bit(OBSOLETE_ZNODE, &znode->flags);
205 if (znode->level != 0) {
207 const int n = zn->child_cnt;
209 /* The children now have new parent */
210 for (i = 0; i < n; i++) {
211 struct ubifs_zbranch *zbr = &zn->zbranch[i];
214 zbr->znode->parent = zn;
218 atomic_long_inc(&c->dirty_zn_cnt);
223 * add_idx_dirt - add dirt due to a dirty znode.
224 * @c: UBIFS file-system description object
225 * @lnum: LEB number of index node
226 * @dirt: size of index node
228 * This function updates lprops dirty space and the new size of the index.
230 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
232 c->calc_idx_sz -= ALIGN(dirt, 8);
233 return ubifs_add_dirt(c, lnum, dirt);
237 * dirty_cow_znode - ensure a znode is not being committed.
238 * @c: UBIFS file-system description object
239 * @zbr: branch of znode to check
241 * Returns dirtied znode on success or negative error code on failure.
243 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
244 struct ubifs_zbranch *zbr)
246 struct ubifs_znode *znode = zbr->znode;
247 struct ubifs_znode *zn;
250 if (!ubifs_zn_cow(znode)) {
251 /* znode is not being committed */
252 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
253 atomic_long_inc(&c->dirty_zn_cnt);
254 atomic_long_dec(&c->clean_zn_cnt);
255 atomic_long_dec(&ubifs_clean_zn_cnt);
256 err = add_idx_dirt(c, zbr->lnum, zbr->len);
263 zn = copy_znode(c, znode);
268 err = insert_old_idx(c, zbr->lnum, zbr->offs);
271 err = add_idx_dirt(c, zbr->lnum, zbr->len);
286 * lnc_add - add a leaf node to the leaf node cache.
287 * @c: UBIFS file-system description object
288 * @zbr: zbranch of leaf node
291 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
292 * purpose of the leaf node cache is to save re-reading the same leaf node over
293 * and over again. Most things are cached by VFS, however the file system must
294 * cache directory entries for readdir and for resolving hash collisions. The
295 * present implementation of the leaf node cache is extremely simple, and
296 * allows for error returns that are not used but that may be needed if a more
297 * complex implementation is created.
299 * Note, this function does not add the @node object to LNC directly, but
300 * allocates a copy of the object and adds the copy to LNC. The reason for this
301 * is that @node has been allocated outside of the TNC subsystem and will be
302 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
303 * may be changed at any time, e.g. freed by the shrinker.
305 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
310 const struct ubifs_dent_node *dent = node;
312 ubifs_assert(c, !zbr->leaf);
313 ubifs_assert(c, zbr->len != 0);
314 ubifs_assert(c, is_hash_key(c, &zbr->key));
316 err = ubifs_validate_entry(c, dent);
319 ubifs_dump_node(c, dent);
323 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
325 /* We don't have to have the cache, so no error */
328 zbr->leaf = lnc_node;
333 * lnc_add_directly - add a leaf node to the leaf-node-cache.
334 * @c: UBIFS file-system description object
335 * @zbr: zbranch of leaf node
338 * This function is similar to 'lnc_add()', but it does not create a copy of
339 * @node but inserts @node to TNC directly.
341 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
346 ubifs_assert(c, !zbr->leaf);
347 ubifs_assert(c, zbr->len != 0);
349 err = ubifs_validate_entry(c, node);
352 ubifs_dump_node(c, node);
361 * lnc_free - remove a leaf node from the leaf node cache.
362 * @zbr: zbranch of leaf node
365 static void lnc_free(struct ubifs_zbranch *zbr)
374 * tnc_read_hashed_node - read a "hashed" leaf node.
375 * @c: UBIFS file-system description object
376 * @zbr: key and position of the node
377 * @node: node is returned here
379 * This function reads a "hashed" node defined by @zbr from the leaf node cache
380 * (in it is there) or from the hash media, in which case the node is also
381 * added to LNC. Returns zero in case of success or a negative negative error
382 * code in case of failure.
384 static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr,
389 ubifs_assert(c, is_hash_key(c, &zbr->key));
392 /* Read from the leaf node cache */
393 ubifs_assert(c, zbr->len != 0);
394 memcpy(node, zbr->leaf, zbr->len);
399 err = fallible_read_node(c, &zbr->key, zbr, node);
401 * When the node was not found, return -ENOENT, 0 otherwise.
402 * Negative return codes stay as-is.
409 err = ubifs_tnc_read_node(c, zbr, node);
414 /* Add the node to the leaf node cache */
415 err = lnc_add(c, zbr, node);
420 * try_read_node - read a node if it is a node.
421 * @c: UBIFS file-system description object
422 * @buf: buffer to read to
424 * @zbr: the zbranch describing the node to read
426 * This function tries to read a node of known type and length, checks it and
427 * stores it in @buf. This function returns %1 if a node is present and %0 if
428 * a node is not present. A negative error code is returned for I/O errors.
429 * This function performs that same function as ubifs_read_node except that
430 * it does not require that there is actually a node present and instead
431 * the return code indicates if a node was read.
433 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
434 * is true (it is controlled by corresponding mount option). However, if
435 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
436 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
437 * because during mounting or re-mounting from R/O mode to R/W mode we may read
438 * journal nodes (when replying the journal or doing the recovery) and the
439 * journal nodes may potentially be corrupted, so checking is required.
441 static int try_read_node(const struct ubifs_info *c, void *buf, int type,
442 struct ubifs_zbranch *zbr)
445 int lnum = zbr->lnum;
446 int offs = zbr->offs;
448 struct ubifs_ch *ch = buf;
449 uint32_t crc, node_crc;
451 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
453 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
455 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
456 type, lnum, offs, err);
460 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
463 if (ch->node_type != type)
466 node_len = le32_to_cpu(ch->len);
470 if (type != UBIFS_DATA_NODE || !c->no_chk_data_crc || c->mounting ||
472 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
473 node_crc = le32_to_cpu(ch->crc);
478 err = ubifs_node_check_hash(c, buf, zbr->hash);
480 ubifs_bad_hash(c, buf, zbr->hash, lnum, offs);
488 * fallible_read_node - try to read a leaf node.
489 * @c: UBIFS file-system description object
490 * @key: key of node to read
491 * @zbr: position of node
492 * @node: node returned
494 * This function tries to read a node and returns %1 if the node is read, %0
495 * if the node is not present, and a negative error code in the case of error.
497 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
498 struct ubifs_zbranch *zbr, void *node)
502 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
504 ret = try_read_node(c, node, key_type(c, key), zbr);
506 union ubifs_key node_key;
507 struct ubifs_dent_node *dent = node;
509 /* All nodes have key in the same place */
510 key_read(c, &dent->key, &node_key);
511 if (keys_cmp(c, key, &node_key) != 0)
514 if (ret == 0 && c->replaying)
515 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
516 zbr->lnum, zbr->offs, zbr->len);
521 * matches_name - determine if a direntry or xattr entry matches a given name.
522 * @c: UBIFS file-system description object
523 * @zbr: zbranch of dent
526 * This function checks if xentry/direntry referred by zbranch @zbr matches name
527 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
528 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
529 * of failure, a negative error code is returned.
531 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
532 const struct fscrypt_name *nm)
534 struct ubifs_dent_node *dent;
537 /* If possible, match against the dent in the leaf node cache */
539 dent = kmalloc(zbr->len, GFP_NOFS);
543 err = ubifs_tnc_read_node(c, zbr, dent);
547 /* Add the node to the leaf node cache */
548 err = lnc_add_directly(c, zbr, dent);
554 nlen = le16_to_cpu(dent->nlen);
555 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
557 if (nlen == fname_len(nm))
559 else if (nlen < fname_len(nm))
574 * get_znode - get a TNC znode that may not be loaded yet.
575 * @c: UBIFS file-system description object
576 * @znode: parent znode
577 * @n: znode branch slot number
579 * This function returns the znode or a negative error code.
581 static struct ubifs_znode *get_znode(struct ubifs_info *c,
582 struct ubifs_znode *znode, int n)
584 struct ubifs_zbranch *zbr;
586 zbr = &znode->zbranch[n];
590 znode = ubifs_load_znode(c, zbr, znode, n);
595 * tnc_next - find next TNC entry.
596 * @c: UBIFS file-system description object
597 * @zn: znode is passed and returned here
598 * @n: znode branch slot number is passed and returned here
600 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
601 * no next entry, or a negative error code otherwise.
603 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
605 struct ubifs_znode *znode = *zn;
609 if (nn < znode->child_cnt) {
614 struct ubifs_znode *zp;
621 if (nn < znode->child_cnt) {
622 znode = get_znode(c, znode, nn);
624 return PTR_ERR(znode);
625 while (znode->level != 0) {
626 znode = get_znode(c, znode, 0);
628 return PTR_ERR(znode);
640 * tnc_prev - find previous TNC entry.
