Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 *
7 * Authors: Adrian Hunter
8 * Artem Bityutskiy (Битюцкий Артём)
9 */
10
11/*
12 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
13 * the UBIFS B-tree.
14 *
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
18 * the mutex locked.
19 */
20
21#include <linux/crc32.h>
22#include <linux/slab.h>
23#include "ubifs.h"
24
25static int try_read_node(const struct ubifs_info *c, void *buf, int type,
26 struct ubifs_zbranch *zbr);
27static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
28 struct ubifs_zbranch *zbr, void *node);
29
30/*
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
35 * first
36 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
37 *
38 * These constants were introduce to improve readability.
39 */
40enum {
41 NAME_LESS = 0,
42 NAME_MATCHES = 1,
43 NAME_GREATER = 2,
44 NOT_ON_MEDIA = 3,
45};
46
47/**
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
52 *
53 * Returns %0 on success, and a negative error code on failure.
54 *
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).
60 *
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.
69 */
70static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
71{
72 struct ubifs_old_idx *old_idx, *o;
73 struct rb_node **p, *parent = NULL;
74
75 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
76 if (unlikely(!old_idx))
77 return -ENOMEM;
78 old_idx->lnum = lnum;
79 old_idx->offs = offs;
80
81 p = &c->old_idx.rb_node;
82 while (*p) {
83 parent = *p;
84 o = rb_entry(parent, struct ubifs_old_idx, rb);
85 if (lnum < o->lnum)
86 p = &(*p)->rb_left;
87 else if (lnum > o->lnum)
88 p = &(*p)->rb_right;
89 else if (offs < o->offs)
90 p = &(*p)->rb_left;
91 else if (offs > o->offs)
92 p = &(*p)->rb_right;
93 else {
94 ubifs_err(c, "old idx added twice!");
95 kfree(old_idx);
96 return 0;
97 }
98 }
99 rb_link_node(&old_idx->rb, parent, p);
100 rb_insert_color(&old_idx->rb, &c->old_idx);
101 return 0;
102}
103
104/**
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
108 *
109 * Returns %0 on success, and a negative error code on failure.
110 */
111int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
112{
113 if (znode->parent) {
114 struct ubifs_zbranch *zbr;
115
116 zbr = &znode->parent->zbranch[znode->iip];
117 if (zbr->len)
118 return insert_old_idx(c, zbr->lnum, zbr->offs);
119 } else
120 if (c->zroot.len)
121 return insert_old_idx(c, c->zroot.lnum,
122 c->zroot.offs);
123 return 0;
124}
125
126/**
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
130 *
131 * Returns %0 on success, and a negative error code on failure.
132 */
133static int ins_clr_old_idx_znode(struct ubifs_info *c,
134 struct ubifs_znode *znode)
135{
136 int err;
137
138 if (znode->parent) {
139 struct ubifs_zbranch *zbr;
140
141 zbr = &znode->parent->zbranch[znode->iip];
142 if (zbr->len) {
143 err = insert_old_idx(c, zbr->lnum, zbr->offs);
144 if (err)
145 return err;
146 zbr->lnum = 0;
147 zbr->offs = 0;
148 zbr->len = 0;
149 }
150 } else
151 if (c->zroot.len) {
152 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
153 if (err)
154 return err;
155 c->zroot.lnum = 0;
156 c->zroot.offs = 0;
157 c->zroot.len = 0;
158 }
159 return 0;
160}
161
162/**
163 * destroy_old_idx - destroy the old_idx RB-tree.
164 * @c: UBIFS file-system description object
165 *
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.
171 */
172void destroy_old_idx(struct ubifs_info *c)
173{
174 struct ubifs_old_idx *old_idx, *n;
175
176 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
177 kfree(old_idx);
178
179 c->old_idx = RB_ROOT;
180}
181
182/**
183 * copy_znode - copy a dirty znode.
184 * @c: UBIFS file-system description object
185 * @znode: znode to copy
186 *
187 * A dirty znode being committed may not be changed, so it is copied.
188 */
189static struct ubifs_znode *copy_znode(struct ubifs_info *c,
190 struct ubifs_znode *znode)
191{
192 struct ubifs_znode *zn;
193
194 zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS);
195 if (unlikely(!zn))
196 return ERR_PTR(-ENOMEM);
197
198 zn->cnext = NULL;
199 __set_bit(DIRTY_ZNODE, &zn->flags);
200 __clear_bit(COW_ZNODE, &zn->flags);
201
202 ubifs_assert(c, !ubifs_zn_obsolete(znode));
203 __set_bit(OBSOLETE_ZNODE, &znode->flags);
204
205 if (znode->level != 0) {
206 int i;
207 const int n = zn->child_cnt;
208
209 /* The children now have new parent */
210 for (i = 0; i < n; i++) {
211 struct ubifs_zbranch *zbr = &zn->zbranch[i];
212
213 if (zbr->znode)
214 zbr->znode->parent = zn;
215 }
216 }
217
218 atomic_long_inc(&c->dirty_zn_cnt);
219 return zn;
220}
221
222/**
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
227 *
228 * This function updates lprops dirty space and the new size of the index.
229 */
230static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
231{
232 c->calc_idx_sz -= ALIGN(dirt, 8);
233 return ubifs_add_dirt(c, lnum, dirt);
234}
235
236/**
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
240 *
241 * Returns dirtied znode on success or negative error code on failure.
242 */
243static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
244 struct ubifs_zbranch *zbr)
245{
246 struct ubifs_znode *znode = zbr->znode;
247 struct ubifs_znode *zn;
248 int err;
249
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);
257 if (unlikely(err))
258 return ERR_PTR(err);
259 }
260 return znode;
261 }
262
263 zn = copy_znode(c, znode);
264 if (IS_ERR(zn))
265 return zn;
266
267 if (zbr->len) {
268 err = insert_old_idx(c, zbr->lnum, zbr->offs);
269 if (unlikely(err))
270 return ERR_PTR(err);
271 err = add_idx_dirt(c, zbr->lnum, zbr->len);
272 } else
273 err = 0;
274
275 zbr->znode = zn;
276 zbr->lnum = 0;
277 zbr->offs = 0;
278 zbr->len = 0;
279
280 if (unlikely(err))
281 return ERR_PTR(err);
282 return zn;
283}
284
285/**
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
289 * @node: leaf node
290 *
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.
298 *
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.
304 */
305static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
306 const void *node)
307{
308 int err;
309 void *lnc_node;
310 const struct ubifs_dent_node *dent = node;
311
312 ubifs_assert(c, !zbr->leaf);
313 ubifs_assert(c, zbr->len != 0);
314 ubifs_assert(c, is_hash_key(c, &zbr->key));
315
316 err = ubifs_validate_entry(c, dent);
317 if (err) {
318 dump_stack();
319 ubifs_dump_node(c, dent);
320 return err;
321 }
322
323 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
324 if (!lnc_node)
325 /* We don't have to have the cache, so no error */
326 return 0;
327
328 zbr->leaf = lnc_node;
329 return 0;
330}
331
332 /**
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
336 * @node: leaf node
337 *
338 * This function is similar to 'lnc_add()', but it does not create a copy of
339 * @node but inserts @node to TNC directly.
340 */
341static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
342 void *node)
343{
344 int err;
345
346 ubifs_assert(c, !zbr->leaf);
347 ubifs_assert(c, zbr->len != 0);
348
349 err = ubifs_validate_entry(c, node);
350 if (err) {
351 dump_stack();
352 ubifs_dump_node(c, node);
353 return err;
354 }
355
356 zbr->leaf = node;
357 return 0;
358}
359
360/**
361 * lnc_free - remove a leaf node from the leaf node cache.
362 * @zbr: zbranch of leaf node
363 * @node: leaf node
364 */
365static void lnc_free(struct ubifs_zbranch *zbr)
366{
367 if (!zbr->leaf)
368 return;
369 kfree(zbr->leaf);
370 zbr->leaf = NULL;
371}
372
373/**
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
378 *
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.
383 */
384static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr,
385 void *node)
386{
387 int err;
388
389 ubifs_assert(c, is_hash_key(c, &zbr->key));
390
391 if (zbr->leaf) {
392 /* Read from the leaf node cache */
393 ubifs_assert(c, zbr->len != 0);
394 memcpy(node, zbr->leaf, zbr->len);
395 return 0;
396 }
397
398 if (c->replaying) {
399 err = fallible_read_node(c, &zbr->key, zbr, node);
400 /*
401 * When the node was not found, return -ENOENT, 0 otherwise.
402 * Negative return codes stay as-is.
403 */
404 if (err == 0)
405 err = -ENOENT;
406 else if (err == 1)
407 err = 0;
408 } else {
409 err = ubifs_tnc_read_node(c, zbr, node);
410 }
411 if (err)
412 return err;
413
414 /* Add the node to the leaf node cache */
415 err = lnc_add(c, zbr, node);
416 return err;
417}
418
419/**
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
423 * @type: node type
424 * @zbr: the zbranch describing the node to read
425 *
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.
432 *
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.
440 */
441static int try_read_node(const struct ubifs_info *c, void *buf, int type,
442 struct ubifs_zbranch *zbr)
443{
444 int len = zbr->len;
445 int lnum = zbr->lnum;
446 int offs = zbr->offs;
447 int err, node_len;
448 struct ubifs_ch *ch = buf;
449 uint32_t crc, node_crc;
450
451 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
452
453 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
454 if (err) {
455 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
456 type, lnum, offs, err);
457 return err;
458 }
459
460 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
461 return 0;
462
463 if (ch->node_type != type)
464 return 0;
465
466 node_len = le32_to_cpu(ch->len);
467 if (node_len != len)
468 return 0;
469
470 if (type != UBIFS_DATA_NODE || !c->no_chk_data_crc || c->mounting ||
471 c->remounting_rw) {
472 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
473 node_crc = le32_to_cpu(ch->crc);
474 if (crc != node_crc)
475 return 0;
476 }
477
478 err = ubifs_node_check_hash(c, buf, zbr->hash);
479 if (err) {
480 ubifs_bad_hash(c, buf, zbr->hash, lnum, offs);
481 return 0;
482 }
483
484 return 1;
485}
486
487/**
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
493 *
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.
496 */
497static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
498 struct ubifs_zbranch *zbr, void *node)
499{
500 int ret;
501
502 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
503
504 ret = try_read_node(c, node, key_type(c, key), zbr);
505 if (ret == 1) {
506 union ubifs_key node_key;
507 struct ubifs_dent_node *dent = node;
508
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)
512 ret = 0;
513 }
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);
517 return ret;
518}
519
520/**
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
524 * @nm: name to match
525 *
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.
530 */
531static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
532 const struct fscrypt_name *nm)
533{
534 struct ubifs_dent_node *dent;
535 int nlen, err;
536
537 /* If possible, match against the dent in the leaf node cache */
538 if (!zbr->leaf) {
539 dent = kmalloc(zbr->len, GFP_NOFS);
540 if (!dent)
541 return -ENOMEM;
542
543 err = ubifs_tnc_read_node(c, zbr, dent);
544 if (err)
545 goto out_free;
546
547 /* Add the node to the leaf node cache */
548 err = lnc_add_directly(c, zbr, dent);
549 if (err)
550 goto out_free;
551 } else
552 dent = zbr->leaf;
553
554 nlen = le16_to_cpu(dent->nlen);
555 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
556 if (err == 0) {
557 if (nlen == fname_len(nm))
558 return NAME_MATCHES;
559 else if (nlen < fname_len(nm))
560 return NAME_LESS;
561 else
562 return NAME_GREATER;
563 } else if (err < 0)
564 return NAME_LESS;
565 else
566 return NAME_GREATER;
567
568out_free:
569 kfree(dent);
570 return err;
571}
572
573/**
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
578 *
579 * This function returns the znode or a negative error code.
580 */
581static struct ubifs_znode *get_znode(struct ubifs_info *c,
582 struct ubifs_znode *znode, int n)
583{
584 struct ubifs_zbranch *zbr;
585
586 zbr = &znode->zbranch[n];
587 if (zbr->znode)
588 znode = zbr->znode;
589 else
590 znode = ubifs_load_znode(c, zbr, znode, n);
591 return znode;
592}
593
594/**
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
599 *
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.
602 */
603static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
604{
605 struct ubifs_znode *znode = *zn;
606 int nn = *n;
607
608 nn += 1;
609 if (nn < znode->child_cnt) {
610 *n = nn;
611 return 0;
612 }
613 while (1) {
614 struct ubifs_znode *zp;
615
616 zp = znode->parent;
617 if (!zp)
618 return -ENOENT;
619 nn = znode->iip + 1;
620 znode = zp;
621 if (nn < znode->child_cnt) {
622 znode = get_znode(c, znode, nn);
623 if (IS_ERR(znode))
624 return PTR_ERR(znode);
625 while (znode->level != 0) {
626 znode = get_znode(c, znode, 0);
627 if (IS_ERR(znode))
628 return PTR_ERR(znode);
629 }
630 nn = 0;
631 break;
632 }
633 }
634 *zn = znode;
635 *n = nn;
636 return 0;
637}
638
639/**
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
644 *
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.
647 */
648static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
649{
650 struct ubifs_znode *znode = *zn;
651 int nn = *n;
652
653 if (nn > 0) {
654 *n = nn - 1;
655 return 0;
656 }
657 while (1) {
658 struct ubifs_znode *zp;
659
660 zp = znode->parent;
661 if (!zp)
662 return -ENOENT;
663 nn = znode->iip - 1;
664 znode = zp;
665 if (nn >= 0) {
666 znode = get_znode(c, znode, nn);
667 if (IS_ERR(znode))
668 return PTR_ERR(znode);
669 while (znode->level != 0) {
670 nn = znode->child_cnt - 1;
671 znode = get_znode(c, znode, nn);
672 if (IS_ERR(znode))
673 return PTR_ERR(znode);
674 }
675 nn = znode->child_cnt - 1;
676 break;
677 }
678 }
679 *zn = znode;
680 *n = nn;
681 return 0;
682}
683
684/**
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
691 *
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.
699 */
700static 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)
703{
704 int err;
705
706 err = matches_name(c, &(*zn)->zbranch[*n], nm);
707 if (unlikely(err < 0))
708 return err;
709 if (err == NAME_MATCHES)
710 return 1;
711
712 if (err == NAME_GREATER) {
713 /* Look left */
714 while (1) {
715 err = tnc_prev(c, zn, n);
716 if (err == -ENOENT) {
717 ubifs_assert(c, *n == 0);
718 *n = -1;
719 return 0;
720 }
721 if (err < 0)
722 return err;
723 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
724 /*
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.
730 *
731 * znode zp
732 * ----------------------
733 * level 1 | Key0 | Key1 |
734 * -----------------------
735 * | |
736 * znode za | | znode zb
737 * ------------ ------------
738 * level 0 | Key0 | | Key2 |
739 * ------------ ------------
740 *
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.
748 *
749 * Note, this problem is not relevant for the
750 * case when we go right, because
751 * 'tnc_insert()' would correct the parent key.
752 */
753 if (*n == (*zn)->child_cnt - 1) {
754 err = tnc_next(c, zn, n);
755 if (err) {
756 /* Should be impossible */
757 ubifs_assert(c, 0);
758 if (err == -ENOENT)
759 err = -EINVAL;
760 return err;
761 }
762 ubifs_assert(c, *n == 0);
763 *n = -1;
764 }
765 return 0;
766 }
767 err = matches_name(c, &(*zn)->zbranch[*n], nm);
768 if (err < 0)
769 return err;
770 if (err == NAME_LESS)
771 return 0;
772 if (err == NAME_MATCHES)
773 return 1;
774 ubifs_assert(c, err == NAME_GREATER);
775 }
776 } else {
777 int nn = *n;
778 struct ubifs_znode *znode = *zn;
779
780 /* Look right */
781 while (1) {
782 err = tnc_next(c, &znode, &nn);
783 if (err == -ENOENT)
784 return 0;
785 if (err < 0)
786 return err;
787 if (keys_cmp(c, &znode->zbranch[nn].key, key))
788 return 0;
789 err = matches_name(c, &znode->zbranch[nn], nm);
790 if (err < 0)
791 return err;
792 if (err == NAME_GREATER)
793 return 0;
794 *zn = znode;
795 *n = nn;
796 if (err == NAME_MATCHES)
797 return 1;
798 ubifs_assert(c, err == NAME_LESS);
799 }
800 }
801}
802
803/**
804 * fallible_matches_name - determine if a dent matches a given name.