641 * @c: UBIFS file-system description object
642 * @zn: znode is returned here
643 * @n: znode branch slot number is passed and returned here
645 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
646 * there is no next entry, or a negative error code otherwise.
648 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
650 struct ubifs_znode *znode = *zn;
658 struct ubifs_znode *zp;
666 znode = get_znode(c, znode, nn);
668 return PTR_ERR(znode);
669 while (znode->level != 0) {
670 nn = znode->child_cnt - 1;
671 znode = get_znode(c, znode, nn);
673 return PTR_ERR(znode);
675 nn = znode->child_cnt - 1;
685 * resolve_collision - resolve a collision.
686 * @c: UBIFS file-system description object
687 * @key: key of a directory or extended attribute entry
688 * @zn: znode is returned here
689 * @n: zbranch number is passed and returned here
690 * @nm: name of the entry
692 * This function is called for "hashed" keys to make sure that the found key
693 * really corresponds to the looked up node (directory or extended attribute
694 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
695 * %0 is returned if @nm is not found and @zn and @n are set to the previous
696 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
697 * This means that @n may be set to %-1 if the leftmost key in @zn is the
698 * previous one. A negative error code is returned on failures.
700 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
701 struct ubifs_znode **zn, int *n,
702 const struct fscrypt_name *nm)
706 err = matches_name(c, &(*zn)->zbranch[*n], nm);
707 if (unlikely(err < 0))
709 if (err == NAME_MATCHES)
712 if (err == NAME_GREATER) {
715 err = tnc_prev(c, zn, n);
716 if (err == -ENOENT) {
717 ubifs_assert(c, *n == 0);
723 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
725 * We have found the branch after which we would
726 * like to insert, but inserting in this znode
727 * may still be wrong. Consider the following 3
728 * znodes, in the case where we are resolving a
729 * collision with Key2.
732 * ----------------------
733 * level 1 | Key0 | Key1 |
734 * -----------------------
736 * znode za | | znode zb
737 * ------------ ------------
738 * level 0 | Key0 | | Key2 |
739 * ------------ ------------
741 * The lookup finds Key2 in znode zb. Lets say
742 * there is no match and the name is greater so
743 * we look left. When we find Key0, we end up
744 * here. If we return now, we will insert into
745 * znode za at slot n = 1. But that is invalid
746 * according to the parent's keys. Key2 must
747 * be inserted into znode zb.
749 * Note, this problem is not relevant for the
750 * case when we go right, because
751 * 'tnc_insert()' would correct the parent key.
753 if (*n == (*zn)->child_cnt - 1) {
754 err = tnc_next(c, zn, n);
756 /* Should be impossible */
762 ubifs_assert(c, *n == 0);
767 err = matches_name(c, &(*zn)->zbranch[*n], nm);
770 if (err == NAME_LESS)
772 if (err == NAME_MATCHES)
774 ubifs_assert(c, err == NAME_GREATER);
778 struct ubifs_znode *znode = *zn;
782 err = tnc_next(c, &znode, &nn);
787 if (keys_cmp(c, &znode->zbranch[nn].key, key))
789 err = matches_name(c, &znode->zbranch[nn], nm);
792 if (err == NAME_GREATER)
796 if (err == NAME_MATCHES)
798 ubifs_assert(c, err == NAME_LESS);
804 * fallible_matches_name - determine if a dent matches a given name.
805 * @c: UBIFS file-system description object
806 * @zbr: zbranch of dent
809 * This is a "fallible" version of 'matches_name()' function which does not
810 * panic if the direntry/xentry referred by @zbr does not exist on the media.
812 * This function checks if xentry/direntry referred by zbranch @zbr matches name
813 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
814 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
815 * if xentry/direntry referred by @zbr does not exist on the media. A negative
816 * error code is returned in case of failure.
818 static int fallible_matches_name(struct ubifs_info *c,
819 struct ubifs_zbranch *zbr,
820 const struct fscrypt_name *nm)
822 struct ubifs_dent_node *dent;
825 /* If possible, match against the dent in the leaf node cache */
827 dent = kmalloc(zbr->len, GFP_NOFS);
831 err = fallible_read_node(c, &zbr->key, zbr, dent);
835 /* The node was not present */
839 ubifs_assert(c, err == 1);
841 err = lnc_add_directly(c, zbr, dent);
847 nlen = le16_to_cpu(dent->nlen);
848 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
850 if (nlen == fname_len(nm))
852 else if (nlen < fname_len(nm))
867 * fallible_resolve_collision - resolve a collision even if nodes are missing.
868 * @c: UBIFS file-system description object
870 * @zn: znode is returned here
871 * @n: branch number is passed and returned here
872 * @nm: name of directory entry
873 * @adding: indicates caller is adding a key to the TNC
875 * This is a "fallible" version of the 'resolve_collision()' function which
876 * does not panic if one of the nodes referred to by TNC does not exist on the
877 * media. This may happen when replaying the journal if a deleted node was
878 * Garbage-collected and the commit was not done. A branch that refers to a node
879 * that is not present is called a dangling branch. The following are the return
880 * codes for this function:
881 * o if @nm was found, %1 is returned and @zn and @n are set to the found
883 * o if we are @adding and @nm was not found, %0 is returned;
884 * o if we are not @adding and @nm was not found, but a dangling branch was
885 * found, then %1 is returned and @zn and @n are set to the dangling branch;
886 * o a negative error code is returned in case of failure.
888 static int fallible_resolve_collision(struct ubifs_info *c,
889 const union ubifs_key *key,
890 struct ubifs_znode **zn, int *n,
891 const struct fscrypt_name *nm,
894 struct ubifs_znode *o_znode = NULL, *znode = *zn;
895 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
897 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
898 if (unlikely(cmp < 0))
900 if (cmp == NAME_MATCHES)
902 if (cmp == NOT_ON_MEDIA) {
906 * We are unlucky and hit a dangling branch straight away.
907 * Now we do not really know where to go to find the needed
908 * branch - to the left or to the right. Well, let's try left.
912 unsure = 1; /* Remove a dangling branch wherever it is */
914 if (cmp == NAME_GREATER || unsure) {
917 err = tnc_prev(c, zn, n);
918 if (err == -ENOENT) {
919 ubifs_assert(c, *n == 0);
925 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
926 /* See comments in 'resolve_collision()' */
927 if (*n == (*zn)->child_cnt - 1) {
928 err = tnc_next(c, zn, n);
930 /* Should be impossible */
936 ubifs_assert(c, *n == 0);
941 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
944 if (err == NAME_MATCHES)
946 if (err == NOT_ON_MEDIA) {
953 if (err == NAME_LESS)
960 if (cmp == NAME_LESS || unsure) {
965 err = tnc_next(c, &znode, &nn);
970 if (keys_cmp(c, &znode->zbranch[nn].key, key))
972 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
975 if (err == NAME_GREATER)
979 if (err == NAME_MATCHES)
981 if (err == NOT_ON_MEDIA) {
988 /* Never match a dangling branch when adding */
989 if (adding || !o_znode)
992 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
993 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
994 o_znode->zbranch[o_n].len);
1001 * matches_position - determine if a zbranch matches a given position.
1002 * @zbr: zbranch of dent
1003 * @lnum: LEB number of dent to match
1004 * @offs: offset of dent to match
1006 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1008 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1010 if (zbr->lnum == lnum && zbr->offs == offs)
1017 * resolve_collision_directly - resolve a collision directly.
1018 * @c: UBIFS file-system description object
1019 * @key: key of directory entry
1020 * @zn: znode is passed and returned here
1021 * @n: zbranch number is passed and returned here
1022 * @lnum: LEB number of dent node to match
1023 * @offs: offset of dent node to match
1025 * This function is used for "hashed" keys to make sure the found directory or
1026 * extended attribute entry node is what was looked for. It is used when the
1027 * flash address of the right node is known (@lnum:@offs) which makes it much
1028 * easier to resolve collisions (no need to read entries and match full
1029 * names). This function returns %1 and sets @zn and @n if the collision is
1030 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1031 * previous directory entry. Otherwise a negative error code is returned.
1033 static int resolve_collision_directly(struct ubifs_info *c,
1034 const union ubifs_key *key,
1035 struct ubifs_znode **zn, int *n,
1038 struct ubifs_znode *znode;
1043 if (matches_position(&znode->zbranch[nn], lnum, offs))
1048 err = tnc_prev(c, &znode, &nn);
1053 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1055 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1066 err = tnc_next(c, &znode, &nn);
1071 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1075 if (matches_position(&znode->zbranch[nn], lnum, offs))
1081 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1082 * @c: UBIFS file-system description object
1083 * @znode: znode to dirty
1085 * If we do not have a unique key that resides in a znode, then we cannot
1086 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1087 * This function records the path back to the last dirty ancestor, and then
1088 * dirties the znodes on that path.
1090 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1091 struct ubifs_znode *znode)
1093 struct ubifs_znode *zp;
1094 int *path = c->bottom_up_buf, p = 0;
1096 ubifs_assert(c, c->zroot.znode);
1097 ubifs_assert(c, znode);
1098 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1099 kfree(c->bottom_up_buf);
1100 c->bottom_up_buf = kmalloc_array(c->zroot.znode->level,
1103 if (!c->bottom_up_buf)
1104 return ERR_PTR(-ENOMEM);
1105 path = c->bottom_up_buf;
1107 if (c->zroot.znode->level) {
1108 /* Go up until parent is dirty */
1116 ubifs_assert(c, p < c->zroot.znode->level);
1118 if (!zp->cnext && ubifs_zn_dirty(znode))
1124 /* Come back down, dirtying as we go */
1126 struct ubifs_zbranch *zbr;
1130 ubifs_assert(c, path[p - 1] >= 0);
1131 ubifs_assert(c, path[p - 1] < zp->child_cnt);
1132 zbr = &zp->zbranch[path[--p]];
1133 znode = dirty_cow_znode(c, zbr);
1135 ubifs_assert(c, znode == c->zroot.znode);
1136 znode = dirty_cow_znode(c, &c->zroot);
1138 if (IS_ERR(znode) || !p)
1140 ubifs_assert(c, path[p - 1] >= 0);
1141 ubifs_assert(c, path[p - 1] < znode->child_cnt);
1142 znode = znode->zbranch[path[p - 1]].znode;
1149 * ubifs_lookup_level0 - search for zero-level znode.
1150 * @c: UBIFS file-system description object
1151 * @key: key to lookup
1152 * @zn: znode is returned here
1153 * @n: znode branch slot number is returned here
1155 * This function looks up the TNC tree and search for zero-level znode which
1156 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1158 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1159 * is returned and slot number of the matched branch is stored in @n;
1160 * o not exact match, which means that zero-level znode does not contain
1161 * @key, then %0 is returned and slot number of the closest branch is stored
1163 * o @key is so small that it is even less than the lowest key of the
1164 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1166 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1167 * function reads corresponding indexing nodes and inserts them to TNC. In
1168 * case of failure, a negative error code is returned.
1170 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1171 struct ubifs_znode **zn, int *n)
1174 struct ubifs_znode *znode;
1175 time64_t time = ktime_get_seconds();
1177 dbg_tnck(key, "search key ");
1178 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
1180 znode = c->zroot.znode;
1181 if (unlikely(!znode)) {
1182 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1184 return PTR_ERR(znode);
1190 struct ubifs_zbranch *zbr;
1192 exact = ubifs_search_zbranch(c, znode, key, n);
1194 if (znode->level == 0)
1199 zbr = &znode->zbranch[*n];
1207 /* znode is not in TNC cache, load it from the media */
1208 znode = ubifs_load_znode(c, zbr, znode, *n);
1210 return PTR_ERR(znode);
1214 if (exact || !is_hash_key(c, key) || *n != -1) {
1215 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1220 * Here is a tricky place. We have not found the key and this is a
1221 * "hashed" key, which may collide. The rest of the code deals with
1222 * situations like this:
1226 * | 3 | 5 | | 6 | 7 | (x)
1228 * Or more a complex example:
1232 * | 1 | 3 | | 5 | 8 |
1234 * | 5 | 5 | | 6 | 7 | (x)
1236 * In the examples, if we are looking for key "5", we may reach nodes
1237 * marked with "(x)". In this case what we have do is to look at the
1238 * left and see if there is "5" key there. If there is, we have to
1241 * Note, this whole situation is possible because we allow to have
1242 * elements which are equivalent to the next key in the parent in the
1243 * children of current znode. For example, this happens if we split a
1244 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1248 * | 3 | 5 | | 5 | 6 | 7 |
1250 * And this becomes what is at the first "picture" after key "5" marked
1251 * with "^" is removed. What could be done is we could prohibit
1252 * splitting in the middle of the colliding sequence. Also, when
1253 * removing the leftmost key, we would have to correct the key of the
1254 * parent node, which would introduce additional complications. Namely,
1255 * if we changed the leftmost key of the parent znode, the garbage
1256 * collector would be unable to find it (GC is doing this when GC'ing
1257 * indexing LEBs). Although we already have an additional RB-tree where
1258 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1259 * after the commit. But anyway, this does not look easy to implement
1260 * so we did not try this.
1262 err = tnc_prev(c, &znode, n);
1263 if (err == -ENOENT) {
1264 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1268 if (unlikely(err < 0))
1270 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1271 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1276 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1282 * lookup_level0_dirty - search for zero-level znode dirtying.
1283 * @c: UBIFS file-system description object
1284 * @key: key to lookup
1285 * @zn: znode is returned here
1286 * @n: znode branch slot number is returned here
1288 * This function looks up the TNC tree and search for zero-level znode which
1289 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1291 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1292 * is returned and slot number of the matched branch is stored in @n;
1293 * o not exact match, which means that zero-level znode does not contain @key
1294 * then %0 is returned and slot number of the closed branch is stored in
1296 * o @key is so small that it is even less than the lowest key of the
1297 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1299 * Additionally all znodes in the path from the root to the located zero-level
1300 * znode are marked as dirty.
1302 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1303 * function reads corresponding indexing nodes and inserts them to TNC. In
1304 * case of failure, a negative error code is returned.
1306 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1307 struct ubifs_znode **zn, int *n)
1310 struct ubifs_znode *znode;
1311 time64_t time = ktime_get_seconds();
1313 dbg_tnck(key, "search and dirty key ");
1315 znode = c->zroot.znode;
1316 if (unlikely(!znode)) {
1317 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1319 return PTR_ERR(znode);
1322 znode = dirty_cow_znode(c, &c->zroot);
1324 return PTR_ERR(znode);
1329 struct ubifs_zbranch *zbr;
1331 exact = ubifs_search_zbranch(c, znode, key, n);
1333 if (znode->level == 0)
1338 zbr = &znode->zbranch[*n];
1342 znode = dirty_cow_znode(c, zbr);
1344 return PTR_ERR(znode);
1348 /* znode is not in TNC cache, load it from the media */
1349 znode = ubifs_load_znode(c, zbr, znode, *n);
1351 return PTR_ERR(znode);
1352 znode = dirty_cow_znode(c, zbr);
1354 return PTR_ERR(znode);
1358 if (exact || !is_hash_key(c, key) || *n != -1) {
1359 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1364 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1367 err = tnc_prev(c, &znode, n);
1368 if (err == -ENOENT) {
1370 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1373 if (unlikely(err < 0))
1375 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1377 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1381 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1382 znode = dirty_cow_bottom_up(c, znode);
1384 return PTR_ERR(znode);
1387 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1393 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1394 * @c: UBIFS file-system description object
1396 * @gc_seq1: garbage collection sequence number
1398 * This function determines if @lnum may have been garbage collected since
1399 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1402 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1404 int gc_seq2, gced_lnum;
1406 gced_lnum = c->gced_lnum;
1408 gc_seq2 = c->gc_seq;
1409 /* Same seq means no GC */
1410 if (gc_seq1 == gc_seq2)
1412 /* Different by more than 1 means we don't know */
1413 if (gc_seq1 + 1 != gc_seq2)
1416 * We have seen the sequence number has increased by 1. Now we need to
1417 * be sure we read the right LEB number, so read it again.
1420 if (gced_lnum != c->gced_lnum)
1422 /* Finally we can check lnum */
1423 if (gced_lnum == lnum)
1429 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1430 * @c: UBIFS file-system description object
1431 * @key: node key to lookup
1432 * @node: the node is returned here
1433 * @lnum: LEB number is returned here
1434 * @offs: offset is returned here
1436 * This function looks up and reads node with key @key. The caller has to make
1437 * sure the @node buffer is large enough to fit the node. Returns zero in case
1438 * of success, %-ENOENT if the node was not found, and a negative error code in
1439 * case of failure. The node location can be returned in @lnum and @offs.
1441 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1442 void *node, int *lnum, int *offs)
1444 int found, n, err, safely = 0, gc_seq1;
1445 struct ubifs_znode *znode;
1446 struct ubifs_zbranch zbr, *zt;
1449 mutex_lock(&c->tnc_mutex);
1450 found = ubifs_lookup_level0(c, key, &znode, &n);
1454 } else if (found < 0) {
1458 zt = &znode->zbranch[n];
1463 if (is_hash_key(c, key)) {
1465 * In this case the leaf node cache gets used, so we pass the
1466 * address of the zbranch and keep the mutex locked
1468 err = tnc_read_hashed_node(c, zt, node);
1472 err = ubifs_tnc_read_node(c, zt, node);
1475 /* Drop the TNC mutex prematurely and race with garbage collection */
1476 zbr = znode->zbranch[n];
1477 gc_seq1 = c->gc_seq;
1478 mutex_unlock(&c->tnc_mutex);
1480 if (ubifs_get_wbuf(c, zbr.lnum)) {
1481 /* We do not GC journal heads */
1482 err = ubifs_tnc_read_node(c, &zbr, node);
1486 err = fallible_read_node(c, key, &zbr, node);
1487 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1489 * The node may have been GC'ed out from under us so try again
1490 * while keeping the TNC mutex locked.