805 * @c: UBIFS file-system description object
806 * @zbr: zbranch of dent
807 * @nm: name to match
808 *
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.
811 *
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.
817 */
818static int fallible_matches_name(struct ubifs_info *c,
819 struct ubifs_zbranch *zbr,
820 const struct fscrypt_name *nm)
821{
822 struct ubifs_dent_node *dent;
823 int nlen, err;
824
825 /* If possible, match against the dent in the leaf node cache */
826 if (!zbr->leaf) {
827 dent = kmalloc(zbr->len, GFP_NOFS);
828 if (!dent)
829 return -ENOMEM;
830
831 err = fallible_read_node(c, &zbr->key, zbr, dent);
832 if (err < 0)
833 goto out_free;
834 if (err == 0) {
835 /* The node was not present */
836 err = NOT_ON_MEDIA;
837 goto out_free;
838 }
839 ubifs_assert(c, err == 1);
840
841 err = lnc_add_directly(c, zbr, dent);
842 if (err)
843 goto out_free;
844 } else
845 dent = zbr->leaf;
846
847 nlen = le16_to_cpu(dent->nlen);
848 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
849 if (err == 0) {
850 if (nlen == fname_len(nm))
851 return NAME_MATCHES;
852 else if (nlen < fname_len(nm))
853 return NAME_LESS;
854 else
855 return NAME_GREATER;
856 } else if (err < 0)
857 return NAME_LESS;
858 else
859 return NAME_GREATER;
860
861out_free:
862 kfree(dent);
863 return err;
864}
865
866/**
867 * fallible_resolve_collision - resolve a collision even if nodes are missing.
868 * @c: UBIFS file-system description object
869 * @key: key
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
874 *
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
882 * branch;
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.
887 */
888static 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,
892 int adding)
893{
894 struct ubifs_znode *o_znode = NULL, *znode = *zn;
895 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
896
897 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
898 if (unlikely(cmp < 0))
899 return cmp;
900 if (cmp == NAME_MATCHES)
901 return 1;
902 if (cmp == NOT_ON_MEDIA) {
903 o_znode = znode;
904 o_n = nn;
905 /*
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.
909 */
910 unsure = 1;
911 } else if (!adding)
912 unsure = 1; /* Remove a dangling branch wherever it is */
913
914 if (cmp == NAME_GREATER || unsure) {
915 /* Look left */
916 while (1) {
917 err = tnc_prev(c, zn, n);
918 if (err == -ENOENT) {
919 ubifs_assert(c, *n == 0);
920 *n = -1;
921 break;
922 }
923 if (err < 0)
924 return err;
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);
929 if (err) {
930 /* Should be impossible */
931 ubifs_assert(c, 0);
932 if (err == -ENOENT)
933 err = -EINVAL;
934 return err;
935 }
936 ubifs_assert(c, *n == 0);
937 *n = -1;
938 }
939 break;
940 }
941 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
942 if (err < 0)
943 return err;
944 if (err == NAME_MATCHES)
945 return 1;
946 if (err == NOT_ON_MEDIA) {
947 o_znode = *zn;
948 o_n = *n;
949 continue;
950 }
951 if (!adding)
952 continue;
953 if (err == NAME_LESS)
954 break;
955 else
956 unsure = 0;
957 }
958 }
959
960 if (cmp == NAME_LESS || unsure) {
961 /* Look right */
962 *zn = znode;
963 *n = nn;
964 while (1) {
965 err = tnc_next(c, &znode, &nn);
966 if (err == -ENOENT)
967 break;
968 if (err < 0)
969 return err;
970 if (keys_cmp(c, &znode->zbranch[nn].key, key))
971 break;
972 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
973 if (err < 0)
974 return err;
975 if (err == NAME_GREATER)
976 break;
977 *zn = znode;
978 *n = nn;
979 if (err == NAME_MATCHES)
980 return 1;
981 if (err == NOT_ON_MEDIA) {
982 o_znode = znode;
983 o_n = nn;
984 }
985 }
986 }
987
988 /* Never match a dangling branch when adding */
989 if (adding || !o_znode)
990 return 0;
991
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);
995 *zn = o_znode;
996 *n = o_n;
997 return 1;
998}
999
1000/**
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
1005 *
1006 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1007 */
1008static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1009{
1010 if (zbr->lnum == lnum && zbr->offs == offs)
1011 return 1;
1012 else
1013 return 0;
1014}
1015
1016/**
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
1024 *
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.
1032 */
1033static int resolve_collision_directly(struct ubifs_info *c,
1034 const union ubifs_key *key,
1035 struct ubifs_znode **zn, int *n,
1036 int lnum, int offs)
1037{
1038 struct ubifs_znode *znode;
1039 int nn, err;
1040
1041 znode = *zn;
1042 nn = *n;
1043 if (matches_position(&znode->zbranch[nn], lnum, offs))
1044 return 1;
1045
1046 /* Look left */
1047 while (1) {
1048 err = tnc_prev(c, &znode, &nn);
1049 if (err == -ENOENT)
1050 break;
1051 if (err < 0)
1052 return err;
1053 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1054 break;
1055 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1056 *zn = znode;
1057 *n = nn;
1058 return 1;
1059 }
1060 }
1061
1062 /* Look right */
1063 znode = *zn;
1064 nn = *n;
1065 while (1) {
1066 err = tnc_next(c, &znode, &nn);
1067 if (err == -ENOENT)
1068 return 0;
1069 if (err < 0)
1070 return err;
1071 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1072 return 0;
1073 *zn = znode;
1074 *n = nn;
1075 if (matches_position(&znode->zbranch[nn], lnum, offs))
1076 return 1;
1077 }
1078}
1079
1080/**
1081 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1082 * @c: UBIFS file-system description object
1083 * @znode: znode to dirty
1084 *
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.
1089 */
1090static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1091 struct ubifs_znode *znode)
1092{
1093 struct ubifs_znode *zp;
1094 int *path = c->bottom_up_buf, p = 0;
1095
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,
1101 sizeof(int),
1102 GFP_NOFS);
1103 if (!c->bottom_up_buf)
1104 return ERR_PTR(-ENOMEM);
1105 path = c->bottom_up_buf;
1106 }
1107 if (c->zroot.znode->level) {
1108 /* Go up until parent is dirty */
1109 while (1) {
1110 int n;
1111
1112 zp = znode->parent;
1113 if (!zp)
1114 break;
1115 n = znode->iip;
1116 ubifs_assert(c, p < c->zroot.znode->level);
1117 path[p++] = n;
1118 if (!zp->cnext && ubifs_zn_dirty(znode))
1119 break;
1120 znode = zp;
1121 }
1122 }
1123
1124 /* Come back down, dirtying as we go */
1125 while (1) {
1126 struct ubifs_zbranch *zbr;
1127
1128 zp = znode->parent;
1129 if (zp) {
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);
1134 } else {
1135 ubifs_assert(c, znode == c->zroot.znode);
1136 znode = dirty_cow_znode(c, &c->zroot);
1137 }
1138 if (IS_ERR(znode) || !p)
1139 break;
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;
1143 }
1144
1145 return znode;
1146}
1147
1148/**
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
1154 *
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
1157 * cases:
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 or %-1
1162 * is stored in @n; In this case calling tnc_next() is mandatory.
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.
1165 *
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.
1169 */
1170int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1171 struct ubifs_znode **zn, int *n)
1172{
1173 int err, exact;
1174 struct ubifs_znode *znode;
1175 time64_t time = ktime_get_seconds();
1176
1177 dbg_tnck(key, "search key ");
1178 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
1179
1180 znode = c->zroot.znode;
1181 if (unlikely(!znode)) {
1182 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1183 if (IS_ERR(znode))
1184 return PTR_ERR(znode);
1185 }
1186
1187 znode->time = time;
1188
1189 while (1) {
1190 struct ubifs_zbranch *zbr;
1191
1192 exact = ubifs_search_zbranch(c, znode, key, n);
1193
1194 if (znode->level == 0)
1195 break;
1196
1197 if (*n < 0)
1198 *n = 0;
1199 zbr = &znode->zbranch[*n];
1200
1201 if (zbr->znode) {
1202 znode->time = time;
1203 znode = zbr->znode;
1204 continue;
1205 }
1206
1207 /* znode is not in TNC cache, load it from the media */
1208 znode = ubifs_load_znode(c, zbr, znode, *n);
1209 if (IS_ERR(znode))
1210 return PTR_ERR(znode);
1211 }
1212
1213 *zn = znode;
1214 if (exact || !is_hash_key(c, key) || *n != -1) {
1215 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1216 return exact;
1217 }
1218
1219 /*
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:
1223 *
1224 * | 3 | 5 |
1225 * / \
1226 * | 3 | 5 | | 6 | 7 | (x)
1227 *
1228 * Or more a complex example:
1229 *
1230 * | 1 | 5 |
1231 * / \
1232 * | 1 | 3 | | 5 | 8 |
1233 * \ /
1234 * | 5 | 5 | | 6 | 7 | (x)
1235 *
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
1239 * return it.
1240 *
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
1245 * like this:
1246 * | 3 | 5 |
1247 * / \
1248 * | 3 | 5 | | 5 | 6 | 7 |
1249 * ^
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.
1261 */
1262 err = tnc_prev(c, &znode, n);
1263 if (err == -ENOENT) {
1264 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1265 *n = -1;
1266 return 0;
1267 }
1268 if (unlikely(err < 0))
1269 return err;
1270 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1271 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1272 *n = -1;
1273 return 0;
1274 }
1275
1276 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1277 *zn = znode;
1278 return 1;
1279}
1280
1281/**
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
1287 *
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
1290 * cases:
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
1295 * @n;
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.
1298 *
1299 * Additionally all znodes in the path from the root to the located zero-level
1300 * znode are marked as dirty.
1301 *
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.
1305 */
1306static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1307 struct ubifs_znode **zn, int *n)
1308{
1309 int err, exact;
1310 struct ubifs_znode *znode;
1311 time64_t time = ktime_get_seconds();
1312
1313 dbg_tnck(key, "search and dirty key ");
1314
1315 znode = c->zroot.znode;
1316 if (unlikely(!znode)) {
1317 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1318 if (IS_ERR(znode))
1319 return PTR_ERR(znode);
1320 }
1321
1322 znode = dirty_cow_znode(c, &c->zroot);
1323 if (IS_ERR(znode))
1324 return PTR_ERR(znode);
1325
1326 znode->time = time;
1327
1328 while (1) {
1329 struct ubifs_zbranch *zbr;
1330
1331 exact = ubifs_search_zbranch(c, znode, key, n);
1332
1333 if (znode->level == 0)
1334 break;
1335
1336 if (*n < 0)
1337 *n = 0;
1338 zbr = &znode->zbranch[*n];
1339
1340 if (zbr->znode) {
1341 znode->time = time;
1342 znode = dirty_cow_znode(c, zbr);
1343 if (IS_ERR(znode))
1344 return PTR_ERR(znode);
1345 continue;
1346 }
1347
1348 /* znode is not in TNC cache, load it from the media */
1349 znode = ubifs_load_znode(c, zbr, znode, *n);
1350 if (IS_ERR(znode))
1351 return PTR_ERR(znode);
1352 znode = dirty_cow_znode(c, zbr);
1353 if (IS_ERR(znode))
1354 return PTR_ERR(znode);
1355 }
1356
1357 *zn = znode;
1358 if (exact || !is_hash_key(c, key) || *n != -1) {
1359 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1360 return exact;
1361 }
1362
1363 /*
1364 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1365 * code.
1366 */
1367 err = tnc_prev(c, &znode, n);
1368 if (err == -ENOENT) {
1369 *n = -1;
1370 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1371 return 0;
1372 }
1373 if (unlikely(err < 0))
1374 return err;
1375 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1376 *n = -1;
1377 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1378 return 0;
1379 }
1380
1381 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1382 znode = dirty_cow_bottom_up(c, znode);
1383 if (IS_ERR(znode))
1384 return PTR_ERR(znode);
1385 }
1386
1387 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1388 *zn = znode;
1389 return 1;
1390}
1391
1392/**
1393 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1394 * @c: UBIFS file-system description object
1395 * @lnum: LEB number
1396 * @gc_seq1: garbage collection sequence number
1397 *
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
1400 * %0 is returned.
1401 */
1402static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1403{
1404 int gc_seq2, gced_lnum;
1405
1406 gced_lnum = c->gced_lnum;
1407 smp_rmb();
1408 gc_seq2 = c->gc_seq;
1409 /* Same seq means no GC */
1410 if (gc_seq1 == gc_seq2)
1411 return 0;
1412 /* Different by more than 1 means we don't know */
1413 if (gc_seq1 + 1 != gc_seq2)
1414 return 1;
1415 /*
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.
1418 */
1419 smp_rmb();
1420 if (gced_lnum != c->gced_lnum)
1421 return 1;
1422 /* Finally we can check lnum */
1423 if (gced_lnum == lnum)
1424 return 1;
1425 return 0;
1426}
1427
1428/**
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
1435 *
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.
1440 */
1441int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1442 void *node, int *lnum, int *offs)
1443{
1444 int found, n, err, safely = 0, gc_seq1;
1445 struct ubifs_znode *znode;
1446 struct ubifs_zbranch zbr, *zt;
1447
1448again:
1449 mutex_lock(&c->tnc_mutex);
1450 found = ubifs_lookup_level0(c, key, &znode, &n);
1451 if (!found) {
1452 err = -ENOENT;
1453 goto out;
1454 } else if (found < 0) {
1455 err = found;
1456 goto out;
1457 }
1458 zt = &znode->zbranch[n];
1459 if (lnum) {
1460 *lnum = zt->lnum;
1461 *offs = zt->offs;
1462 }
1463 if (is_hash_key(c, key)) {
1464 /*
1465 * In this case the leaf node cache gets used, so we pass the
1466 * address of the zbranch and keep the mutex locked
1467 */
1468 err = tnc_read_hashed_node(c, zt, node);
1469 goto out;
1470 }
1471 if (safely) {
1472 err = ubifs_tnc_read_node(c, zt, node);
1473 goto out;
1474 }
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);
1479
1480 if (ubifs_get_wbuf(c, zbr.lnum)) {
1481 /* We do not GC journal heads */
1482 err = ubifs_tnc_read_node(c, &zbr, node);
1483 return err;
1484 }
1485
1486 err = fallible_read_node(c, key, &zbr, node);
1487 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1488 /*
1489 * The node may have been GC'ed out from under us so try again
1490 * while keeping the TNC mutex locked.
1491 */
1492 safely = 1;
1493 goto again;
1494 }
1495 return 0;
1496
1497out:
1498 mutex_unlock(&c->tnc_mutex);
1499 return err;
1500}
1501
1502/**
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
1506 *
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.
1510 *
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.
1514 */
1515int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1516{
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;
1521
1522 bu->cnt = 0;
1523 bu->blk_cnt = 0;
1524 bu->eof = 0;
1525
1526 mutex_lock(&c->tnc_mutex);
1527 /* Find first key */
1528 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1529 if (err < 0)
1530 goto out;
1531 if (err) {
1532 /* Key found */
1533 len = znode->zbranch[n].len;
1534 /* The buffer must be big enough for at least 1 node */
1535 if (len > bu->buf_len) {
1536 err = -EINVAL;
1537 goto out;
1538 }
1539 /* Add this key */
1540 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1541 bu->blk_cnt += 1;
1542 lnum = znode->zbranch[n].lnum;
1543 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1544 }
1545 while (1) {
1546 struct ubifs_zbranch *zbr;
1547 union ubifs_key *key;
1548 unsigned int next_block;
1549
1550 /* Find next key */
1551 err = tnc_next(c, &znode, &n);
1552 if (err)
1553 goto out;
1554 zbr = &znode->zbranch[n];
1555 key = &zbr->key;
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) {
1559 err = -ENOENT;
1560 goto out;
1561 }
1562 if (lnum < 0) {
1563 /* First key found */
1564 lnum = zbr->lnum;
1565 offs = ALIGN(zbr->offs + zbr->len, 8);
1566 len = zbr->len;
1567 if (len > bu->buf_len) {
1568 err = -EINVAL;
1569 goto out;
1570 }
1571 } else {
1572 /*
1573 * The data nodes must be in consecutive positions in
1574 * the same LEB.