1498 mutex_unlock(&c->tnc_mutex);
1503 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1504 * @c: UBIFS file-system description object
1505 * @bu: bulk-read parameters and results
1507 * Lookup consecutive data node keys for the same inode that reside
1508 * consecutively in the same LEB. This function returns zero in case of success
1509 * and a negative error code in case of failure.
1511 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1512 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1513 * maximum possible amount of nodes for bulk-read.
1515 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1517 int n, err = 0, lnum = -1, uninitialized_var(offs);
1518 int uninitialized_var(len);
1519 unsigned int block = key_block(c, &bu->key);
1520 struct ubifs_znode *znode;
1526 mutex_lock(&c->tnc_mutex);
1527 /* Find first key */
1528 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1533 len = znode->zbranch[n].len;
1534 /* The buffer must be big enough for at least 1 node */
1535 if (len > bu->buf_len) {
1540 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1542 lnum = znode->zbranch[n].lnum;
1543 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1546 struct ubifs_zbranch *zbr;
1547 union ubifs_key *key;
1548 unsigned int next_block;
1551 err = tnc_next(c, &znode, &n);
1554 zbr = &znode->zbranch[n];
1556 /* See if there is another data key for this file */
1557 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1558 key_type(c, key) != UBIFS_DATA_KEY) {
1563 /* First key found */
1565 offs = ALIGN(zbr->offs + zbr->len, 8);
1567 if (len > bu->buf_len) {
1573 * The data nodes must be in consecutive positions in
1576 if (zbr->lnum != lnum || zbr->offs != offs)
1578 offs += ALIGN(zbr->len, 8);
1579 len = ALIGN(len, 8) + zbr->len;
1580 /* Must not exceed buffer length */
1581 if (len > bu->buf_len)
1584 /* Allow for holes */
1585 next_block = key_block(c, key);
1586 bu->blk_cnt += (next_block - block - 1);
1587 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1591 bu->zbranch[bu->cnt++] = *zbr;
1593 /* See if we have room for more */
1594 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1596 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1600 if (err == -ENOENT) {
1604 bu->gc_seq = c->gc_seq;
1605 mutex_unlock(&c->tnc_mutex);
1609 * An enormous hole could cause bulk-read to encompass too many
1610 * page cache pages, so limit the number here.
1612 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1613 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1615 * Ensure that bulk-read covers a whole number of page cache
1618 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1619 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1622 /* At the end of file we can round up */
1623 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1626 /* Exclude data nodes that do not make up a whole page cache page */
1627 block = key_block(c, &bu->key) + bu->blk_cnt;
1628 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1630 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1638 * read_wbuf - bulk-read from a LEB with a wbuf.
1639 * @wbuf: wbuf that may overlap the read
1640 * @buf: buffer into which to read
1642 * @lnum: LEB number from which to read
1643 * @offs: offset from which to read
1645 * This functions returns %0 on success or a negative error code on failure.
1647 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1650 const struct ubifs_info *c = wbuf->c;
1653 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1654 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1655 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1656 ubifs_assert(c, offs + len <= c->leb_size);
1658 spin_lock(&wbuf->lock);
1659 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1661 /* We may safely unlock the write-buffer and read the data */
1662 spin_unlock(&wbuf->lock);
1663 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1666 /* Don't read under wbuf */
1667 rlen = wbuf->offs - offs;
1671 /* Copy the rest from the write-buffer */
1672 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1673 spin_unlock(&wbuf->lock);
1676 /* Read everything that goes before write-buffer */
1677 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1683 * validate_data_node - validate data nodes for bulk-read.
1684 * @c: UBIFS file-system description object
1685 * @buf: buffer containing data node to validate
1686 * @zbr: zbranch of data node to validate
1688 * This functions returns %0 on success or a negative error code on failure.
1690 static int validate_data_node(struct ubifs_info *c, void *buf,
1691 struct ubifs_zbranch *zbr)
1693 union ubifs_key key1;
1694 struct ubifs_ch *ch = buf;
1697 if (ch->node_type != UBIFS_DATA_NODE) {
1698 ubifs_err(c, "bad node type (%d but expected %d)",
1699 ch->node_type, UBIFS_DATA_NODE);
1703 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1705 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1709 err = ubifs_node_check_hash(c, buf, zbr->hash);
1711 ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs);
1715 len = le32_to_cpu(ch->len);
1716 if (len != zbr->len) {
1717 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1721 /* Make sure the key of the read node is correct */
1722 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1723 if (!keys_eq(c, &zbr->key, &key1)) {
1724 ubifs_err(c, "bad key in node at LEB %d:%d",
1725 zbr->lnum, zbr->offs);
1726 dbg_tnck(&zbr->key, "looked for key ");
1727 dbg_tnck(&key1, "found node's key ");
1736 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1737 ubifs_dump_node(c, buf);
1743 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1744 * @c: UBIFS file-system description object
1745 * @bu: bulk-read parameters and results
1747 * This functions reads and validates the data nodes that were identified by the
1748 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1749 * -EAGAIN to indicate a race with GC, or another negative error code on
1752 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1754 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1755 struct ubifs_wbuf *wbuf;
1758 len = bu->zbranch[bu->cnt - 1].offs;
1759 len += bu->zbranch[bu->cnt - 1].len - offs;
1760 if (len > bu->buf_len) {
1761 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1766 wbuf = ubifs_get_wbuf(c, lnum);
1768 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1770 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1772 /* Check for a race with GC */
1773 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1776 if (err && err != -EBADMSG) {
1777 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1780 dbg_tnck(&bu->key, "key ");
1784 /* Validate the nodes read */
1786 for (i = 0; i < bu->cnt; i++) {
1787 err = validate_data_node(c, buf, &bu->zbranch[i]);
1790 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1797 * do_lookup_nm- look up a "hashed" node.
1798 * @c: UBIFS file-system description object
1799 * @key: node key to lookup
1800 * @node: the node is returned here
1803 * This function looks up and reads a node which contains name hash in the key.
1804 * Since the hash may have collisions, there may be many nodes with the same
1805 * key, so we have to sequentially look to all of them until the needed one is
1806 * found. This function returns zero in case of success, %-ENOENT if the node
1807 * was not found, and a negative error code in case of failure.
1809 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1810 void *node, const struct fscrypt_name *nm)
1813 struct ubifs_znode *znode;
1815 dbg_tnck(key, "key ");
1816 mutex_lock(&c->tnc_mutex);
1817 found = ubifs_lookup_level0(c, key, &znode, &n);
1821 } else if (found < 0) {
1826 ubifs_assert(c, n >= 0);
1828 err = resolve_collision(c, key, &znode, &n, nm);
1829 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1830 if (unlikely(err < 0))
1837 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1840 mutex_unlock(&c->tnc_mutex);
1845 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1846 * @c: UBIFS file-system description object
1847 * @key: node key to lookup
1848 * @node: the node is returned here
1851 * This function looks up and reads a node which contains name hash in the key.
1852 * Since the hash may have collisions, there may be many nodes with the same
1853 * key, so we have to sequentially look to all of them until the needed one is
1854 * found. This function returns zero in case of success, %-ENOENT if the node
1855 * was not found, and a negative error code in case of failure.
1857 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1858 void *node, const struct fscrypt_name *nm)
1861 const struct ubifs_dent_node *dent = node;
1864 * We assume that in most of the cases there are no name collisions and
1865 * 'ubifs_tnc_lookup()' returns us the right direntry.
1867 err = ubifs_tnc_lookup(c, key, node);
1871 len = le16_to_cpu(dent->nlen);
1872 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1876 * Unluckily, there are hash collisions and we have to iterate over
1877 * them look at each direntry with colliding name hash sequentially.
1880 return do_lookup_nm(c, key, node, nm);
1883 static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1884 struct ubifs_dent_node *dent, uint32_t cookie,
1885 struct ubifs_znode **zn, int *n)
1888 struct ubifs_znode *znode = *zn;
1889 struct ubifs_zbranch *zbr;
1890 union ubifs_key *dkey;
1893 zbr = &znode->zbranch[*n];
1896 if (key_inum(c, dkey) != key_inum(c, key) ||
1897 key_type(c, dkey) != key_type(c, key)) {
1901 err = tnc_read_hashed_node(c, zbr, dent);
1905 if (key_hash(c, key) == key_hash(c, dkey) &&
1906 le32_to_cpu(dent->cookie) == cookie) {
1911 err = tnc_next(c, &znode, n);
1917 static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1918 struct ubifs_dent_node *dent, uint32_t cookie)
1921 struct ubifs_znode *znode;
1922 union ubifs_key start_key;
1924 ubifs_assert(c, is_hash_key(c, key));
1926 lowest_dent_key(c, &start_key, key_inum(c, key));
1928 mutex_lock(&c->tnc_mutex);
1929 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1930 if (unlikely(err < 0))
1933 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
1936 mutex_unlock(&c->tnc_mutex);
1941 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1942 * @c: UBIFS file-system description object
1943 * @key: node key to lookup
1944 * @node: the node is returned here
1945 * @cookie: node cookie for collision resolution
1947 * This function looks up and reads a node which contains name hash in the key.
1948 * Since the hash may have collisions, there may be many nodes with the same
1949 * key, so we have to sequentially look to all of them until the needed one
1950 * with the same cookie value is found.
1951 * This function returns zero in case of success, %-ENOENT if the node
1952 * was not found, and a negative error code in case of failure.
1954 int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1955 void *node, uint32_t cookie)
1958 const struct ubifs_dent_node *dent = node;
1960 if (!c->double_hash)
1964 * We assume that in most of the cases there are no name collisions and
1965 * 'ubifs_tnc_lookup()' returns us the right direntry.
1967 err = ubifs_tnc_lookup(c, key, node);
1971 if (le32_to_cpu(dent->cookie) == cookie)
1975 * Unluckily, there are hash collisions and we have to iterate over
1976 * them look at each direntry with colliding name hash sequentially.
1978 return do_lookup_dh(c, key, node, cookie);
1982 * correct_parent_keys - correct parent znodes' keys.
1983 * @c: UBIFS file-system description object
1984 * @znode: znode to correct parent znodes for
1986 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1987 * zbranch changes, keys of parent znodes have to be corrected. This helper
1988 * function is called in such situations and corrects the keys if needed.
1990 static void correct_parent_keys(const struct ubifs_info *c,
1991 struct ubifs_znode *znode)
1993 union ubifs_key *key, *key1;
1995 ubifs_assert(c, znode->parent);
1996 ubifs_assert(c, znode->iip == 0);
1998 key = &znode->zbranch[0].key;
1999 key1 = &znode->parent->zbranch[0].key;
2001 while (keys_cmp(c, key, key1) < 0) {
2002 key_copy(c, key, key1);
2003 znode = znode->parent;
2005 if (!znode->parent || znode->iip)
2007 key1 = &znode->parent->zbranch[0].key;
2012 * insert_zbranch - insert a zbranch into a znode.
2013 * @c: UBIFS file-system description object
2014 * @znode: znode into which to insert
2015 * @zbr: zbranch to insert
2016 * @n: slot number to insert to
2018 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2019 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2020 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2021 * slot, zbranches starting from @n have to be moved right.
2023 static void insert_zbranch(struct ubifs_info *c, struct ubifs_znode *znode,
2024 const struct ubifs_zbranch *zbr, int n)
2028 ubifs_assert(c, ubifs_zn_dirty(znode));
2031 for (i = znode->child_cnt; i > n; i--) {
2032 znode->zbranch[i] = znode->zbranch[i - 1];
2033 if (znode->zbranch[i].znode)
2034 znode->zbranch[i].znode->iip = i;
2037 zbr->znode->iip = n;
2039 for (i = znode->child_cnt; i > n; i--)
2040 znode->zbranch[i] = znode->zbranch[i - 1];
2042 znode->zbranch[n] = *zbr;
2043 znode->child_cnt += 1;
2046 * After inserting at slot zero, the lower bound of the key range of
2047 * this znode may have changed. If this znode is subsequently split
2048 * then the upper bound of the key range may change, and furthermore
2049 * it could change to be lower than the original lower bound. If that
2050 * happens, then it will no longer be possible to find this znode in the
2051 * TNC using the key from the index node on flash. That is bad because
2052 * if it is not found, we will assume it is obsolete and may overwrite
2053 * it. Then if there is an unclean unmount, we will start using the
2054 * old index which will be broken.
2056 * So we first mark znodes that have insertions at slot zero, and then
2057 * if they are split we add their lnum/offs to the old_idx tree.
2064 * tnc_insert - insert a node into TNC.
2065 * @c: UBIFS file-system description object
2066 * @znode: znode to insert into
2067 * @zbr: branch to insert
2068 * @n: slot number to insert new zbranch to
2070 * This function inserts a new node described by @zbr into znode @znode. If
2071 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2072 * are splat as well if needed. Returns zero in case of success or a negative
2073 * error code in case of failure.
2075 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2076 struct ubifs_zbranch *zbr, int n)
2078 struct ubifs_znode *zn, *zi, *zp;
2079 int i, keep, move, appending = 0;
2080 union ubifs_key *key = &zbr->key, *key1;
2082 ubifs_assert(c, n >= 0 && n <= c->fanout);
2084 /* Implement naive insert for now */
2087 if (znode->child_cnt < c->fanout) {
2088 ubifs_assert(c, n != c->fanout);
2089 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2091 insert_zbranch(c, znode, zbr, n);
2093 /* Ensure parent's key is correct */
2094 if (n == 0 && zp && znode->iip == 0)
2095 correct_parent_keys(c, znode);
2101 * Unfortunately, @znode does not have more empty slots and we have to
2104 dbg_tnck(key, "splitting level %d, key ", znode->level);
2108 * We can no longer be sure of finding this znode by key, so we
2109 * record it in the old_idx tree.
2111 ins_clr_old_idx_znode(c, znode);
2113 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2117 zn->level = znode->level;
2119 /* Decide where to split */
2120 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2121 /* Try not to split consecutive data keys */
2122 if (n == c->fanout) {
2123 key1 = &znode->zbranch[n - 1].key;
2124 if (key_inum(c, key1) == key_inum(c, key) &&
2125 key_type(c, key1) == UBIFS_DATA_KEY)
2129 } else if (appending && n != c->fanout) {
2130 /* Try not to split consecutive data keys */
2133 if (n >= (c->fanout + 1) / 2) {
2134 key1 = &znode->zbranch[0].key;
2135 if (key_inum(c, key1) == key_inum(c, key) &&
2136 key_type(c, key1) == UBIFS_DATA_KEY) {
2137 key1 = &znode->zbranch[n].key;
2138 if (key_inum(c, key1) != key_inum(c, key) ||
2139 key_type(c, key1) != UBIFS_DATA_KEY) {
2141 move = c->fanout - keep;
2153 keep = (c->fanout + 1) / 2;
2154 move = c->fanout - keep;
2158 * Although we don't at present, we could look at the neighbors and see
2159 * if we can move some zbranches there.
2163 /* Insert into existing znode */
2168 /* Insert into new znode */
2173 zbr->znode->parent = zn;
2178 __set_bit(DIRTY_ZNODE, &zn->flags);
2179 atomic_long_inc(&c->dirty_zn_cnt);
2181 zn->child_cnt = move;
2182 znode->child_cnt = keep;
2184 dbg_tnc("moving %d, keeping %d", move, keep);
2187 for (i = 0; i < move; i++) {
2188 zn->zbranch[i] = znode->zbranch[keep + i];
2191 if (zn->zbranch[i].znode) {
2192 zn->zbranch[i].znode->parent = zn;
2193 zn->zbranch[i].znode->iip = i;
2197 /* Insert new key and branch */
2198 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2200 insert_zbranch(c, zi, zbr, n);
2202 /* Insert new znode (produced by spitting) into the parent */
2204 if (n == 0 && zi == znode && znode->iip == 0)
2205 correct_parent_keys(c, znode);
2207 /* Locate insertion point */
2210 /* Tail recursion */
2211 zbr->key = zn->zbranch[0].key;
2221 /* We have to split root znode */
2222 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2224 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2229 zi->level = znode->level + 1;
2231 __set_bit(DIRTY_ZNODE, &zi->flags);
2232 atomic_long_inc(&c->dirty_zn_cnt);
2234 zi->zbranch[0].key = znode->zbranch[0].key;
2235 zi->zbranch[0].znode = znode;
2236 zi->zbranch[0].lnum = c->zroot.lnum;
2237 zi->zbranch[0].offs = c->zroot.offs;
2238 zi->zbranch[0].len = c->zroot.len;
2239 zi->zbranch[1].key = zn->zbranch[0].key;
2240 zi->zbranch[1].znode = zn;
2245 c->zroot.znode = zi;
2256 * ubifs_tnc_add - add a node to TNC.