1575 */
1576 if (zbr->lnum != lnum || zbr->offs != offs)
1577 goto out;
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)
1582 goto out;
1583 }
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)
1588 goto out;
1589 block = next_block;
1590 /* Add this key */
1591 bu->zbranch[bu->cnt++] = *zbr;
1592 bu->blk_cnt += 1;
1593 /* See if we have room for more */
1594 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1595 goto out;
1596 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1597 goto out;
1598 }
1599out:
1600 if (err == -ENOENT) {
1601 bu->eof = 1;
1602 err = 0;
1603 }
1604 bu->gc_seq = c->gc_seq;
1605 mutex_unlock(&c->tnc_mutex);
1606 if (err)
1607 return err;
1608 /*
1609 * An enormous hole could cause bulk-read to encompass too many
1610 * page cache pages, so limit the number here.
1611 */
1612 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1613 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1614 /*
1615 * Ensure that bulk-read covers a whole number of page cache
1616 * pages.
1617 */
1618 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1619 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1620 return 0;
1621 if (bu->eof) {
1622 /* At the end of file we can round up */
1623 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1624 return 0;
1625 }
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);
1629 while (bu->cnt) {
1630 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1631 break;
1632 bu->cnt -= 1;
1633 }
1634 return 0;
1635}
1636
1637/**
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
1641 * @len: read length
1642 * @lnum: LEB number from which to read
1643 * @offs: offset from which to read
1644 *
1645 * This functions returns %0 on success or a negative error code on failure.
1646 */
1647static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1648 int offs)
1649{
1650 const struct ubifs_info *c = wbuf->c;
1651 int rlen, overlap;
1652
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);
1657
1658 spin_lock(&wbuf->lock);
1659 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1660 if (!overlap) {
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);
1664 }
1665
1666 /* Don't read under wbuf */
1667 rlen = wbuf->offs - offs;
1668 if (rlen < 0)
1669 rlen = 0;
1670
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);
1674
1675 if (rlen > 0)
1676 /* Read everything that goes before write-buffer */
1677 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1678
1679 return 0;
1680}
1681
1682/**
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
1687 *
1688 * This functions returns %0 on success or a negative error code on failure.
1689 */
1690static int validate_data_node(struct ubifs_info *c, void *buf,
1691 struct ubifs_zbranch *zbr)
1692{
1693 union ubifs_key key1;
1694 struct ubifs_ch *ch = buf;
1695 int err, len;
1696
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);
1700 goto out_err;
1701 }
1702
1703 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1704 if (err) {
1705 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1706 goto out;
1707 }
1708
1709 err = ubifs_node_check_hash(c, buf, zbr->hash);
1710 if (err) {
1711 ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs);
1712 return err;
1713 }
1714
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);
1718 goto out_err;
1719 }
1720
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 ");
1728 goto out_err;
1729 }
1730
1731 return 0;
1732
1733out_err:
1734 err = -EINVAL;
1735out:
1736 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1737 ubifs_dump_node(c, buf);
1738 dump_stack();
1739 return err;
1740}
1741
1742/**
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
1746 *
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
1750 * failure.
1751 */
1752int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1753{
1754 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1755 struct ubifs_wbuf *wbuf;
1756 void *buf;
1757
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);
1762 return -EINVAL;
1763 }
1764
1765 /* Do the read */
1766 wbuf = ubifs_get_wbuf(c, lnum);
1767 if (wbuf)
1768 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1769 else
1770 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1771
1772 /* Check for a race with GC */
1773 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1774 return -EAGAIN;
1775
1776 if (err && err != -EBADMSG) {
1777 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1778 lnum, offs, err);
1779 dump_stack();
1780 dbg_tnck(&bu->key, "key ");
1781 return err;
1782 }
1783
1784 /* Validate the nodes read */
1785 buf = bu->buf;
1786 for (i = 0; i < bu->cnt; i++) {
1787 err = validate_data_node(c, buf, &bu->zbranch[i]);
1788 if (err)
1789 return err;
1790 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1791 }
1792
1793 return 0;
1794}
1795
1796/**
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
1801 * @nm: node name
1802 *
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.
1808 */
1809static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1810 void *node, const struct fscrypt_name *nm)
1811{
1812 int found, n, err;
1813 struct ubifs_znode *znode;
1814
1815 dbg_tnck(key, "key ");
1816 mutex_lock(&c->tnc_mutex);
1817 found = ubifs_lookup_level0(c, key, &znode, &n);
1818 if (!found) {
1819 err = -ENOENT;
1820 goto out_unlock;
1821 } else if (found < 0) {
1822 err = found;
1823 goto out_unlock;
1824 }
1825
1826 ubifs_assert(c, n >= 0);
1827
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))
1831 goto out_unlock;
1832 if (err == 0) {
1833 err = -ENOENT;
1834 goto out_unlock;
1835 }
1836
1837 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1838
1839out_unlock:
1840 mutex_unlock(&c->tnc_mutex);
1841 return err;
1842}
1843
1844/**
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
1849 * @nm: node name
1850 *
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.
1856 */
1857int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1858 void *node, const struct fscrypt_name *nm)
1859{
1860 int err, len;
1861 const struct ubifs_dent_node *dent = node;
1862
1863 /*
1864 * We assume that in most of the cases there are no name collisions and
1865 * 'ubifs_tnc_lookup()' returns us the right direntry.
1866 */
1867 err = ubifs_tnc_lookup(c, key, node);
1868 if (err)
1869 return err;
1870
1871 len = le16_to_cpu(dent->nlen);
1872 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1873 return 0;
1874
1875 /*
1876 * Unluckily, there are hash collisions and we have to iterate over
1877 * them look at each direntry with colliding name hash sequentially.
1878 */
1879
1880 return do_lookup_nm(c, key, node, nm);
1881}
1882
1883static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1884 struct ubifs_dent_node *dent, uint32_t cookie,
1885 struct ubifs_znode **zn, int *n, int exact)
1886{
1887 int err;
1888 struct ubifs_znode *znode = *zn;
1889 struct ubifs_zbranch *zbr;
1890 union ubifs_key *dkey;
1891
1892 if (!exact) {
1893 err = tnc_next(c, &znode, n);
1894 if (err)
1895 return err;
1896 }
1897
1898 for (;;) {
1899 zbr = &znode->zbranch[*n];
1900 dkey = &zbr->key;
1901
1902 if (key_inum(c, dkey) != key_inum(c, key) ||
1903 key_type(c, dkey) != key_type(c, key)) {
1904 return -ENOENT;
1905 }
1906
1907 err = tnc_read_hashed_node(c, zbr, dent);
1908 if (err)
1909 return err;
1910
1911 if (key_hash(c, key) == key_hash(c, dkey) &&
1912 le32_to_cpu(dent->cookie) == cookie) {
1913 *zn = znode;
1914 return 0;
1915 }
1916
1917 err = tnc_next(c, &znode, n);
1918 if (err)
1919 return err;
1920 }
1921}
1922
1923static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1924 struct ubifs_dent_node *dent, uint32_t cookie)
1925{
1926 int n, err;
1927 struct ubifs_znode *znode;
1928 union ubifs_key start_key;
1929
1930 ubifs_assert(c, is_hash_key(c, key));
1931
1932 lowest_dent_key(c, &start_key, key_inum(c, key));
1933
1934 mutex_lock(&c->tnc_mutex);
1935 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1936 if (unlikely(err < 0))
1937 goto out_unlock;
1938
1939 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
1940
1941out_unlock:
1942 mutex_unlock(&c->tnc_mutex);
1943 return err;
1944}
1945
1946/**
1947 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1948 * @c: UBIFS file-system description object
1949 * @key: node key to lookup
1950 * @node: the node is returned here
1951 * @cookie: node cookie for collision resolution
1952 *
1953 * This function looks up and reads a node which contains name hash in the key.
1954 * Since the hash may have collisions, there may be many nodes with the same
1955 * key, so we have to sequentially look to all of them until the needed one
1956 * with the same cookie value is found.
1957 * This function returns zero in case of success, %-ENOENT if the node
1958 * was not found, and a negative error code in case of failure.
1959 */
1960int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1961 void *node, uint32_t cookie)
1962{
1963 int err;
1964 const struct ubifs_dent_node *dent = node;
1965
1966 if (!c->double_hash)
1967 return -EOPNOTSUPP;
1968
1969 /*
1970 * We assume that in most of the cases there are no name collisions and
1971 * 'ubifs_tnc_lookup()' returns us the right direntry.
1972 */
1973 err = ubifs_tnc_lookup(c, key, node);
1974 if (err)
1975 return err;
1976
1977 if (le32_to_cpu(dent->cookie) == cookie)
1978 return 0;
1979
1980 /*
1981 * Unluckily, there are hash collisions and we have to iterate over
1982 * them look at each direntry with colliding name hash sequentially.
1983 */
1984 return do_lookup_dh(c, key, node, cookie);
1985}
1986
1987/**
1988 * correct_parent_keys - correct parent znodes' keys.
1989 * @c: UBIFS file-system description object
1990 * @znode: znode to correct parent znodes for
1991 *
1992 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1993 * zbranch changes, keys of parent znodes have to be corrected. This helper
1994 * function is called in such situations and corrects the keys if needed.
1995 */
1996static void correct_parent_keys(const struct ubifs_info *c,
1997 struct ubifs_znode *znode)
1998{
1999 union ubifs_key *key, *key1;
2000
2001 ubifs_assert(c, znode->parent);
2002 ubifs_assert(c, znode->iip == 0);
2003
2004 key = &znode->zbranch[0].key;
2005 key1 = &znode->parent->zbranch[0].key;
2006
2007 while (keys_cmp(c, key, key1) < 0) {
2008 key_copy(c, key, key1);
2009 znode = znode->parent;
2010 znode->alt = 1;
2011 if (!znode->parent || znode->iip)
2012 break;
2013 key1 = &znode->parent->zbranch[0].key;
2014 }
2015}
2016
2017/**
2018 * insert_zbranch - insert a zbranch into a znode.
2019 * @c: UBIFS file-system description object
2020 * @znode: znode into which to insert
2021 * @zbr: zbranch to insert
2022 * @n: slot number to insert to
2023 *
2024 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2025 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2026 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2027 * slot, zbranches starting from @n have to be moved right.
2028 */
2029static void insert_zbranch(struct ubifs_info *c, struct ubifs_znode *znode,
2030 const struct ubifs_zbranch *zbr, int n)
2031{
2032 int i;
2033
2034 ubifs_assert(c, ubifs_zn_dirty(znode));
2035
2036 if (znode->level) {
2037 for (i = znode->child_cnt; i > n; i--) {
2038 znode->zbranch[i] = znode->zbranch[i - 1];
2039 if (znode->zbranch[i].znode)
2040 znode->zbranch[i].znode->iip = i;
2041 }
2042 if (zbr->znode)
2043 zbr->znode->iip = n;
2044 } else
2045 for (i = znode->child_cnt; i > n; i--)
2046 znode->zbranch[i] = znode->zbranch[i - 1];
2047
2048 znode->zbranch[n] = *zbr;
2049 znode->child_cnt += 1;
2050
2051 /*
2052 * After inserting at slot zero, the lower bound of the key range of
2053 * this znode may have changed. If this znode is subsequently split
2054 * then the upper bound of the key range may change, and furthermore
2055 * it could change to be lower than the original lower bound. If that
2056 * happens, then it will no longer be possible to find this znode in the
2057 * TNC using the key from the index node on flash. That is bad because
2058 * if it is not found, we will assume it is obsolete and may overwrite
2059 * it. Then if there is an unclean unmount, we will start using the
2060 * old index which will be broken.
2061 *
2062 * So we first mark znodes that have insertions at slot zero, and then
2063 * if they are split we add their lnum/offs to the old_idx tree.
2064 */
2065 if (n == 0)
2066 znode->alt = 1;
2067}
2068
2069/**
2070 * tnc_insert - insert a node into TNC.
2071 * @c: UBIFS file-system description object
2072 * @znode: znode to insert into
2073 * @zbr: branch to insert
2074 * @n: slot number to insert new zbranch to
2075 *
2076 * This function inserts a new node described by @zbr into znode @znode. If
2077 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2078 * are splat as well if needed. Returns zero in case of success or a negative
2079 * error code in case of failure.
2080 */
2081static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2082 struct ubifs_zbranch *zbr, int n)
2083{
2084 struct ubifs_znode *zn, *zi, *zp;
2085 int i, keep, move, appending = 0;
2086 union ubifs_key *key = &zbr->key, *key1;
2087
2088 ubifs_assert(c, n >= 0 && n <= c->fanout);
2089
2090 /* Implement naive insert for now */
2091again:
2092 zp = znode->parent;
2093 if (znode->child_cnt < c->fanout) {
2094 ubifs_assert(c, n != c->fanout);
2095 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2096
2097 insert_zbranch(c, znode, zbr, n);
2098
2099 /* Ensure parent's key is correct */
2100 if (n == 0 && zp && znode->iip == 0)
2101 correct_parent_keys(c, znode);
2102
2103 return 0;
2104 }
2105
2106 /*
2107 * Unfortunately, @znode does not have more empty slots and we have to
2108 * split it.
2109 */
2110 dbg_tnck(key, "splitting level %d, key ", znode->level);
2111
2112 if (znode->alt)
2113 /*
2114 * We can no longer be sure of finding this znode by key, so we
2115 * record it in the old_idx tree.
2116 */
2117 ins_clr_old_idx_znode(c, znode);
2118
2119 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2120 if (!zn)
2121 return -ENOMEM;
2122 zn->parent = zp;
2123 zn->level = znode->level;
2124
2125 /* Decide where to split */
2126 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2127 /* Try not to split consecutive data keys */
2128 if (n == c->fanout) {
2129 key1 = &znode->zbranch[n - 1].key;
2130 if (key_inum(c, key1) == key_inum(c, key) &&
2131 key_type(c, key1) == UBIFS_DATA_KEY)
2132 appending = 1;
2133 } else
2134 goto check_split;
2135 } else if (appending && n != c->fanout) {
2136 /* Try not to split consecutive data keys */
2137 appending = 0;
2138check_split:
2139 if (n >= (c->fanout + 1) / 2) {
2140 key1 = &znode->zbranch[0].key;
2141 if (key_inum(c, key1) == key_inum(c, key) &&
2142 key_type(c, key1) == UBIFS_DATA_KEY) {
2143 key1 = &znode->zbranch[n].key;
2144 if (key_inum(c, key1) != key_inum(c, key) ||
2145 key_type(c, key1) != UBIFS_DATA_KEY) {
2146 keep = n;
2147 move = c->fanout - keep;
2148 zi = znode;
2149 goto do_split;
2150 }
2151 }
2152 }
2153 }
2154
2155 if (appending) {
2156 keep = c->fanout;
2157 move = 0;
2158 } else {
2159 keep = (c->fanout + 1) / 2;
2160 move = c->fanout - keep;
2161 }
2162
2163 /*
2164 * Although we don't at present, we could look at the neighbors and see
2165 * if we can move some zbranches there.