2257 * @c: UBIFS file-system description object
2259 * @lnum: LEB number of node
2260 * @offs: node offset
2262 * @hash: The hash over the node
2264 * This function adds a node with key @key to TNC. The node may be new or it may
2265 * obsolete some existing one. Returns %0 on success or negative error code on
2268 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2269 int offs, int len, const u8 *hash)
2271 int found, n, err = 0;
2272 struct ubifs_znode *znode;
2274 mutex_lock(&c->tnc_mutex);
2275 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2276 found = lookup_level0_dirty(c, key, &znode, &n);
2278 struct ubifs_zbranch zbr;
2284 ubifs_copy_hash(c, hash, zbr.hash);
2285 key_copy(c, key, &zbr.key);
2286 err = tnc_insert(c, znode, &zbr, n + 1);
2287 } else if (found == 1) {
2288 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2291 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2295 ubifs_copy_hash(c, hash, zbr->hash);
2299 err = dbg_check_tnc(c, 0);
2300 mutex_unlock(&c->tnc_mutex);
2306 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2307 * @c: UBIFS file-system description object
2309 * @old_lnum: LEB number of old node
2310 * @old_offs: old node offset
2311 * @lnum: LEB number of node
2312 * @offs: node offset
2315 * This function replaces a node with key @key in the TNC only if the old node
2316 * is found. This function is called by garbage collection when node are moved.
2317 * Returns %0 on success or negative error code on failure.
2319 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2320 int old_lnum, int old_offs, int lnum, int offs, int len)
2322 int found, n, err = 0;
2323 struct ubifs_znode *znode;
2325 mutex_lock(&c->tnc_mutex);
2326 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2327 old_offs, lnum, offs, len);
2328 found = lookup_level0_dirty(c, key, &znode, &n);
2335 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2338 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2340 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2347 } else if (is_hash_key(c, key)) {
2348 found = resolve_collision_directly(c, key, &znode, &n,
2349 old_lnum, old_offs);
2350 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2351 found, znode, n, old_lnum, old_offs);
2358 /* Ensure the znode is dirtied */
2359 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2360 znode = dirty_cow_bottom_up(c, znode);
2361 if (IS_ERR(znode)) {
2362 err = PTR_ERR(znode);
2366 zbr = &znode->zbranch[n];
2368 err = ubifs_add_dirt(c, zbr->lnum,
2380 err = ubifs_add_dirt(c, lnum, len);
2383 err = dbg_check_tnc(c, 0);
2386 mutex_unlock(&c->tnc_mutex);
2391 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2392 * @c: UBIFS file-system description object
2394 * @lnum: LEB number of node
2395 * @offs: node offset
2397 * @hash: The hash over the node
2400 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2401 * may have collisions, like directory entry keys.
2403 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2404 int lnum, int offs, int len, const u8 *hash,
2405 const struct fscrypt_name *nm)
2407 int found, n, err = 0;
2408 struct ubifs_znode *znode;
2410 mutex_lock(&c->tnc_mutex);
2411 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2412 found = lookup_level0_dirty(c, key, &znode, &n);
2420 found = fallible_resolve_collision(c, key, &znode, &n,
2423 found = resolve_collision(c, key, &znode, &n, nm);
2424 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2430 /* Ensure the znode is dirtied */
2431 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2432 znode = dirty_cow_bottom_up(c, znode);
2433 if (IS_ERR(znode)) {
2434 err = PTR_ERR(znode);
2440 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2443 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2447 ubifs_copy_hash(c, hash, zbr->hash);
2453 struct ubifs_zbranch zbr;
2459 ubifs_copy_hash(c, hash, zbr.hash);
2460 key_copy(c, key, &zbr.key);
2461 err = tnc_insert(c, znode, &zbr, n + 1);
2466 * We did not find it in the index so there may be a
2467 * dangling branch still in the index. So we remove it
2468 * by passing 'ubifs_tnc_remove_nm()' the same key but
2469 * an unmatchable name.
2471 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2473 err = dbg_check_tnc(c, 0);
2474 mutex_unlock(&c->tnc_mutex);
2477 return ubifs_tnc_remove_nm(c, key, &noname);
2483 err = dbg_check_tnc(c, 0);
2484 mutex_unlock(&c->tnc_mutex);
2489 * tnc_delete - delete a znode form TNC.
2490 * @c: UBIFS file-system description object
2491 * @znode: znode to delete from
2492 * @n: zbranch slot number to delete
2494 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2495 * case of success and a negative error code in case of failure.
2497 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2499 struct ubifs_zbranch *zbr;
2500 struct ubifs_znode *zp;
2503 /* Delete without merge for now */
2504 ubifs_assert(c, znode->level == 0);
2505 ubifs_assert(c, n >= 0 && n < c->fanout);
2506 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2508 zbr = &znode->zbranch[n];
2511 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2513 ubifs_dump_znode(c, znode);
2517 /* We do not "gap" zbranch slots */
2518 for (i = n; i < znode->child_cnt - 1; i++)
2519 znode->zbranch[i] = znode->zbranch[i + 1];
2520 znode->child_cnt -= 1;
2522 if (znode->child_cnt > 0)
2526 * This was the last zbranch, we have to delete this znode from the
2531 ubifs_assert(c, !ubifs_zn_obsolete(znode));
2532 ubifs_assert(c, ubifs_zn_dirty(znode));
2537 atomic_long_dec(&c->dirty_zn_cnt);
2539 err = insert_old_idx_znode(c, znode);
2544 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2545 atomic_long_inc(&c->clean_zn_cnt);
2546 atomic_long_inc(&ubifs_clean_zn_cnt);
2550 } while (znode->child_cnt == 1); /* while removing last child */
2552 /* Remove from znode, entry n - 1 */
2553 znode->child_cnt -= 1;
2554 ubifs_assert(c, znode->level != 0);
2555 for (i = n; i < znode->child_cnt; i++) {
2556 znode->zbranch[i] = znode->zbranch[i + 1];
2557 if (znode->zbranch[i].znode)
2558 znode->zbranch[i].znode->iip = i;
2562 * If this is the root and it has only 1 child then
2563 * collapse the tree.
2565 if (!znode->parent) {
2566 while (znode->child_cnt == 1 && znode->level != 0) {
2568 zbr = &znode->zbranch[0];
2569 znode = get_znode(c, znode, 0);
2571 return PTR_ERR(znode);
2572 znode = dirty_cow_znode(c, zbr);
2574 return PTR_ERR(znode);
2575 znode->parent = NULL;
2578 err = insert_old_idx(c, c->zroot.lnum,
2583 c->zroot.lnum = zbr->lnum;
2584 c->zroot.offs = zbr->offs;
2585 c->zroot.len = zbr->len;
2586 c->zroot.znode = znode;
2587 ubifs_assert(c, !ubifs_zn_obsolete(zp));
2588 ubifs_assert(c, ubifs_zn_dirty(zp));
2589 atomic_long_dec(&c->dirty_zn_cnt);
2592 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2593 atomic_long_inc(&c->clean_zn_cnt);
2594 atomic_long_inc(&ubifs_clean_zn_cnt);
2604 * ubifs_tnc_remove - remove an index entry of a node.
2605 * @c: UBIFS file-system description object
2608 * Returns %0 on success or negative error code on failure.
2610 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2612 int found, n, err = 0;
2613 struct ubifs_znode *znode;
2615 mutex_lock(&c->tnc_mutex);
2616 dbg_tnck(key, "key ");
2617 found = lookup_level0_dirty(c, key, &znode, &n);
2623 err = tnc_delete(c, znode, n);
2625 err = dbg_check_tnc(c, 0);
2628 mutex_unlock(&c->tnc_mutex);
2633 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2634 * @c: UBIFS file-system description object
2636 * @nm: directory entry name
2638 * Returns %0 on success or negative error code on failure.
2640 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2641 const struct fscrypt_name *nm)
2644 struct ubifs_znode *znode;
2646 mutex_lock(&c->tnc_mutex);
2647 dbg_tnck(key, "key ");
2648 err = lookup_level0_dirty(c, key, &znode, &n);
2654 err = fallible_resolve_collision(c, key, &znode, &n,
2657 err = resolve_collision(c, key, &znode, &n, nm);
2658 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2662 /* Ensure the znode is dirtied */
2663 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2664 znode = dirty_cow_bottom_up(c, znode);
2665 if (IS_ERR(znode)) {
2666 err = PTR_ERR(znode);
2670 err = tnc_delete(c, znode, n);
2676 err = dbg_check_tnc(c, 0);
2677 mutex_unlock(&c->tnc_mutex);
2682 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2683 * @c: UBIFS file-system description object
2685 * @cookie: node cookie for collision resolution
2687 * Returns %0 on success or negative error code on failure.
2689 int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2693 struct ubifs_znode *znode;
2694 struct ubifs_dent_node *dent;
2695 struct ubifs_zbranch *zbr;
2697 if (!c->double_hash)
2700 mutex_lock(&c->tnc_mutex);
2701 err = lookup_level0_dirty(c, key, &znode, &n);
2705 zbr = &znode->zbranch[n];
2706 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2712 err = tnc_read_hashed_node(c, zbr, dent);
2716 /* If the cookie does not match, we're facing a hash collision. */
2717 if (le32_to_cpu(dent->cookie) != cookie) {
2718 union ubifs_key start_key;
2720 lowest_dent_key(c, &start_key, key_inum(c, key));
2722 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2723 if (unlikely(err < 0))
2726 err = search_dh_cookie(c, key, dent, cookie, &znode, &n);
2731 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2732 znode = dirty_cow_bottom_up(c, znode);
2733 if (IS_ERR(znode)) {
2734 err = PTR_ERR(znode);
2738 err = tnc_delete(c, znode, n);
2744 err = dbg_check_tnc(c, 0);
2745 mutex_unlock(&c->tnc_mutex);
2750 * key_in_range - determine if a key falls within a range of keys.
2751 * @c: UBIFS file-system description object
2752 * @key: key to check
2753 * @from_key: lowest key in range
2754 * @to_key: highest key in range
2756 * This function returns %1 if the key is in range and %0 otherwise.
2758 static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2759 union ubifs_key *from_key, union ubifs_key *to_key)
2761 if (keys_cmp(c, key, from_key) < 0)
2763 if (keys_cmp(c, key, to_key) > 0)
2769 * ubifs_tnc_remove_range - remove index entries in range.
2770 * @c: UBIFS file-system description object
2771 * @from_key: lowest key to remove
2772 * @to_key: highest key to remove
2774 * This function removes index entries starting at @from_key and ending at
2775 * @to_key. This function returns zero in case of success and a negative error
2776 * code in case of failure.
2778 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2779 union ubifs_key *to_key)
2781 int i, n, k, err = 0;
2782 struct ubifs_znode *znode;
2783 union ubifs_key *key;
2785 mutex_lock(&c->tnc_mutex);
2787 /* Find first level 0 znode that contains keys to remove */
2788 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2795 err = tnc_next(c, &znode, &n);
2796 if (err == -ENOENT) {
2802 key = &znode->zbranch[n].key;
2803 if (!key_in_range(c, key, from_key, to_key)) {
2809 /* Ensure the znode is dirtied */
2810 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2811 znode = dirty_cow_bottom_up(c, znode);
2812 if (IS_ERR(znode)) {
2813 err = PTR_ERR(znode);
2818 /* Remove all keys in range except the first */
2819 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2820 key = &znode->zbranch[i].key;
2821 if (!key_in_range(c, key, from_key, to_key))
2823 lnc_free(&znode->zbranch[i]);
2824 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2825 znode->zbranch[i].len);
2827 ubifs_dump_znode(c, znode);
2830 dbg_tnck(key, "removing key ");
2833 for (i = n + 1 + k; i < znode->child_cnt; i++)
2834 znode->zbranch[i - k] = znode->zbranch[i];
2835 znode->child_cnt -= k;
2838 /* Now delete the first */
2839 err = tnc_delete(c, znode, n);
2846 err = dbg_check_tnc(c, 0);
2847 mutex_unlock(&c->tnc_mutex);
2852 * ubifs_tnc_remove_ino - remove an inode from TNC.
2853 * @c: UBIFS file-system description object
2854 * @inum: inode number to remove
2856 * This function remove inode @inum and all the extended attributes associated
2857 * with the anode from TNC and returns zero in case of success or a negative
2858 * error code in case of failure.
2860 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2862 union ubifs_key key1, key2;
2863 struct ubifs_dent_node *xent, *pxent = NULL;
2864 struct fscrypt_name nm = {0};
2866 dbg_tnc("ino %lu", (unsigned long)inum);
2869 * Walk all extended attribute entries and remove them together with
2870 * corresponding extended attribute inodes.
2872 lowest_xent_key(c, &key1, inum);
2877 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2879 err = PTR_ERR(xent);
2885 xattr_inum = le64_to_cpu(xent->inum);
2886 dbg_tnc("xent '%s', ino %lu", xent->name,
2887 (unsigned long)xattr_inum);
2889 ubifs_evict_xattr_inode(c, xattr_inum);
2891 fname_name(&nm) = xent->name;
2892 fname_len(&nm) = le16_to_cpu(xent->nlen);
2893 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2899 lowest_ino_key(c, &key1, xattr_inum);
2900 highest_ino_key(c, &key2, xattr_inum);
2901 err = ubifs_tnc_remove_range(c, &key1, &key2);
2909 key_read(c, &xent->key, &key1);
2913 lowest_ino_key(c, &key1, inum);
2914 highest_ino_key(c, &key2, inum);
2916 return ubifs_tnc_remove_range(c, &key1, &key2);
2920 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2921 * @c: UBIFS file-system description object
2922 * @key: key of last entry
2923 * @nm: name of last entry found or %NULL
2925 * This function finds and reads the next directory or extended attribute entry
2926 * after the given key (@key) if there is one. @nm is used to resolve
2929 * If the name of the current entry is not known and only the key is known,
2930 * @nm->name has to be %NULL. In this case the semantics of this function is a
2931 * little bit different and it returns the entry corresponding to this key, not
2932 * the next one. If the key was not found, the closest "right" entry is
2935 * If the fist entry has to be found, @key has to contain the lowest possible
2936 * key value for this inode and @name has to be %NULL.
2938 * This function returns the found directory or extended attribute entry node
2939 * in case of success, %-ENOENT is returned if no entry was found, and a
2940 * negative error code is returned in case of failure.
2942 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2943 union ubifs_key *key,
2944 const struct fscrypt_name *nm)
2946 int n, err, type = key_type(c, key);
2947 struct ubifs_znode *znode;
2948 struct ubifs_dent_node *dent;
2949 struct ubifs_zbranch *zbr;
2950 union ubifs_key *dkey;
2952 dbg_tnck(key, "key ");
2953 ubifs_assert(c, is_hash_key(c, key));
2955 mutex_lock(&c->tnc_mutex);
2956 err = ubifs_lookup_level0(c, key, &znode, &n);
2957 if (unlikely(err < 0))
2960 if (fname_len(nm) > 0) {
2962 /* Handle collisions */
2964 err = fallible_resolve_collision(c, key, &znode, &n,
2967 err = resolve_collision(c, key, &znode, &n, nm);
2968 dbg_tnc("rc returned %d, znode %p, n %d",
2970 if (unlikely(err < 0))
2974 /* Now find next entry */
2975 err = tnc_next(c, &znode, &n);
2980 * The full name of the entry was not given, in which case the
2981 * behavior of this function is a little different and it
2982 * returns current entry, not the next one.
2986 * However, the given key does not exist in the TNC
2987 * tree and @znode/@n variables contain the closest
2988 * "preceding" element. Switch to the next one.
2990 err = tnc_next(c, &znode, &n);
2996 zbr = &znode->zbranch[n];
2997 dent = kmalloc(zbr->len, GFP_NOFS);
2998 if (unlikely(!dent)) {
3004 * The above 'tnc_next()' call could lead us to the next inode, check
3008 if (key_inum(c, dkey) != key_inum(c, key) ||
3009 key_type(c, dkey) != type) {
3014 err = tnc_read_hashed_node(c, zbr, dent);
3018 mutex_unlock(&c->tnc_mutex);
3024 mutex_unlock(&c->tnc_mutex);
3025 return ERR_PTR(err);
3029 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3030 * @c: UBIFS file-system description object
3032 * Destroy left-over obsolete znodes from a failed commit.
3034 static void tnc_destroy_cnext(struct ubifs_info *c)
3036 struct ubifs_znode *cnext;
3040 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3043 struct ubifs_znode *znode = cnext;
3045 cnext = cnext->cnext;
3046 if (ubifs_zn_obsolete(znode))
3048 } while (cnext && cnext != c->cnext);
3052 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3053 * @c: UBIFS file-system description object
3055 void ubifs_tnc_close(struct ubifs_info *c)
3057 tnc_destroy_cnext(c);
3058 if (c->zroot.znode) {
3061 n = atomic_long_read(&c->clean_zn_cnt);
3062 freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode);
3063 ubifs_assert(c, freed == n);
3064 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3072 * left_znode - get the znode to the left.
3073 * @c: UBIFS file-system description object
3076 * This function returns a pointer to the znode to the left of @znode or NULL if
3077 * there is not one. A negative error code is returned on failure.
3079 static struct ubifs_znode *left_znode(struct ubifs_info *c,
3080 struct ubifs_znode *znode)
3082 int level = znode->level;
3085 int n = znode->iip - 1;
3087 /* Go up until we can go left */
3088 znode = znode->parent;
3092 /* Now go down the rightmost branch to 'level' */
3093 znode = get_znode(c, znode, n);
3096 while (znode->level != level) {
3097 n = znode->child_cnt - 1;
3098 znode = get_znode(c, znode, n);
3109 * right_znode - get the znode to the right.