2166 */
2167
2168 if (n < keep) {
2169 /* Insert into existing znode */
2170 zi = znode;
2171 move += 1;
2172 keep -= 1;
2173 } else {
2174 /* Insert into new znode */
2175 zi = zn;
2176 n -= keep;
2177 /* Re-parent */
2178 if (zn->level != 0)
2179 zbr->znode->parent = zn;
2180 }
2181
2182do_split:
2183
2184 __set_bit(DIRTY_ZNODE, &zn->flags);
2185 atomic_long_inc(&c->dirty_zn_cnt);
2186
2187 zn->child_cnt = move;
2188 znode->child_cnt = keep;
2189
2190 dbg_tnc("moving %d, keeping %d", move, keep);
2191
2192 /* Move zbranch */
2193 for (i = 0; i < move; i++) {
2194 zn->zbranch[i] = znode->zbranch[keep + i];
2195 /* Re-parent */
2196 if (zn->level != 0)
2197 if (zn->zbranch[i].znode) {
2198 zn->zbranch[i].znode->parent = zn;
2199 zn->zbranch[i].znode->iip = i;
2200 }
2201 }
2202
2203 /* Insert new key and branch */
2204 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2205
2206 insert_zbranch(c, zi, zbr, n);
2207
2208 /* Insert new znode (produced by spitting) into the parent */
2209 if (zp) {
2210 if (n == 0 && zi == znode && znode->iip == 0)
2211 correct_parent_keys(c, znode);
2212
2213 /* Locate insertion point */
2214 n = znode->iip + 1;
2215
2216 /* Tail recursion */
2217 zbr->key = zn->zbranch[0].key;
2218 zbr->znode = zn;
2219 zbr->lnum = 0;
2220 zbr->offs = 0;
2221 zbr->len = 0;
2222 znode = zp;
2223
2224 goto again;
2225 }
2226
2227 /* We have to split root znode */
2228 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2229
2230 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2231 if (!zi)
2232 return -ENOMEM;
2233
2234 zi->child_cnt = 2;
2235 zi->level = znode->level + 1;
2236
2237 __set_bit(DIRTY_ZNODE, &zi->flags);
2238 atomic_long_inc(&c->dirty_zn_cnt);
2239
2240 zi->zbranch[0].key = znode->zbranch[0].key;
2241 zi->zbranch[0].znode = znode;
2242 zi->zbranch[0].lnum = c->zroot.lnum;
2243 zi->zbranch[0].offs = c->zroot.offs;
2244 zi->zbranch[0].len = c->zroot.len;
2245 zi->zbranch[1].key = zn->zbranch[0].key;
2246 zi->zbranch[1].znode = zn;
2247
2248 c->zroot.lnum = 0;
2249 c->zroot.offs = 0;
2250 c->zroot.len = 0;
2251 c->zroot.znode = zi;
2252
2253 zn->parent = zi;
2254 zn->iip = 1;
2255 znode->parent = zi;
2256 znode->iip = 0;
2257
2258 return 0;
2259}
2260
2261/**
2262 * ubifs_tnc_add - add a node to TNC.
2263 * @c: UBIFS file-system description object
2264 * @key: key to add
2265 * @lnum: LEB number of node
2266 * @offs: node offset
2267 * @len: node length
2268 * @hash: The hash over the node
2269 *
2270 * This function adds a node with key @key to TNC. The node may be new or it may
2271 * obsolete some existing one. Returns %0 on success or negative error code on
2272 * failure.
2273 */
2274int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2275 int offs, int len, const u8 *hash)
2276{
2277 int found, n, err = 0;
2278 struct ubifs_znode *znode;
2279
2280 mutex_lock(&c->tnc_mutex);
2281 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2282 found = lookup_level0_dirty(c, key, &znode, &n);
2283 if (!found) {
2284 struct ubifs_zbranch zbr;
2285
2286 zbr.znode = NULL;
2287 zbr.lnum = lnum;
2288 zbr.offs = offs;
2289 zbr.len = len;
2290 ubifs_copy_hash(c, hash, zbr.hash);
2291 key_copy(c, key, &zbr.key);
2292 err = tnc_insert(c, znode, &zbr, n + 1);
2293 } else if (found == 1) {
2294 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2295
2296 lnc_free(zbr);
2297 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2298 zbr->lnum = lnum;
2299 zbr->offs = offs;
2300 zbr->len = len;
2301 ubifs_copy_hash(c, hash, zbr->hash);
2302 } else
2303 err = found;
2304 if (!err)
2305 err = dbg_check_tnc(c, 0);
2306 mutex_unlock(&c->tnc_mutex);
2307
2308 return err;
2309}
2310
2311/**
2312 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2313 * @c: UBIFS file-system description object
2314 * @key: key to add
2315 * @old_lnum: LEB number of old node
2316 * @old_offs: old node offset
2317 * @lnum: LEB number of node
2318 * @offs: node offset
2319 * @len: node length
2320 *
2321 * This function replaces a node with key @key in the TNC only if the old node
2322 * is found. This function is called by garbage collection when node are moved.
2323 * Returns %0 on success or negative error code on failure.
2324 */
2325int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2326 int old_lnum, int old_offs, int lnum, int offs, int len)
2327{
2328 int found, n, err = 0;
2329 struct ubifs_znode *znode;
2330
2331 mutex_lock(&c->tnc_mutex);
2332 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2333 old_offs, lnum, offs, len);
2334 found = lookup_level0_dirty(c, key, &znode, &n);
2335 if (found < 0) {
2336 err = found;
2337 goto out_unlock;
2338 }
2339
2340 if (found == 1) {
2341 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2342
2343 found = 0;
2344 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2345 lnc_free(zbr);
2346 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2347 if (err)
2348 goto out_unlock;
2349 zbr->lnum = lnum;
2350 zbr->offs = offs;
2351 zbr->len = len;
2352 found = 1;
2353 } else if (is_hash_key(c, key)) {
2354 found = resolve_collision_directly(c, key, &znode, &n,
2355 old_lnum, old_offs);
2356 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2357 found, znode, n, old_lnum, old_offs);
2358 if (found < 0) {
2359 err = found;
2360 goto out_unlock;
2361 }
2362
2363 if (found) {
2364 /* Ensure the znode is dirtied */
2365 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2366 znode = dirty_cow_bottom_up(c, znode);
2367 if (IS_ERR(znode)) {
2368 err = PTR_ERR(znode);
2369 goto out_unlock;
2370 }
2371 }
2372 zbr = &znode->zbranch[n];
2373 lnc_free(zbr);
2374 err = ubifs_add_dirt(c, zbr->lnum,
2375 zbr->len);
2376 if (err)
2377 goto out_unlock;
2378 zbr->lnum = lnum;
2379 zbr->offs = offs;
2380 zbr->len = len;
2381 }
2382 }
2383 }
2384
2385 if (!found)
2386 err = ubifs_add_dirt(c, lnum, len);
2387
2388 if (!err)
2389 err = dbg_check_tnc(c, 0);
2390
2391out_unlock:
2392 mutex_unlock(&c->tnc_mutex);
2393 return err;
2394}
2395
2396/**
2397 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2398 * @c: UBIFS file-system description object
2399 * @key: key to add
2400 * @lnum: LEB number of node
2401 * @offs: node offset
2402 * @len: node length
2403 * @hash: The hash over the node
2404 * @nm: node name
2405 *
2406 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2407 * may have collisions, like directory entry keys.
2408 */
2409int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2410 int lnum, int offs, int len, const u8 *hash,
2411 const struct fscrypt_name *nm)
2412{
2413 int found, n, err = 0;
2414 struct ubifs_znode *znode;
2415
2416 mutex_lock(&c->tnc_mutex);
2417 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2418 found = lookup_level0_dirty(c, key, &znode, &n);
2419 if (found < 0) {
2420 err = found;
2421 goto out_unlock;
2422 }
2423
2424 if (found == 1) {
2425 if (c->replaying)
2426 found = fallible_resolve_collision(c, key, &znode, &n,
2427 nm, 1);
2428 else
2429 found = resolve_collision(c, key, &znode, &n, nm);
2430 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2431 if (found < 0) {
2432 err = found;
2433 goto out_unlock;
2434 }
2435
2436 /* Ensure the znode is dirtied */
2437 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2438 znode = dirty_cow_bottom_up(c, znode);
2439 if (IS_ERR(znode)) {
2440 err = PTR_ERR(znode);
2441 goto out_unlock;
2442 }
2443 }
2444
2445 if (found == 1) {
2446 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2447
2448 lnc_free(zbr);
2449 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2450 zbr->lnum = lnum;
2451 zbr->offs = offs;
2452 zbr->len = len;
2453 ubifs_copy_hash(c, hash, zbr->hash);
2454 goto out_unlock;
2455 }
2456 }
2457
2458 if (!found) {
2459 struct ubifs_zbranch zbr;
2460
2461 zbr.znode = NULL;
2462 zbr.lnum = lnum;
2463 zbr.offs = offs;
2464 zbr.len = len;
2465 ubifs_copy_hash(c, hash, zbr.hash);
2466 key_copy(c, key, &zbr.key);
2467 err = tnc_insert(c, znode, &zbr, n + 1);
2468 if (err)
2469 goto out_unlock;
2470 if (c->replaying) {
2471 /*
2472 * We did not find it in the index so there may be a
2473 * dangling branch still in the index. So we remove it
2474 * by passing 'ubifs_tnc_remove_nm()' the same key but
2475 * an unmatchable name.
2476 */
2477 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2478
2479 err = dbg_check_tnc(c, 0);
2480 mutex_unlock(&c->tnc_mutex);
2481 if (err)
2482 return err;
2483 return ubifs_tnc_remove_nm(c, key, &noname);
2484 }
2485 }
2486
2487out_unlock:
2488 if (!err)
2489 err = dbg_check_tnc(c, 0);
2490 mutex_unlock(&c->tnc_mutex);
2491 return err;
2492}
2493
2494/**
2495 * tnc_delete - delete a znode form TNC.
2496 * @c: UBIFS file-system description object
2497 * @znode: znode to delete from
2498 * @n: zbranch slot number to delete
2499 *
2500 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2501 * case of success and a negative error code in case of failure.
2502 */
2503static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2504{
2505 struct ubifs_zbranch *zbr;
2506 struct ubifs_znode *zp;
2507 int i, err;
2508
2509 /* Delete without merge for now */
2510 ubifs_assert(c, znode->level == 0);
2511 ubifs_assert(c, n >= 0 && n < c->fanout);
2512 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2513
2514 zbr = &znode->zbranch[n];
2515 lnc_free(zbr);
2516
2517 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2518 if (err) {
2519 ubifs_dump_znode(c, znode);
2520 return err;
2521 }
2522
2523 /* We do not "gap" zbranch slots */
2524 for (i = n; i < znode->child_cnt - 1; i++)
2525 znode->zbranch[i] = znode->zbranch[i + 1];
2526 znode->child_cnt -= 1;
2527
2528 if (znode->child_cnt > 0)
2529 return 0;
2530
2531 /*
2532 * This was the last zbranch, we have to delete this znode from the
2533 * parent.
2534 */
2535
2536 do {
2537 ubifs_assert(c, !ubifs_zn_obsolete(znode));
2538 ubifs_assert(c, ubifs_zn_dirty(znode));
2539
2540 zp = znode->parent;
2541 n = znode->iip;
2542
2543 atomic_long_dec(&c->dirty_zn_cnt);
2544
2545 err = insert_old_idx_znode(c, znode);
2546 if (err)
2547 return err;
2548
2549 if (znode->cnext) {
2550 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2551 atomic_long_inc(&c->clean_zn_cnt);
2552 atomic_long_inc(&ubifs_clean_zn_cnt);
2553 } else
2554 kfree(znode);
2555 znode = zp;
2556 } while (znode->child_cnt == 1); /* while removing last child */
2557
2558 /* Remove from znode, entry n - 1 */
2559 znode->child_cnt -= 1;
2560 ubifs_assert(c, znode->level != 0);
2561 for (i = n; i < znode->child_cnt; i++) {
2562 znode->zbranch[i] = znode->zbranch[i + 1];
2563 if (znode->zbranch[i].znode)
2564 znode->zbranch[i].znode->iip = i;
2565 }
2566
2567 /*
2568 * If this is the root and it has only 1 child then
2569 * collapse the tree.
2570 */
2571 if (!znode->parent) {
2572 while (znode->child_cnt == 1 && znode->level != 0) {
2573 zp = znode;
2574 zbr = &znode->zbranch[0];
2575 znode = get_znode(c, znode, 0);
2576 if (IS_ERR(znode))
2577 return PTR_ERR(znode);
2578 znode = dirty_cow_znode(c, zbr);
2579 if (IS_ERR(znode))
2580 return PTR_ERR(znode);
2581 znode->parent = NULL;
2582 znode->iip = 0;
2583 if (c->zroot.len) {
2584 err = insert_old_idx(c, c->zroot.lnum,
2585 c->zroot.offs);
2586 if (err)
2587 return err;
2588 }
2589 c->zroot.lnum = zbr->lnum;
2590 c->zroot.offs = zbr->offs;
2591 c->zroot.len = zbr->len;
2592 c->zroot.znode = znode;
2593 ubifs_assert(c, !ubifs_zn_obsolete(zp));
2594 ubifs_assert(c, ubifs_zn_dirty(zp));
2595 atomic_long_dec(&c->dirty_zn_cnt);
2596
2597 if (zp->cnext) {
2598 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2599 atomic_long_inc(&c->clean_zn_cnt);
2600 atomic_long_inc(&ubifs_clean_zn_cnt);
2601 } else
2602 kfree(zp);
2603 }
2604 }
2605
2606 return 0;
2607}
2608
2609/**
2610 * ubifs_tnc_remove - remove an index entry of a node.
2611 * @c: UBIFS file-system description object
2612 * @key: key of node
2613 *
2614 * Returns %0 on success or negative error code on failure.
2615 */
2616int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2617{
2618 int found, n, err = 0;
2619 struct ubifs_znode *znode;
2620
2621 mutex_lock(&c->tnc_mutex);
2622 dbg_tnck(key, "key ");
2623 found = lookup_level0_dirty(c, key, &znode, &n);
2624 if (found < 0) {
2625 err = found;
2626 goto out_unlock;
2627 }
2628 if (found == 1)
2629 err = tnc_delete(c, znode, n);
2630 if (!err)
2631 err = dbg_check_tnc(c, 0);
2632
2633out_unlock:
2634 mutex_unlock(&c->tnc_mutex);
2635 return err;
2636}
2637
2638/**
2639 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2640 * @c: UBIFS file-system description object
2641 * @key: key of node
2642 * @nm: directory entry name
2643 *
2644 * Returns %0 on success or negative error code on failure.
2645 */
2646int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2647 const struct fscrypt_name *nm)
2648{
2649 int n, err;
2650 struct ubifs_znode *znode;
2651
2652 mutex_lock(&c->tnc_mutex);
2653 dbg_tnck(key, "key ");
2654 err = lookup_level0_dirty(c, key, &znode, &n);
2655 if (err < 0)
2656 goto out_unlock;
2657
2658 if (err) {
2659 if (c->replaying)
2660 err = fallible_resolve_collision(c, key, &znode, &n,
2661 nm, 0);
2662 else
2663 err = resolve_collision(c, key, &znode, &n, nm);
2664 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2665 if (err < 0)
2666 goto out_unlock;
2667 if (err) {
2668 /* Ensure the znode is dirtied */
2669 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2670 znode = dirty_cow_bottom_up(c, znode);
2671 if (IS_ERR(znode)) {
2672 err = PTR_ERR(znode);
2673 goto out_unlock;
2674 }
2675 }
2676 err = tnc_delete(c, znode, n);
2677 }
2678 }
2679
2680out_unlock:
2681 if (!err)
2682 err = dbg_check_tnc(c, 0);
2683 mutex_unlock(&c->tnc_mutex);
2684 return err;
2685}
2686
2687/**
2688 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2689 * @c: UBIFS file-system description object
2690 * @key: key of node
2691 * @cookie: node cookie for collision resolution
2692 *
2693 * Returns %0 on success or negative error code on failure.
2694 */
2695int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2696 uint32_t cookie)
2697{
2698 int n, err;
2699 struct ubifs_znode *znode;
2700 struct ubifs_dent_node *dent;
2701 struct ubifs_zbranch *zbr;
2702
2703 if (!c->double_hash)
2704 return -EOPNOTSUPP;
2705
2706 mutex_lock(&c->tnc_mutex);
2707 err = lookup_level0_dirty(c, key, &znode, &n);
2708 if (err <= 0)
2709 goto out_unlock;
2710
2711 zbr = &znode->zbranch[n];
2712 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2713 if (!dent) {
2714 err = -ENOMEM;
2715 goto out_unlock;
2716 }
2717
2718 err = tnc_read_hashed_node(c, zbr, dent);
2719 if (err)
2720 goto out_free;
2721
2722 /* If the cookie does not match, we're facing a hash collision. */
2723 if (le32_to_cpu(dent->cookie) != cookie) {
2724 union ubifs_key start_key;
2725
2726 lowest_dent_key(c, &start_key, key_inum(c, key));
2727
2728 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2729 if (unlikely(err < 0))
2730 goto out_free;
2731
2732 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
2733 if (err)
2734 goto out_free;
2735 }
2736
2737 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2738 znode = dirty_cow_bottom_up(c, znode);
2739 if (IS_ERR(znode)) {
2740 err = PTR_ERR(znode);
2741 goto out_free;
2742 }
2743 }
2744 err = tnc_delete(c, znode, n);
2745
2746out_free:
2747 kfree(dent);
2748out_unlock:
2749 if (!err)
2750 err = dbg_check_tnc(c, 0);
2751 mutex_unlock(&c->tnc_mutex);
2752 return err;
2753}
2754
2755/**
2756 * key_in_range - determine if a key falls within a range of keys.
2757 * @c: UBIFS file-system description object
2758 * @key: key to check
2759 * @from_key: lowest key in range
2760 * @to_key: highest key in range
2761 *
2762 * This function returns %1 if the key is in range and %0 otherwise.
2763 */
2764static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2765 union ubifs_key *from_key, union ubifs_key *to_key)
2766{
2767 if (keys_cmp(c, key, from_key) < 0)
2768 return 0;
2769 if (keys_cmp(c, key, to_key) > 0)
2770 return 0;
2771 return 1;
2772}
2773
2774/**
2775 * ubifs_tnc_remove_range - remove index entries in range.
2776 * @c: UBIFS file-system description object
2777 * @from_key: lowest key to remove
2778 * @to_key: highest key to remove
2779 *
2780 * This function removes index entries starting at @from_key and ending at
2781 * @to_key. This function returns zero in case of success and a negative error
2782 * code in case of failure.
2783 */
2784int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2785 union ubifs_key *to_key)
2786{
2787 int i, n, k, err = 0;
2788 struct ubifs_znode *znode;
2789 union ubifs_key *key;
2790
2791 mutex_lock(&c->tnc_mutex);
2792 while (1) {
2793 /* Find first level 0 znode that contains keys to remove */
2794 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2795 if (err < 0)
2796 goto out_unlock;
2797
2798 if (err)
2799 key = from_key;
2800 else {
2801 err = tnc_next(c, &znode, &n);
2802 if (err == -ENOENT) {
2803 err = 0;
2804 goto out_unlock;
2805 }
2806 if (err < 0)
2807 goto out_unlock;
2808 key = &znode->zbranch[n].key;
2809 if (!key_in_range(c, key, from_key, to_key)) {
2810 err = 0;
2811 goto out_unlock;
2812 }
2813 }
2814
2815 /* Ensure the znode is dirtied */
2816 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2817 znode = dirty_cow_bottom_up(c, znode);
2818 if (IS_ERR(znode)) {
2819 err = PTR_ERR(znode);
2820 goto out_unlock;
2821 }
2822 }
2823
2824 /* Remove all keys in range except the first */
2825 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2826 key = &znode->zbranch[i].key;
2827 if (!key_in_range(c, key, from_key, to_key))
2828 break;
2829 lnc_free(&znode->zbranch[i]);
2830 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2831 znode->zbranch[i].len);
2832 if (err) {
2833 ubifs_dump_znode(c, znode);
2834 goto out_unlock;
2835 }
2836 dbg_tnck(key, "removing key ");
2837 }
2838 if (k) {
2839 for (i = n + 1 + k; i < znode->child_cnt; i++)
2840 znode->zbranch[i - k] = znode->zbranch[i];
2841 znode->child_cnt -= k;
2842 }
2843
2844 /* Now delete the first */
2845 err = tnc_delete(c, znode, n);
2846 if (err)
2847 goto out_unlock;
2848 }
2849
2850out_unlock:
2851 if (!err)
2852 err = dbg_check_tnc(c, 0);
2853 mutex_unlock(&c->tnc_mutex);
2854 return err;
2855}
2856
2857/**
2858 * ubifs_tnc_remove_ino - remove an inode from TNC.
2859 * @c: UBIFS file-system description object
2860 * @inum: inode number to remove
2861 *
2862 * This function remove inode @inum and all the extended attributes associated
2863 * with the anode from TNC and returns zero in case of success or a negative
2864 * error code in case of failure.
2865 */
2866int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2867{
2868 union ubifs_key key1, key2;
2869 struct ubifs_dent_node *xent, *pxent = NULL;
2870 struct fscrypt_name nm = {0};
2871
2872 dbg_tnc("ino %lu", (unsigned long)inum);
2873
2874 /*
2875 * Walk all extended attribute entries and remove them together with
2876 * corresponding extended attribute inodes.
2877 */
2878 lowest_xent_key(c, &key1, inum);
2879 while (1) {
2880 ino_t xattr_inum;
2881 int err;
2882
2883 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2884 if (IS_ERR(xent)) {
2885 err = PTR_ERR(xent);
2886 if (err == -ENOENT)
2887 break;
2888 return err;
2889 }
2890
2891 xattr_inum = le64_to_cpu(xent->inum);
2892 dbg_tnc("xent '%s', ino %lu", xent->name,
2893 (unsigned long)xattr_inum);
2894
2895 ubifs_evict_xattr_inode(c, xattr_inum);
2896
2897 fname_name(&nm) = xent->name;
2898 fname_len(&nm) = le16_to_cpu(xent->nlen);
2899 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2900 if (err) {
2901 kfree(xent);
2902 return err;
2903 }
2904
2905 lowest_ino_key(c, &key1, xattr_inum);
2906 highest_ino_key(c, &key2, xattr_inum);
2907 err = ubifs_tnc_remove_range(c, &key1, &key2);
2908 if (err) {
2909 kfree(xent);
2910 return err;
2911 }
2912
2913 kfree(pxent);
2914 pxent = xent;
2915 key_read(c, &xent->key, &key1);
2916 }
2917
2918 kfree(pxent);
2919 lowest_ino_key(c, &key1, inum);
2920 highest_ino_key(c, &key2, inum);
2921
2922 return ubifs_tnc_remove_range(c, &key1, &key2);
2923}
2924
2925/**
2926 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2927 * @c: UBIFS file-system description object
2928 * @key: key of last entry
2929 * @nm: name of last entry found or %NULL
2930 *
2931 * This function finds and reads the next directory or extended attribute entry
2932 * after the given key (@key) if there is one. @nm is used to resolve
2933 * collisions.
2934 *
2935 * If the name of the current entry is not known and only the key is known,
2936 * @nm->name has to be %NULL. In this case the semantics of this function is a
2937 * little bit different and it returns the entry corresponding to this key, not
2938 * the next one. If the key was not found, the closest "right" entry is
2939 * returned.
2940 *
2941 * If the fist entry has to be found, @key has to contain the lowest possible
2942 * key value for this inode and @name has to be %NULL.
2943 *
2944 * This function returns the found directory or extended attribute entry node
2945 * in case of success, %-ENOENT is returned if no entry was found, and a
2946 * negative error code is returned in case of failure.
2947 */
2948struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2949 union ubifs_key *key,
2950 const struct fscrypt_name *nm)
2951{
2952 int n, err, type = key_type(c, key);
2953 struct ubifs_znode *znode;
2954 struct ubifs_dent_node *dent;
2955 struct ubifs_zbranch *zbr;
2956 union ubifs_key *dkey;
2957
2958 dbg_tnck(key, "key ");
2959 ubifs_assert(c, is_hash_key(c, key));
2960
2961 mutex_lock(&c->tnc_mutex);
2962 err = ubifs_lookup_level0(c, key, &znode, &n);
2963 if (unlikely(err < 0))
2964 goto out_unlock;
2965
2966 if (fname_len(nm) > 0) {
2967 if (err) {
2968 /* Handle collisions */
2969 if (c->replaying)
2970 err = fallible_resolve_collision(c, key, &znode, &n,
2971 nm, 0);
2972 else
2973 err = resolve_collision(c, key, &znode, &n, nm);
2974 dbg_tnc("rc returned %d, znode %p, n %d",
2975 err, znode, n);
2976 if (unlikely(err < 0))
2977 goto out_unlock;
2978 }
2979
2980 /* Now find next entry */
2981 err = tnc_next(c, &znode, &n);
2982 if (unlikely(err))
2983 goto out_unlock;
2984 } else {
2985 /*
2986 * The full name of the entry was not given, in which case the
2987 * behavior of this function is a little different and it
2988 * returns current entry, not the next one.
2989 */
2990 if (!err) {
2991 /*
2992 * However, the given key does not exist in the TNC
2993 * tree and @znode/@n variables contain the closest
2994 * "preceding" element. Switch to the next one.
2995 */
2996 err = tnc_next(c, &znode, &n);
2997 if (err)
2998 goto out_unlock;
2999 }
3000 }
3001
3002 zbr = &znode->zbranch[n];
3003 dent = kmalloc(zbr->len, GFP_NOFS);
3004 if (unlikely(!dent)) {
3005 err = -ENOMEM;
3006 goto out_unlock;
3007 }
3008
3009 /*
3010 * The above 'tnc_next()' call could lead us to the next inode, check
3011 * this.
3012 */
3013 dkey = &zbr->key;
3014 if (key_inum(c, dkey) != key_inum(c, key) ||
3015 key_type(c, dkey) != type) {
3016 err = -ENOENT;
3017 goto out_free;
3018 }
3019
3020 err = tnc_read_hashed_node(c, zbr, dent);
3021 if (unlikely(err))
3022 goto out_free;
3023
3024 mutex_unlock(&c->tnc_mutex);
3025 return dent;
3026
3027out_free:
3028 kfree(dent);
3029out_unlock:
3030 mutex_unlock(&c->tnc_mutex);
3031 return ERR_PTR(err);
3032}
3033
3034/**
3035 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3036 * @c: UBIFS file-system description object
3037 *
3038 * Destroy left-over obsolete znodes from a failed commit.
3039 */
3040static void tnc_destroy_cnext(struct ubifs_info *c)
3041{
3042 struct ubifs_znode *cnext;
3043
3044 if (!c->cnext)
3045 return;
3046 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3047 cnext = c->cnext;
3048 do {
3049 struct ubifs_znode *znode = cnext;
3050
3051 cnext = cnext->cnext;
3052 if (ubifs_zn_obsolete(znode))
3053 kfree(znode);
3054 } while (cnext && cnext != c->cnext);
3055}
3056
3057/**
3058 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3059 * @c: UBIFS file-system description object
3060 */
3061void ubifs_tnc_close(struct ubifs_info *c)
3062{
3063 tnc_destroy_cnext(c);
3064 if (c->zroot.znode) {
3065 long n, freed;
3066
3067 n = atomic_long_read(&c->clean_zn_cnt);
3068 freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode);
3069 ubifs_assert(c, freed == n);
3070 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3071 }
3072 kfree(c->gap_lebs);
3073 kfree(c->ilebs);
3074 destroy_old_idx(c);
3075}
3076
3077/**
3078 * left_znode - get the znode to the left.
3079 * @c: UBIFS file-system description object
3080 * @znode: znode
3081 *
3082 * This function returns a pointer to the znode to the left of @znode or NULL if
3083 * there is not one. A negative error code is returned on failure.
3084 */
3085static struct ubifs_znode *left_znode(struct ubifs_info *c,
3086 struct ubifs_znode *znode)
3087{
3088 int level = znode->level;
3089
3090 while (1) {
3091 int n = znode->iip - 1;
3092
3093 /* Go up until we can go left */
3094 znode = znode->parent;
3095 if (!znode)
3096 return NULL;
3097 if (n >= 0) {
3098 /* Now go down the rightmost branch to 'level' */
3099 znode = get_znode(c, znode, n);
3100 if (IS_ERR(znode))
3101 return znode;
3102 while (znode->level != level) {
3103 n = znode->child_cnt - 1;
3104 znode = get_znode(c, znode, n);
3105 if (IS_ERR(znode))
3106 return znode;
3107 }
3108 break;
3109 }
3110 }
3111 return znode;
3112}
3113
3114/**
3115 * right_znode - get the znode to the right.
3116 * @c: UBIFS file-system description object
3117 * @znode: znode
3118 *
3119 * This function returns a pointer to the znode to the right of @znode or NULL
3120 * if there is not one. A negative error code is returned on failure.
3121 */
3122static struct ubifs_znode *right_znode(struct ubifs_info *c,
3123 struct ubifs_znode *znode)
3124{
3125 int level = znode->level;
3126
3127 while (1) {
3128 int n = znode->iip + 1;
3129
3130 /* Go up until we can go right */
3131 znode = znode->parent;
3132 if (!znode)
3133 return NULL;
3134 if (n < znode->child_cnt) {
3135 /* Now go down the leftmost branch to 'level' */
3136 znode = get_znode(c, znode, n);
3137 if (IS_ERR(znode))
3138 return znode;
3139 while (znode->level != level) {
3140 znode = get_znode(c, znode, 0);
3141 if (IS_ERR(znode))
3142 return znode;
3143 }
3144 break;
3145 }
3146 }
3147 return znode;
3148}
3149
3150/**
3151 * lookup_znode - find a particular indexing node from TNC.
3152 * @c: UBIFS file-system description object
3153 * @key: index node key to lookup
3154 * @level: index node level
3155 * @lnum: index node LEB number
3156 * @offs: index node offset
3157 *
3158 * This function searches an indexing node by its first key @key and its
3159 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3160 * nodes it traverses to TNC. This function is called for indexing nodes which
3161 * were found on the media by scanning, for example when garbage-collecting or
3162 * when doing in-the-gaps commit. This means that the indexing node which is
3163 * looked for does not have to have exactly the same leftmost key @key, because
3164 * the leftmost key may have been changed, in which case TNC will contain a
3165 * dirty znode which still refers the same @lnum:@offs. This function is clever
3166 * enough to recognize such indexing nodes.
3167 *
3168 * Note, if a znode was deleted or changed too much, then this function will
3169 * not find it. For situations like this UBIFS has the old index RB-tree
3170 * (indexed by @lnum:@offs).
3171 *
3172 * This function returns a pointer to the znode found or %NULL if it is not
3173 * found. A negative error code is returned on failure.
3174 */
3175static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3176 union ubifs_key *key, int level,
3177 int lnum, int offs)
3178{
3179 struct ubifs_znode *znode, *zn;
3180 int n, nn;
3181
3182 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3183
3184 /*
3185 * The arguments have probably been read off flash, so don't assume
3186 * they are valid.
3187 */
3188 if (level < 0)
3189 return ERR_PTR(-EINVAL);
3190
3191 /* Get the root znode */
3192 znode = c->zroot.znode;
3193 if (!znode) {
3194 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3195 if (IS_ERR(znode))
3196 return znode;
3197 }
3198 /* Check if it is the one we are looking for */
3199 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3200 return znode;
3201 /* Descend to the parent level i.e. (level + 1) */
3202 if (level >= znode->level)
3203 return NULL;
3204 while (1) {
3205 ubifs_search_zbranch(c, znode, key, &n);
3206 if (n < 0) {
3207 /*
3208 * We reached a znode where the leftmost key is greater
3209 * than the key we are searching for. This is the same
3210 * situation as the one described in a huge comment at
3211 * the end of the 'ubifs_lookup_level0()' function. And
3212 * for exactly the same reasons we have to try to look
3213 * left before giving up.