3110 * @c: UBIFS file-system description object
3113 * This function returns a pointer to the znode to the right of @znode or NULL
3114 * if there is not one. A negative error code is returned on failure.
3116 static struct ubifs_znode *right_znode(struct ubifs_info *c,
3117 struct ubifs_znode *znode)
3119 int level = znode->level;
3122 int n = znode->iip + 1;
3124 /* Go up until we can go right */
3125 znode = znode->parent;
3128 if (n < znode->child_cnt) {
3129 /* Now go down the leftmost branch to 'level' */
3130 znode = get_znode(c, znode, n);
3133 while (znode->level != level) {
3134 znode = get_znode(c, znode, 0);
3145 * lookup_znode - find a particular indexing node from TNC.
3146 * @c: UBIFS file-system description object
3147 * @key: index node key to lookup
3148 * @level: index node level
3149 * @lnum: index node LEB number
3150 * @offs: index node offset
3152 * This function searches an indexing node by its first key @key and its
3153 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3154 * nodes it traverses to TNC. This function is called for indexing nodes which
3155 * were found on the media by scanning, for example when garbage-collecting or
3156 * when doing in-the-gaps commit. This means that the indexing node which is
3157 * looked for does not have to have exactly the same leftmost key @key, because
3158 * the leftmost key may have been changed, in which case TNC will contain a
3159 * dirty znode which still refers the same @lnum:@offs. This function is clever
3160 * enough to recognize such indexing nodes.
3162 * Note, if a znode was deleted or changed too much, then this function will
3163 * not find it. For situations like this UBIFS has the old index RB-tree
3164 * (indexed by @lnum:@offs).
3166 * This function returns a pointer to the znode found or %NULL if it is not
3167 * found. A negative error code is returned on failure.
3169 static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3170 union ubifs_key *key, int level,
3173 struct ubifs_znode *znode, *zn;
3176 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3179 * The arguments have probably been read off flash, so don't assume
3183 return ERR_PTR(-EINVAL);
3185 /* Get the root znode */
3186 znode = c->zroot.znode;
3188 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3192 /* Check if it is the one we are looking for */
3193 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3195 /* Descend to the parent level i.e. (level + 1) */
3196 if (level >= znode->level)
3199 ubifs_search_zbranch(c, znode, key, &n);
3202 * We reached a znode where the leftmost key is greater
3203 * than the key we are searching for. This is the same
3204 * situation as the one described in a huge comment at
3205 * the end of the 'ubifs_lookup_level0()' function. And
3206 * for exactly the same reasons we have to try to look
3207 * left before giving up.
3209 znode = left_znode(c, znode);
3214 ubifs_search_zbranch(c, znode, key, &n);
3215 ubifs_assert(c, n >= 0);
3217 if (znode->level == level + 1)
3219 znode = get_znode(c, znode, n);
3223 /* Check if the child is the one we are looking for */
3224 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3225 return get_znode(c, znode, n);
3226 /* If the key is unique, there is nowhere else to look */
3227 if (!is_hash_key(c, key))
3230 * The key is not unique and so may be also in the znodes to either
3237 /* Move one branch to the left */
3241 znode = left_znode(c, znode);
3246 n = znode->child_cnt - 1;
3249 if (znode->zbranch[n].lnum == lnum &&
3250 znode->zbranch[n].offs == offs)
3251 return get_znode(c, znode, n);
3252 /* Stop if the key is less than the one we are looking for */
3253 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3256 /* Back to the middle */
3261 /* Move one branch to the right */
3262 if (++n >= znode->child_cnt) {
3263 znode = right_znode(c, znode);
3271 if (znode->zbranch[n].lnum == lnum &&
3272 znode->zbranch[n].offs == offs)
3273 return get_znode(c, znode, n);
3274 /* Stop if the key is greater than the one we are looking for */
3275 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3282 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3283 * @c: UBIFS file-system description object
3284 * @key: key of index node
3285 * @level: index node level
3286 * @lnum: LEB number of index node
3287 * @offs: offset of index node
3289 * This function returns %0 if the index node is not referred to in the TNC, %1
3290 * if the index node is referred to in the TNC and the corresponding znode is
3291 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3292 * znode is clean, and a negative error code in case of failure.
3294 * Note, the @key argument has to be the key of the first child. Also note,
3295 * this function relies on the fact that 0:0 is never a valid LEB number and
3296 * offset for a main-area node.
3298 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3301 struct ubifs_znode *znode;
3303 znode = lookup_znode(c, key, level, lnum, offs);
3307 return PTR_ERR(znode);
3309 return ubifs_zn_dirty(znode) ? 1 : 2;
3313 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3314 * @c: UBIFS file-system description object
3316 * @lnum: node LEB number
3317 * @offs: node offset
3319 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3320 * not, and a negative error code in case of failure.
3322 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3323 * and offset for a main-area node.
3325 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3328 struct ubifs_zbranch *zbr;
3329 struct ubifs_znode *znode, *zn;
3330 int n, found, err, nn;
3331 const int unique = !is_hash_key(c, key);
3333 found = ubifs_lookup_level0(c, key, &znode, &n);
3335 return found; /* Error code */
3338 zbr = &znode->zbranch[n];
3339 if (lnum == zbr->lnum && offs == zbr->offs)
3340 return 1; /* Found it */
3344 * Because the key is not unique, we have to look left
3351 err = tnc_prev(c, &znode, &n);
3356 if (keys_cmp(c, key, &znode->zbranch[n].key))
3358 zbr = &znode->zbranch[n];
3359 if (lnum == zbr->lnum && offs == zbr->offs)
3360 return 1; /* Found it */
3366 err = tnc_next(c, &znode, &n);
3372 if (keys_cmp(c, key, &znode->zbranch[n].key))
3374 zbr = &znode->zbranch[n];
3375 if (lnum == zbr->lnum && offs == zbr->offs)
3376 return 1; /* Found it */
3382 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3383 * @c: UBIFS file-system description object
3385 * @level: index node level (if it is an index node)
3386 * @lnum: node LEB number
3387 * @offs: node offset
3388 * @is_idx: non-zero if the node is an index node
3390 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3391 * negative error code in case of failure. For index nodes, @key has to be the
3392 * key of the first child. An index node is considered to be in the TNC only if
3393 * the corresponding znode is clean or has not been loaded.
3395 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3396 int lnum, int offs, int is_idx)
3400 mutex_lock(&c->tnc_mutex);
3402 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3406 /* The index node was found but it was dirty */
3409 /* The index node was found and it was clean */
3414 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3417 mutex_unlock(&c->tnc_mutex);
3422 * ubifs_dirty_idx_node - dirty an index node.
3423 * @c: UBIFS file-system description object
3424 * @key: index node key
3425 * @level: index node level
3426 * @lnum: index node LEB number
3427 * @offs: index node offset
3429 * This function loads and dirties an index node so that it can be garbage
3430 * collected. The @key argument has to be the key of the first child. This
3431 * function relies on the fact that 0:0 is never a valid LEB number and offset
3432 * for a main-area node. Returns %0 on success and a negative error code on
3435 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3438 struct ubifs_znode *znode;
3441 mutex_lock(&c->tnc_mutex);
3442 znode = lookup_znode(c, key, level, lnum, offs);
3445 if (IS_ERR(znode)) {
3446 err = PTR_ERR(znode);
3449 znode = dirty_cow_bottom_up(c, znode);
3450 if (IS_ERR(znode)) {
3451 err = PTR_ERR(znode);
3456 mutex_unlock(&c->tnc_mutex);
3461 * dbg_check_inode_size - check if inode size is correct.
3462 * @c: UBIFS file-system description object
3463 * @inum: inode number
3466 * This function makes sure that the inode size (@size) is correct and it does
3467 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3468 * if it has a data page beyond @size, and other negative error code in case of
3471 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3475 union ubifs_key from_key, to_key, *key;
3476 struct ubifs_znode *znode;
3479 if (!S_ISREG(inode->i_mode))
3481 if (!dbg_is_chk_gen(c))
3484 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3485 data_key_init(c, &from_key, inode->i_ino, block);
3486 highest_data_key(c, &to_key, inode->i_ino);
3488 mutex_lock(&c->tnc_mutex);
3489 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3498 err = tnc_next(c, &znode, &n);
3499 if (err == -ENOENT) {
3506 ubifs_assert(c, err == 0);
3507 key = &znode->zbranch[n].key;
3508 if (!key_in_range(c, key, &from_key, &to_key))
3512 block = key_block(c, key);
3513 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3514 (unsigned long)inode->i_ino, size,
3515 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3516 mutex_unlock(&c->tnc_mutex);
3517 ubifs_dump_inode(c, inode);
3522 mutex_unlock(&c->tnc_mutex);