3214 */
3215 znode = left_znode(c, znode);
3216 if (!znode)
3217 return NULL;
3218 if (IS_ERR(znode))
3219 return znode;
3220 ubifs_search_zbranch(c, znode, key, &n);
3221 ubifs_assert(c, n >= 0);
3222 }
3223 if (znode->level == level + 1)
3224 break;
3225 znode = get_znode(c, znode, n);
3226 if (IS_ERR(znode))
3227 return znode;
3228 }
3229 /* Check if the child is the one we are looking for */
3230 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3231 return get_znode(c, znode, n);
3232 /* If the key is unique, there is nowhere else to look */
3233 if (!is_hash_key(c, key))
3234 return NULL;
3235 /*
3236 * The key is not unique and so may be also in the znodes to either
3237 * side.
3238 */
3239 zn = znode;
3240 nn = n;
3241 /* Look left */
3242 while (1) {
3243 /* Move one branch to the left */
3244 if (n)
3245 n -= 1;
3246 else {
3247 znode = left_znode(c, znode);
3248 if (!znode)
3249 break;
3250 if (IS_ERR(znode))
3251 return znode;
3252 n = znode->child_cnt - 1;
3253 }
3254 /* Check it */
3255 if (znode->zbranch[n].lnum == lnum &&
3256 znode->zbranch[n].offs == offs)
3257 return get_znode(c, znode, n);
3258 /* Stop if the key is less than the one we are looking for */
3259 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3260 break;
3261 }
3262 /* Back to the middle */
3263 znode = zn;
3264 n = nn;
3265 /* Look right */
3266 while (1) {
3267 /* Move one branch to the right */
3268 if (++n >= znode->child_cnt) {
3269 znode = right_znode(c, znode);
3270 if (!znode)
3271 break;
3272 if (IS_ERR(znode))
3273 return znode;
3274 n = 0;
3275 }
3276 /* Check it */
3277 if (znode->zbranch[n].lnum == lnum &&
3278 znode->zbranch[n].offs == offs)
3279 return get_znode(c, znode, n);
3280 /* Stop if the key is greater than the one we are looking for */
3281 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3282 break;
3283 }
3284 return NULL;
3285}
3286
3287/**
3288 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3289 * @c: UBIFS file-system description object
3290 * @key: key of index node
3291 * @level: index node level
3292 * @lnum: LEB number of index node
3293 * @offs: offset of index node
3294 *
3295 * This function returns %0 if the index node is not referred to in the TNC, %1
3296 * if the index node is referred to in the TNC and the corresponding znode is
3297 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3298 * znode is clean, and a negative error code in case of failure.
3299 *
3300 * Note, the @key argument has to be the key of the first child. Also note,
3301 * this function relies on the fact that 0:0 is never a valid LEB number and
3302 * offset for a main-area node.
3303 */
3304int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3305 int lnum, int offs)
3306{
3307 struct ubifs_znode *znode;
3308
3309 znode = lookup_znode(c, key, level, lnum, offs);
3310 if (!znode)
3311 return 0;
3312 if (IS_ERR(znode))
3313 return PTR_ERR(znode);
3314
3315 return ubifs_zn_dirty(znode) ? 1 : 2;
3316}
3317
3318/**
3319 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3320 * @c: UBIFS file-system description object
3321 * @key: node key
3322 * @lnum: node LEB number
3323 * @offs: node offset
3324 *
3325 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3326 * not, and a negative error code in case of failure.
3327 *
3328 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3329 * and offset for a main-area node.
3330 */
3331static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3332 int lnum, int offs)
3333{
3334 struct ubifs_zbranch *zbr;
3335 struct ubifs_znode *znode, *zn;
3336 int n, found, err, nn;
3337 const int unique = !is_hash_key(c, key);
3338
3339 found = ubifs_lookup_level0(c, key, &znode, &n);
3340 if (found < 0)
3341 return found; /* Error code */
3342 if (!found)
3343 return 0;
3344 zbr = &znode->zbranch[n];
3345 if (lnum == zbr->lnum && offs == zbr->offs)
3346 return 1; /* Found it */
3347 if (unique)
3348 return 0;
3349 /*
3350 * Because the key is not unique, we have to look left
3351 * and right as well
3352 */
3353 zn = znode;
3354 nn = n;
3355 /* Look left */
3356 while (1) {
3357 err = tnc_prev(c, &znode, &n);
3358 if (err == -ENOENT)
3359 break;
3360 if (err)
3361 return err;
3362 if (keys_cmp(c, key, &znode->zbranch[n].key))
3363 break;
3364 zbr = &znode->zbranch[n];
3365 if (lnum == zbr->lnum && offs == zbr->offs)
3366 return 1; /* Found it */
3367 }
3368 /* Look right */
3369 znode = zn;
3370 n = nn;
3371 while (1) {
3372 err = tnc_next(c, &znode, &n);
3373 if (err) {
3374 if (err == -ENOENT)
3375 return 0;
3376 return err;
3377 }
3378 if (keys_cmp(c, key, &znode->zbranch[n].key))
3379 break;
3380 zbr = &znode->zbranch[n];
3381 if (lnum == zbr->lnum && offs == zbr->offs)
3382 return 1; /* Found it */
3383 }
3384 return 0;
3385}
3386
3387/**
3388 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3389 * @c: UBIFS file-system description object
3390 * @key: node key
3391 * @level: index node level (if it is an index node)
3392 * @lnum: node LEB number
3393 * @offs: node offset
3394 * @is_idx: non-zero if the node is an index node
3395 *
3396 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3397 * negative error code in case of failure. For index nodes, @key has to be the
3398 * key of the first child. An index node is considered to be in the TNC only if
3399 * the corresponding znode is clean or has not been loaded.
3400 */
3401int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3402 int lnum, int offs, int is_idx)
3403{
3404 int err;
3405
3406 mutex_lock(&c->tnc_mutex);
3407 if (is_idx) {
3408 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3409 if (err < 0)
3410 goto out_unlock;
3411 if (err == 1)
3412 /* The index node was found but it was dirty */
3413 err = 0;
3414 else if (err == 2)
3415 /* The index node was found and it was clean */
3416 err = 1;
3417 else
3418 BUG_ON(err != 0);
3419 } else
3420 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3421
3422out_unlock:
3423 mutex_unlock(&c->tnc_mutex);
3424 return err;
3425}
3426
3427/**
3428 * ubifs_dirty_idx_node - dirty an index node.
3429 * @c: UBIFS file-system description object
3430 * @key: index node key
3431 * @level: index node level
3432 * @lnum: index node LEB number
3433 * @offs: index node offset
3434 *
3435 * This function loads and dirties an index node so that it can be garbage
3436 * collected. The @key argument has to be the key of the first child. This
3437 * function relies on the fact that 0:0 is never a valid LEB number and offset
3438 * for a main-area node. Returns %0 on success and a negative error code on
3439 * failure.
3440 */
3441int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3442 int lnum, int offs)
3443{
3444 struct ubifs_znode *znode;
3445 int err = 0;
3446
3447 mutex_lock(&c->tnc_mutex);
3448 znode = lookup_znode(c, key, level, lnum, offs);
3449 if (!znode)
3450 goto out_unlock;
3451 if (IS_ERR(znode)) {
3452 err = PTR_ERR(znode);
3453 goto out_unlock;
3454 }
3455 znode = dirty_cow_bottom_up(c, znode);
3456 if (IS_ERR(znode)) {
3457 err = PTR_ERR(znode);
3458 goto out_unlock;
3459 }
3460
3461out_unlock:
3462 mutex_unlock(&c->tnc_mutex);
3463 return err;
3464}
3465
3466/**
3467 * dbg_check_inode_size - check if inode size is correct.
3468 * @c: UBIFS file-system description object
3469 * @inum: inode number
3470 * @size: inode size
3471 *
3472 * This function makes sure that the inode size (@size) is correct and it does
3473 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3474 * if it has a data page beyond @size, and other negative error code in case of
3475 * other errors.
3476 */
3477int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3478 loff_t size)
3479{
3480 int err, n;
3481 union ubifs_key from_key, to_key, *key;
3482 struct ubifs_znode *znode;
3483 unsigned int block;
3484
3485 if (!S_ISREG(inode->i_mode))
3486 return 0;
3487 if (!dbg_is_chk_gen(c))
3488 return 0;
3489
3490 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3491 data_key_init(c, &from_key, inode->i_ino, block);
3492 highest_data_key(c, &to_key, inode->i_ino);
3493
3494 mutex_lock(&c->tnc_mutex);
3495 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3496 if (err < 0)
3497 goto out_unlock;
3498
3499 if (err) {
3500 key = &from_key;
3501 goto out_dump;
3502 }
3503
3504 err = tnc_next(c, &znode, &n);
3505 if (err == -ENOENT) {
3506 err = 0;
3507 goto out_unlock;
3508 }
3509 if (err < 0)
3510 goto out_unlock;
3511
3512 ubifs_assert(c, err == 0);
3513 key = &znode->zbranch[n].key;
3514 if (!key_in_range(c, key, &from_key, &to_key))
3515 goto out_unlock;
3516
3517out_dump:
3518 block = key_block(c, key);
3519 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3520 (unsigned long)inode->i_ino, size,
3521 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3522 mutex_unlock(&c->tnc_mutex);
3523 ubifs_dump_inode(c, inode);
3524 dump_stack();
3525 return -EINVAL;
3526
3527out_unlock:
3528 mutex_unlock(&c->tnc_mutex);
3529 return err;
3530}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 *
7 * Authors: Adrian Hunter
8 * Artem Bityutskiy (Битюцкий Артём)
9 */
10
11/*
12 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
13 * the UBIFS B-tree.
14 *
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
18 * the mutex locked.
19 */
20
21#include <linux/crc32.h>
22#include <linux/slab.h>
23#include "ubifs.h"
24
25static int try_read_node(const struct ubifs_info *c, void *buf, int type,
26 struct ubifs_zbranch *zbr);
27static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
28 struct ubifs_zbranch *zbr, void *node);
29
30/*
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
35 * first
36 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
37 *
38 * These constants were introduce to improve readability.
39 */
40enum {
41 NAME_LESS = 0,
42 NAME_MATCHES = 1,
43 NAME_GREATER = 2,
44 NOT_ON_MEDIA = 3,
45};
46
47static void do_insert_old_idx(struct ubifs_info *c,
48 struct ubifs_old_idx *old_idx)
49{
50 struct ubifs_old_idx *o;
51 struct rb_node **p, *parent = NULL;
52
53 p = &c->old_idx.rb_node;
54 while (*p) {
55 parent = *p;
56 o = rb_entry(parent, struct ubifs_old_idx, rb);
57 if (old_idx->lnum < o->lnum)
58 p = &(*p)->rb_left;
59 else if (old_idx->lnum > o->lnum)
60 p = &(*p)->rb_right;
61 else if (old_idx->offs < o->offs)
62 p = &(*p)->rb_left;
63 else if (old_idx->offs > o->offs)
64 p = &(*p)->rb_right;
65 else {
66 ubifs_err(c, "old idx added twice!");
67 kfree(old_idx);
68 return;
69 }
70 }
71 rb_link_node(&old_idx->rb, parent, p);
72 rb_insert_color(&old_idx->rb, &c->old_idx);
73}
74
75/**
76 * insert_old_idx - record an index node obsoleted since the last commit start.
77 * @c: UBIFS file-system description object
78 * @lnum: LEB number of obsoleted index node
79 * @offs: offset of obsoleted index node
80 *
81 * Returns %0 on success, and a negative error code on failure.
82 *
83 * For recovery, there must always be a complete intact version of the index on
84 * flash at all times. That is called the "old index". It is the index as at the
85 * time of the last successful commit. Many of the index nodes in the old index
86 * may be dirty, but they must not be erased until the next successful commit
87 * (at which point that index becomes the old index).
88 *
89 * That means that the garbage collection and the in-the-gaps method of
90 * committing must be able to determine if an index node is in the old index.
91 * Most of the old index nodes can be found by looking up the TNC using the
92 * 'lookup_znode()' function. However, some of the old index nodes may have
93 * been deleted from the current index or may have been changed so much that
94 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
95 * That is what this function does. The RB-tree is ordered by LEB number and
96 * offset because they uniquely identify the old index node.
97 */
98static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
99{
100 struct ubifs_old_idx *old_idx;
101
102 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
103 if (unlikely(!old_idx))
104 return -ENOMEM;
105 old_idx->lnum = lnum;
106 old_idx->offs = offs;
107 do_insert_old_idx(c, old_idx);
108
109 return 0;
110}
111
112/**
113 * insert_old_idx_znode - record a znode obsoleted since last commit start.
114 * @c: UBIFS file-system description object
115 * @znode: znode of obsoleted index node
116 *
117 * Returns %0 on success, and a negative error code on failure.
118 */
119int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
120{
121 if (znode->parent) {
122 struct ubifs_zbranch *zbr;
123
124 zbr = &znode->parent->zbranch[znode->iip];
125 if (zbr->len)
126 return insert_old_idx(c, zbr->lnum, zbr->offs);
127 } else
128 if (c->zroot.len)
129 return insert_old_idx(c, c->zroot.lnum,
130 c->zroot.offs);
131 return 0;
132}
133
134/**
135 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
136 * @c: UBIFS file-system description object
137 * @znode: znode of obsoleted index node
138 *
139 * Returns %0 on success, and a negative error code on failure.
140 */
141static int ins_clr_old_idx_znode(struct ubifs_info *c,
142 struct ubifs_znode *znode)
143{
144 int err;
145
146 if (znode->parent) {
147 struct ubifs_zbranch *zbr;
148
149 zbr = &znode->parent->zbranch[znode->iip];
150 if (zbr->len) {
151 err = insert_old_idx(c, zbr->lnum, zbr->offs);
152 if (err)
153 return err;
154 zbr->lnum = 0;
155 zbr->offs = 0;
156 zbr->len = 0;
157 }
158 } else
159 if (c->zroot.len) {
160 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
161 if (err)
162 return err;
163 c->zroot.lnum = 0;
164 c->zroot.offs = 0;
165 c->zroot.len = 0;
166 }
167 return 0;
168}
169
170/**
171 * destroy_old_idx - destroy the old_idx RB-tree.
172 * @c: UBIFS file-system description object
173 *
174 * During start commit, the old_idx RB-tree is used to avoid overwriting index
175 * nodes that were in the index last commit but have since been deleted. This
176 * is necessary for recovery i.e. the old index must be kept intact until the
177 * new index is successfully written. The old-idx RB-tree is used for the
178 * in-the-gaps method of writing index nodes and is destroyed every commit.
179 */
180void destroy_old_idx(struct ubifs_info *c)
181{
182 struct ubifs_old_idx *old_idx, *n;
183
184 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb)
185 kfree(old_idx);
186
187 c->old_idx = RB_ROOT;
188}
189
190/**
191 * copy_znode - copy a dirty znode.
192 * @c: UBIFS file-system description object
193 * @znode: znode to copy
194 *
195 * A dirty znode being committed may not be changed, so it is copied.
196 */
197static struct ubifs_znode *copy_znode(struct ubifs_info *c,
198 struct ubifs_znode *znode)
199{
200 struct ubifs_znode *zn;
201
202 zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS);
203 if (unlikely(!zn))
204 return ERR_PTR(-ENOMEM);
205
206 zn->cnext = NULL;
207 __set_bit(DIRTY_ZNODE, &zn->flags);
208 __clear_bit(COW_ZNODE, &zn->flags);
209
210 return zn;
211}
212
213/**
214 * add_idx_dirt - add dirt due to a dirty znode.
215 * @c: UBIFS file-system description object
216 * @lnum: LEB number of index node
217 * @dirt: size of index node
218 *
219 * This function updates lprops dirty space and the new size of the index.
220 */
221static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
222{
223 c->calc_idx_sz -= ALIGN(dirt, 8);
224 return ubifs_add_dirt(c, lnum, dirt);
225}
226
227/**
228 * replace_znode - replace old znode with new znode.
229 * @c: UBIFS file-system description object
230 * @new_zn: new znode
231 * @old_zn: old znode
232 * @zbr: the branch of parent znode
233 *
234 * Replace old znode with new znode in TNC.
235 */
236static void replace_znode(struct ubifs_info *c, struct ubifs_znode *new_zn,
237 struct ubifs_znode *old_zn, struct ubifs_zbranch *zbr)
238{
239 ubifs_assert(c, !ubifs_zn_obsolete(old_zn));
240 __set_bit(OBSOLETE_ZNODE, &old_zn->flags);
241
242 if (old_zn->level != 0) {
243 int i;
244 const int n = new_zn->child_cnt;
245
246 /* The children now have new parent */
247 for (i = 0; i < n; i++) {
248 struct ubifs_zbranch *child = &new_zn->zbranch[i];
249
250 if (child->znode)
251 child->znode->parent = new_zn;
252 }
253 }
254
255 zbr->znode = new_zn;
256 zbr->lnum = 0;
257 zbr->offs = 0;
258 zbr->len = 0;
259
260 atomic_long_inc(&c->dirty_zn_cnt);
261}
262
263/**
264 * dirty_cow_znode - ensure a znode is not being committed.
265 * @c: UBIFS file-system description object
266 * @zbr: branch of znode to check
267 *
268 * Returns dirtied znode on success or negative error code on failure.
269 */
270static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
271 struct ubifs_zbranch *zbr)
272{
273 struct ubifs_znode *znode = zbr->znode;
274 struct ubifs_znode *zn;
275 int err;
276
277 if (!ubifs_zn_cow(znode)) {
278 /* znode is not being committed */
279 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
280 atomic_long_inc(&c->dirty_zn_cnt);
281 atomic_long_dec(&c->clean_zn_cnt);
282 atomic_long_dec(&ubifs_clean_zn_cnt);
283 err = add_idx_dirt(c, zbr->lnum, zbr->len);
284 if (unlikely(err))
285 return ERR_PTR(err);
286 }
287 return znode;
288 }
289
290 zn = copy_znode(c, znode);
291 if (IS_ERR(zn))
292 return zn;
293
294 if (zbr->len) {
295 struct ubifs_old_idx *old_idx;
296
297 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
298 if (unlikely(!old_idx)) {
299 err = -ENOMEM;
300 goto out;
301 }
302 old_idx->lnum = zbr->lnum;
303 old_idx->offs = zbr->offs;
304
305 err = add_idx_dirt(c, zbr->lnum, zbr->len);
306 if (err) {
307 kfree(old_idx);
308 goto out;
309 }
310
311 do_insert_old_idx(c, old_idx);
312 }
313
314 replace_znode(c, zn, znode, zbr);
315
316 return zn;
317
318out:
319 kfree(zn);
320 return ERR_PTR(err);
321}
322
323/**
324 * lnc_add - add a leaf node to the leaf node cache.
325 * @c: UBIFS file-system description object
326 * @zbr: zbranch of leaf node
327 * @node: leaf node
328 *
329 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
330 * purpose of the leaf node cache is to save re-reading the same leaf node over
331 * and over again. Most things are cached by VFS, however the file system must
332 * cache directory entries for readdir and for resolving hash collisions. The
333 * present implementation of the leaf node cache is extremely simple, and
334 * allows for error returns that are not used but that may be needed if a more
335 * complex implementation is created.
336 *
337 * Note, this function does not add the @node object to LNC directly, but
338 * allocates a copy of the object and adds the copy to LNC. The reason for this
339 * is that @node has been allocated outside of the TNC subsystem and will be
340 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
341 * may be changed at any time, e.g. freed by the shrinker.
342 */
343static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
344 const void *node)
345{
346 int err;
347 void *lnc_node;
348 const struct ubifs_dent_node *dent = node;
349
350 ubifs_assert(c, !zbr->leaf);
351 ubifs_assert(c, zbr->len != 0);
352 ubifs_assert(c, is_hash_key(c, &zbr->key));
353
354 err = ubifs_validate_entry(c, dent);
355 if (err) {
356 dump_stack();
357 ubifs_dump_node(c, dent, zbr->len);
358 return err;
359 }
360
361 lnc_node = kmemdup(node, zbr->len, GFP_NOFS);
362 if (!lnc_node)
363 /* We don't have to have the cache, so no error */
364 return 0;
365
366 zbr->leaf = lnc_node;
367 return 0;
368}
369
370 /**
371 * lnc_add_directly - add a leaf node to the leaf-node-cache.
372 * @c: UBIFS file-system description object
373 * @zbr: zbranch of leaf node
374 * @node: leaf node
375 *
376 * This function is similar to 'lnc_add()', but it does not create a copy of
377 * @node but inserts @node to TNC directly.
378 */
379static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
380 void *node)
381{
382 int err;
383
384 ubifs_assert(c, !zbr->leaf);
385 ubifs_assert(c, zbr->len != 0);
386
387 err = ubifs_validate_entry(c, node);
388 if (err) {
389 dump_stack();
390 ubifs_dump_node(c, node, zbr->len);
391 return err;
392 }
393
394 zbr->leaf = node;
395 return 0;
396}
397
398/**
399 * lnc_free - remove a leaf node from the leaf node cache.
400 * @zbr: zbranch of leaf node
401 */
402static void lnc_free(struct ubifs_zbranch *zbr)
403{
404 if (!zbr->leaf)
405 return;
406 kfree(zbr->leaf);
407 zbr->leaf = NULL;
408}
409
410/**
411 * tnc_read_hashed_node - read a "hashed" leaf node.
412 * @c: UBIFS file-system description object
413 * @zbr: key and position of the node
414 * @node: node is returned here
415 *
416 * This function reads a "hashed" node defined by @zbr from the leaf node cache
417 * (in it is there) or from the hash media, in which case the node is also
418 * added to LNC. Returns zero in case of success or a negative error
419 * code in case of failure.
420 */
421static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr,
422 void *node)
423{
424 int err;
425
426 ubifs_assert(c, is_hash_key(c, &zbr->key));
427
428 if (zbr->leaf) {
429 /* Read from the leaf node cache */
430 ubifs_assert(c, zbr->len != 0);
431 memcpy(node, zbr->leaf, zbr->len);
432 return 0;
433 }
434
435 if (c->replaying) {
436 err = fallible_read_node(c, &zbr->key, zbr, node);
437 /*
438 * When the node was not found, return -ENOENT, 0 otherwise.
439 * Negative return codes stay as-is.
440 */
441 if (err == 0)
442 err = -ENOENT;
443 else if (err == 1)
444 err = 0;
445 } else {
446 err = ubifs_tnc_read_node(c, zbr, node);
447 }
448 if (err)
449 return err;
450
451 /* Add the node to the leaf node cache */
452 err = lnc_add(c, zbr, node);
453 return err;
454}
455
456/**
457 * try_read_node - read a node if it is a node.
458 * @c: UBIFS file-system description object
459 * @buf: buffer to read to
460 * @type: node type
461 * @zbr: the zbranch describing the node to read
462 *
463 * This function tries to read a node of known type and length, checks it and
464 * stores it in @buf. This function returns %1 if a node is present and %0 if
465 * a node is not present. A negative error code is returned for I/O errors.
466 * This function performs that same function as ubifs_read_node except that
467 * it does not require that there is actually a node present and instead
468 * the return code indicates if a node was read.
469 *
470 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
471 * is true (it is controlled by corresponding mount option). However, if
472 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
473 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
474 * because during mounting or re-mounting from R/O mode to R/W mode we may read
475 * journal nodes (when replying the journal or doing the recovery) and the
476 * journal nodes may potentially be corrupted, so checking is required.
477 */
478static int try_read_node(const struct ubifs_info *c, void *buf, int type,
479 struct ubifs_zbranch *zbr)
480{
481 int len = zbr->len;
482 int lnum = zbr->lnum;
483 int offs = zbr->offs;
484 int err, node_len;
485 struct ubifs_ch *ch = buf;
486 uint32_t crc, node_crc;
487
488 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
489
490 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
491 if (err) {
492 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d",
493 type, lnum, offs, err);
494 return err;
495 }
496
497 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
498 return 0;
499
500 if (ch->node_type != type)
501 return 0;
502
503 node_len = le32_to_cpu(ch->len);
504 if (node_len != len)
505 return 0;
506
507 if (type != UBIFS_DATA_NODE || !c->no_chk_data_crc || c->mounting ||
508 c->remounting_rw) {
509 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
510 node_crc = le32_to_cpu(ch->crc);
511 if (crc != node_crc)
512 return 0;
513 }
514
515 err = ubifs_node_check_hash(c, buf, zbr->hash);
516 if (err) {
517 ubifs_bad_hash(c, buf, zbr->hash, lnum, offs);
518 return 0;
519 }
520
521 return 1;
522}
523
524/**
525 * fallible_read_node - try to read a leaf node.
526 * @c: UBIFS file-system description object
527 * @key: key of node to read
528 * @zbr: position of node
529 * @node: node returned
530 *
531 * This function tries to read a node and returns %1 if the node is read, %0
532 * if the node is not present, and a negative error code in the case of error.
533 */
534static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
535 struct ubifs_zbranch *zbr, void *node)
536{
537 int ret;
538
539 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs);
540
541 ret = try_read_node(c, node, key_type(c, key), zbr);
542 if (ret == 1) {
543 union ubifs_key node_key;
544 struct ubifs_dent_node *dent = node;
545
546 /* All nodes have key in the same place */
547 key_read(c, &dent->key, &node_key);
548 if (keys_cmp(c, key, &node_key) != 0)
549 ret = 0;
550 }
551 if (ret == 0 && c->replaying)
552 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ",
553 zbr->lnum, zbr->offs, zbr->len);
554 return ret;
555}
556
557/**
558 * matches_name - determine if a direntry or xattr entry matches a given name.
559 * @c: UBIFS file-system description object
560 * @zbr: zbranch of dent
561 * @nm: name to match
562 *
563 * This function checks if xentry/direntry referred by zbranch @zbr matches name
564 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
565 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
566 * of failure, a negative error code is returned.
567 */
568static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
569 const struct fscrypt_name *nm)
570{
571 struct ubifs_dent_node *dent;
572 int nlen, err;
573
574 /* If possible, match against the dent in the leaf node cache */
575 if (!zbr->leaf) {
576 dent = kmalloc(zbr->len, GFP_NOFS);
577 if (!dent)
578 return -ENOMEM;
579
580 err = ubifs_tnc_read_node(c, zbr, dent);
581 if (err)
582 goto out_free;
583
584 /* Add the node to the leaf node cache */
585 err = lnc_add_directly(c, zbr, dent);
586 if (err)
587 goto out_free;
588 } else
589 dent = zbr->leaf;
590
591 nlen = le16_to_cpu(dent->nlen);
592 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
593 if (err == 0) {
594 if (nlen == fname_len(nm))
595 return NAME_MATCHES;
596 else if (nlen < fname_len(nm))
597 return NAME_LESS;
598 else
599 return NAME_GREATER;
600 } else if (err < 0)
601 return NAME_LESS;
602 else
603 return NAME_GREATER;
604
605out_free:
606 kfree(dent);
607 return err;
608}
609
610/**
611 * get_znode - get a TNC znode that may not be loaded yet.
612 * @c: UBIFS file-system description object
613 * @znode: parent znode
614 * @n: znode branch slot number
615 *
616 * This function returns the znode or a negative error code.
617 */
618static struct ubifs_znode *get_znode(struct ubifs_info *c,
619 struct ubifs_znode *znode, int n)
620{
621 struct ubifs_zbranch *zbr;
622
623 zbr = &znode->zbranch[n];
624 if (zbr->znode)
625 znode = zbr->znode;
626 else
627 znode = ubifs_load_znode(c, zbr, znode, n);
628 return znode;
629}
630
631/**
632 * tnc_next - find next TNC entry.
633 * @c: UBIFS file-system description object
634 * @zn: znode is passed and returned here
635 * @n: znode branch slot number is passed and returned here
636 *
637 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
638 * no next entry, or a negative error code otherwise.
639 */
640static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
641{
642 struct ubifs_znode *znode = *zn;
643 int nn = *n;
644
645 nn += 1;
646 if (nn < znode->child_cnt) {
647 *n = nn;
648 return 0;
649 }
650 while (1) {
651 struct ubifs_znode *zp;
652
653 zp = znode->parent;
654 if (!zp)
655 return -ENOENT;
656 nn = znode->iip + 1;
657 znode = zp;
658 if (nn < znode->child_cnt) {
659 znode = get_znode(c, znode, nn);
660 if (IS_ERR(znode))
661 return PTR_ERR(znode);
662 while (znode->level != 0) {
663 znode = get_znode(c, znode, 0);
664 if (IS_ERR(znode))
665 return PTR_ERR(znode);
666 }
667 nn = 0;
668 break;
669 }
670 }
671 *zn = znode;
672 *n = nn;
673 return 0;
674}
675
676/**
677 * tnc_prev - find previous TNC entry.
678 * @c: UBIFS file-system description object
679 * @zn: znode is returned here
680 * @n: znode branch slot number is passed and returned here
681 *
682 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
683 * there is no next entry, or a negative error code otherwise.
684 */
685static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
686{
687 struct ubifs_znode *znode = *zn;
688 int nn = *n;
689
690 if (nn > 0) {
691 *n = nn - 1;
692 return 0;
693 }
694 while (1) {
695 struct ubifs_znode *zp;
696
697 zp = znode->parent;
698 if (!zp)
699 return -ENOENT;
700 nn = znode->iip - 1;
701 znode = zp;
702 if (nn >= 0) {
703 znode = get_znode(c, znode, nn);
704 if (IS_ERR(znode))
705 return PTR_ERR(znode);
706 while (znode->level != 0) {
707 nn = znode->child_cnt - 1;
708 znode = get_znode(c, znode, nn);
709 if (IS_ERR(znode))
710 return PTR_ERR(znode);
711 }
712 nn = znode->child_cnt - 1;
713 break;
714 }
715 }
716 *zn = znode;
717 *n = nn;
718 return 0;
719}
720
721/**
722 * resolve_collision - resolve a collision.
723 * @c: UBIFS file-system description object
724 * @key: key of a directory or extended attribute entry
725 * @zn: znode is returned here
726 * @n: zbranch number is passed and returned here
727 * @nm: name of the entry
728 *
729 * This function is called for "hashed" keys to make sure that the found key
730 * really corresponds to the looked up node (directory or extended attribute
731 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
732 * %0 is returned if @nm is not found and @zn and @n are set to the previous
733 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
734 * This means that @n may be set to %-1 if the leftmost key in @zn is the
735 * previous one. A negative error code is returned on failures.
736 */
737static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
738 struct ubifs_znode **zn, int *n,
739 const struct fscrypt_name *nm)
740{
741 int err;
742
743 err = matches_name(c, &(*zn)->zbranch[*n], nm);
744 if (unlikely(err < 0))
745 return err;
746 if (err == NAME_MATCHES)
747 return 1;
748
749 if (err == NAME_GREATER) {
750 /* Look left */
751 while (1) {
752 err = tnc_prev(c, zn, n);
753 if (err == -ENOENT) {
754 ubifs_assert(c, *n == 0);
755 *n = -1;
756 return 0;
757 }
758 if (err < 0)
759 return err;
760 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
761 /*
762 * We have found the branch after which we would
763 * like to insert, but inserting in this znode
764 * may still be wrong. Consider the following 3
765 * znodes, in the case where we are resolving a
766 * collision with Key2.
767 *
768 * znode zp
769 * ----------------------
770 * level 1 | Key0 | Key1 |
771 * -----------------------
772 * | |
773 * znode za | | znode zb
774 * ------------ ------------
775 * level 0 | Key0 | | Key2 |
776 * ------------ ------------
777 *
778 * The lookup finds Key2 in znode zb. Lets say
779 * there is no match and the name is greater so
780 * we look left. When we find Key0, we end up
781 * here. If we return now, we will insert into
782 * znode za at slot n = 1. But that is invalid
783 * according to the parent's keys. Key2 must
784 * be inserted into znode zb.
785 *
786 * Note, this problem is not relevant for the
787 * case when we go right, because
788 * 'tnc_insert()' would correct the parent key.
789 */
790 if (*n == (*zn)->child_cnt - 1) {
791 err = tnc_next(c, zn, n);
792 if (err) {
793 /* Should be impossible */
794 ubifs_assert(c, 0);
795 if (err == -ENOENT)
796 err = -EINVAL;
797 return err;
798 }
799 ubifs_assert(c, *n == 0);
800 *n = -1;
801 }
802 return 0;
803 }
804 err = matches_name(c, &(*zn)->zbranch[*n], nm);
805 if (err < 0)
806 return err;
807 if (err == NAME_LESS)
808 return 0;
809 if (err == NAME_MATCHES)
810 return 1;
811 ubifs_assert(c, err == NAME_GREATER);
812 }
813 } else {
814 int nn = *n;
815 struct ubifs_znode *znode = *zn;
816
817 /* Look right */
818 while (1) {
819 err = tnc_next(c, &znode, &nn);
820 if (err == -ENOENT)
821 return 0;
822 if (err < 0)
823 return err;
824 if (keys_cmp(c, &znode->zbranch[nn].key, key))
825 return 0;
826 err = matches_name(c, &znode->zbranch[nn], nm);
827 if (err < 0)
828 return err;
829 if (err == NAME_GREATER)
830 return 0;
831 *zn = znode;
832 *n = nn;
833 if (err == NAME_MATCHES)
834 return 1;
835 ubifs_assert(c, err == NAME_LESS);
836 }
837 }
838}
839
840/**
841 * fallible_matches_name - determine if a dent matches a given name.
842 * @c: UBIFS file-system description object
843 * @zbr: zbranch of dent
844 * @nm: name to match
845 *
846 * This is a "fallible" version of 'matches_name()' function which does not
847 * panic if the direntry/xentry referred by @zbr does not exist on the media.
848 *
849 * This function checks if xentry/direntry referred by zbranch @zbr matches name
850 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
851 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
852 * if xentry/direntry referred by @zbr does not exist on the media. A negative
853 * error code is returned in case of failure.
854 */
855static int fallible_matches_name(struct ubifs_info *c,
856 struct ubifs_zbranch *zbr,
857 const struct fscrypt_name *nm)
858{
859 struct ubifs_dent_node *dent;
860 int nlen, err;
861
862 /* If possible, match against the dent in the leaf node cache */
863 if (!zbr->leaf) {
864 dent = kmalloc(zbr->len, GFP_NOFS);
865 if (!dent)
866 return -ENOMEM;
867
868 err = fallible_read_node(c, &zbr->key, zbr, dent);
869 if (err < 0)
870 goto out_free;
871 if (err == 0) {
872 /* The node was not present */
873 err = NOT_ON_MEDIA;
874 goto out_free;
875 }
876 ubifs_assert(c, err == 1);
877
878 err = lnc_add_directly(c, zbr, dent);
879 if (err)
880 goto out_free;
881 } else
882 dent = zbr->leaf;
883
884 nlen = le16_to_cpu(dent->nlen);
885 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm)));
886 if (err == 0) {
887 if (nlen == fname_len(nm))
888 return NAME_MATCHES;
889 else if (nlen < fname_len(nm))
890 return NAME_LESS;
891 else
892 return NAME_GREATER;
893 } else if (err < 0)
894 return NAME_LESS;
895 else
896 return NAME_GREATER;
897
898out_free:
899 kfree(dent);
900 return err;
901}
902
903/**
904 * fallible_resolve_collision - resolve a collision even if nodes are missing.
905 * @c: UBIFS file-system description object
906 * @key: key
907 * @zn: znode is returned here
908 * @n: branch number is passed and returned here
909 * @nm: name of directory entry
910 * @adding: indicates caller is adding a key to the TNC
911 *
912 * This is a "fallible" version of the 'resolve_collision()' function which
913 * does not panic if one of the nodes referred to by TNC does not exist on the
914 * media. This may happen when replaying the journal if a deleted node was
915 * Garbage-collected and the commit was not done. A branch that refers to a node
916 * that is not present is called a dangling branch. The following are the return
917 * codes for this function:
918 * o if @nm was found, %1 is returned and @zn and @n are set to the found
919 * branch;
920 * o if we are @adding and @nm was not found, %0 is returned;
921 * o if we are not @adding and @nm was not found, but a dangling branch was
922 * found, then %1 is returned and @zn and @n are set to the dangling branch;
923 * o a negative error code is returned in case of failure.
924 */
925static int fallible_resolve_collision(struct ubifs_info *c,
926 const union ubifs_key *key,
927 struct ubifs_znode **zn, int *n,
928 const struct fscrypt_name *nm,
929 int adding)
930{
931 struct ubifs_znode *o_znode = NULL, *znode = *zn;
932 int o_n, err, cmp, unsure = 0, nn = *n;
933
934 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
935 if (unlikely(cmp < 0))
936 return cmp;
937 if (cmp == NAME_MATCHES)
938 return 1;
939 if (cmp == NOT_ON_MEDIA) {
940 o_znode = znode;
941 o_n = nn;
942 /*
943 * We are unlucky and hit a dangling branch straight away.
944 * Now we do not really know where to go to find the needed
945 * branch - to the left or to the right. Well, let's try left.
946 */
947 unsure = 1;
948 } else if (!adding)
949 unsure = 1; /* Remove a dangling branch wherever it is */
950
951 if (cmp == NAME_GREATER || unsure) {
952 /* Look left */
953 while (1) {
954 err = tnc_prev(c, zn, n);
955 if (err == -ENOENT) {
956 ubifs_assert(c, *n == 0);
957 *n = -1;
958 break;
959 }
960 if (err < 0)
961 return err;
962 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
963 /* See comments in 'resolve_collision()' */
964 if (*n == (*zn)->child_cnt - 1) {
965 err = tnc_next(c, zn, n);
966 if (err) {
967 /* Should be impossible */
968 ubifs_assert(c, 0);
969 if (err == -ENOENT)
970 err = -EINVAL;
971 return err;
972 }
973 ubifs_assert(c, *n == 0);
974 *n = -1;
975 }
976 break;
977 }
978 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
979 if (err < 0)
980 return err;
981 if (err == NAME_MATCHES)
982 return 1;
983 if (err == NOT_ON_MEDIA) {
984 o_znode = *zn;
985 o_n = *n;
986 continue;
987 }
988 if (!adding)
989 continue;
990 if (err == NAME_LESS)
991 break;
992 else
993 unsure = 0;
994 }
995 }
996
997 if (cmp == NAME_LESS || unsure) {
998 /* Look right */
999 *zn = znode;
1000 *n = nn;
1001 while (1) {
1002 err = tnc_next(c, &znode, &nn);
1003 if (err == -ENOENT)
1004 break;
1005 if (err < 0)
1006 return err;
1007 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1008 break;
1009 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
1010 if (err < 0)
1011 return err;
1012 if (err == NAME_GREATER)
1013 break;
1014 *zn = znode;
1015 *n = nn;
1016 if (err == NAME_MATCHES)
1017 return 1;
1018 if (err == NOT_ON_MEDIA) {
1019 o_znode = znode;
1020 o_n = nn;
1021 }
1022 }
1023 }
1024
1025 /* Never match a dangling branch when adding */
1026 if (adding || !o_znode)
1027 return 0;
1028
1029 dbg_mntk(key, "dangling match LEB %d:%d len %d key ",
1030 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1031 o_znode->zbranch[o_n].len);
1032 *zn = o_znode;
1033 *n = o_n;
1034 return 1;
1035}
1036
1037/**
1038 * matches_position - determine if a zbranch matches a given position.
1039 * @zbr: zbranch of dent
1040 * @lnum: LEB number of dent to match
1041 * @offs: offset of dent to match
1042 *
1043 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1044 */
1045static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1046{
1047 if (zbr->lnum == lnum && zbr->offs == offs)
1048 return 1;
1049 else
1050 return 0;
1051}
1052
1053/**
1054 * resolve_collision_directly - resolve a collision directly.
1055 * @c: UBIFS file-system description object
1056 * @key: key of directory entry
1057 * @zn: znode is passed and returned here
1058 * @n: zbranch number is passed and returned here
1059 * @lnum: LEB number of dent node to match
1060 * @offs: offset of dent node to match
1061 *
1062 * This function is used for "hashed" keys to make sure the found directory or
1063 * extended attribute entry node is what was looked for. It is used when the
1064 * flash address of the right node is known (@lnum:@offs) which makes it much
1065 * easier to resolve collisions (no need to read entries and match full
1066 * names). This function returns %1 and sets @zn and @n if the collision is
1067 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1068 * previous directory entry. Otherwise a negative error code is returned.
1069 */
1070static int resolve_collision_directly(struct ubifs_info *c,
1071 const union ubifs_key *key,
1072 struct ubifs_znode **zn, int *n,
1073 int lnum, int offs)
1074{
1075 struct ubifs_znode *znode;
1076 int nn, err;
1077
1078 znode = *zn;
1079 nn = *n;
1080 if (matches_position(&znode->zbranch[nn], lnum, offs))
1081 return 1;
1082
1083 /* Look left */
1084 while (1) {
1085 err = tnc_prev(c, &znode, &nn);
1086 if (err == -ENOENT)
1087 break;
1088 if (err < 0)
1089 return err;
1090 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1091 break;
1092 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1093 *zn = znode;
1094 *n = nn;
1095 return 1;
1096 }
1097 }
1098
1099 /* Look right */
1100 znode = *zn;
1101 nn = *n;
1102 while (1) {
1103 err = tnc_next(c, &znode, &nn);
1104 if (err == -ENOENT)
1105 return 0;
1106 if (err < 0)
1107 return err;
1108 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1109 return 0;
1110 *zn = znode;
1111 *n = nn;
1112 if (matches_position(&znode->zbranch[nn], lnum, offs))
1113 return 1;
1114 }
1115}
1116
1117/**
1118 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1119 * @c: UBIFS file-system description object
1120 * @znode: znode to dirty
1121 *
1122 * If we do not have a unique key that resides in a znode, then we cannot
1123 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1124 * This function records the path back to the last dirty ancestor, and then
1125 * dirties the znodes on that path.
1126 */
1127static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1128 struct ubifs_znode *znode)
1129{
1130 struct ubifs_znode *zp;
1131 int *path = c->bottom_up_buf, p = 0;
1132
1133 ubifs_assert(c, c->zroot.znode);
1134 ubifs_assert(c, znode);
1135 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1136 kfree(c->bottom_up_buf);
1137 c->bottom_up_buf = kmalloc_array(c->zroot.znode->level,
1138 sizeof(int),
1139 GFP_NOFS);
1140 if (!c->bottom_up_buf)
1141 return ERR_PTR(-ENOMEM);
1142 path = c->bottom_up_buf;
1143 }
1144 if (c->zroot.znode->level) {
1145 /* Go up until parent is dirty */
1146 while (1) {
1147 int n;
1148
1149 zp = znode->parent;
1150 if (!zp)
1151 break;
1152 n = znode->iip;
1153 ubifs_assert(c, p < c->zroot.znode->level);
1154 path[p++] = n;
1155 if (!zp->cnext && ubifs_zn_dirty(znode))
1156 break;
1157 znode = zp;
1158 }
1159 }
1160
1161 /* Come back down, dirtying as we go */
1162 while (1) {
1163 struct ubifs_zbranch *zbr;
1164
1165 zp = znode->parent;
1166 if (zp) {
1167 ubifs_assert(c, path[p - 1] >= 0);
1168 ubifs_assert(c, path[p - 1] < zp->child_cnt);
1169 zbr = &zp->zbranch[path[--p]];
1170 znode = dirty_cow_znode(c, zbr);
1171 } else {
1172 ubifs_assert(c, znode == c->zroot.znode);
1173 znode = dirty_cow_znode(c, &c->zroot);
1174 }
1175 if (IS_ERR(znode) || !p)
1176 break;
1177 ubifs_assert(c, path[p - 1] >= 0);
1178 ubifs_assert(c, path[p - 1] < znode->child_cnt);
1179 znode = znode->zbranch[path[p - 1]].znode;
1180 }
1181
1182 return znode;
1183}
1184
1185/**
1186 * ubifs_lookup_level0 - search for zero-level znode.
1187 * @c: UBIFS file-system description object
1188 * @key: key to lookup
1189 * @zn: znode is returned here
1190 * @n: znode branch slot number is returned here
1191 *
1192 * This function looks up the TNC tree and search for zero-level znode which
1193 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1194 * cases:
1195 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1196 * is returned and slot number of the matched branch is stored in @n;
1197 * o not exact match, which means that zero-level znode does not contain
1198 * @key, then %0 is returned and slot number of the closest branch or %-1
1199 * is stored in @n; In this case calling tnc_next() is mandatory.
1200 * o @key is so small that it is even less than the lowest key of the
1201 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1202 *
1203 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1204 * function reads corresponding indexing nodes and inserts them to TNC. In
1205 * case of failure, a negative error code is returned.
1206 */
1207int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1208 struct ubifs_znode **zn, int *n)
1209{
1210 int err, exact;
1211 struct ubifs_znode *znode;
1212 time64_t time = ktime_get_seconds();
1213
1214 dbg_tnck(key, "search key ");
1215 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
1216
1217 znode = c->zroot.znode;
1218 if (unlikely(!znode)) {
1219 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1220 if (IS_ERR(znode))
1221 return PTR_ERR(znode);
1222 }
1223
1224 znode->time = time;
1225
1226 while (1) {
1227 struct ubifs_zbranch *zbr;
1228
1229 exact = ubifs_search_zbranch(c, znode, key, n);
1230
1231 if (znode->level == 0)
1232 break;
1233
1234 if (*n < 0)
1235 *n = 0;
1236 zbr = &znode->zbranch[*n];
1237
1238 if (zbr->znode) {
1239 znode->time = time;
1240 znode = zbr->znode;
1241 continue;
1242 }
1243
1244 /* znode is not in TNC cache, load it from the media */
1245 znode = ubifs_load_znode(c, zbr, znode, *n);
1246 if (IS_ERR(znode))
1247 return PTR_ERR(znode);
1248 }
1249
1250 *zn = znode;
1251 if (exact || !is_hash_key(c, key) || *n != -1) {
1252 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1253 return exact;
1254 }
1255
1256 /*
1257 * Here is a tricky place. We have not found the key and this is a
1258 * "hashed" key, which may collide. The rest of the code deals with
1259 * situations like this:
1260 *
1261 * | 3 | 5 |
1262 * / \
1263 * | 3 | 5 | | 6 | 7 | (x)
1264 *
1265 * Or more a complex example:
1266 *
1267 * | 1 | 5 |
1268 * / \
1269 * | 1 | 3 | | 5 | 8 |
1270 * \ /
1271 * | 5 | 5 | | 6 | 7 | (x)
1272 *
1273 * In the examples, if we are looking for key "5", we may reach nodes
1274 * marked with "(x)". In this case what we have do is to look at the
1275 * left and see if there is "5" key there. If there is, we have to
1276 * return it.
1277 *
1278 * Note, this whole situation is possible because we allow to have
1279 * elements which are equivalent to the next key in the parent in the
1280 * children of current znode. For example, this happens if we split a
1281 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1282 * like this:
1283 * | 3 | 5 |
1284 * / \
1285 * | 3 | 5 | | 5 | 6 | 7 |
1286 * ^
1287 * And this becomes what is at the first "picture" after key "5" marked
1288 * with "^" is removed. What could be done is we could prohibit
1289 * splitting in the middle of the colliding sequence. Also, when
1290 * removing the leftmost key, we would have to correct the key of the
1291 * parent node, which would introduce additional complications. Namely,
1292 * if we changed the leftmost key of the parent znode, the garbage
1293 * collector would be unable to find it (GC is doing this when GC'ing
1294 * indexing LEBs). Although we already have an additional RB-tree where
1295 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1296 * after the commit. But anyway, this does not look easy to implement
1297 * so we did not try this.
1298 */
1299 err = tnc_prev(c, &znode, n);
1300 if (err == -ENOENT) {
1301 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1302 *n = -1;
1303 return 0;
1304 }
1305 if (unlikely(err < 0))
1306 return err;
1307 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1308 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1309 *n = -1;
1310 return 0;
1311 }
1312
1313 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1314 *zn = znode;
1315 return 1;
1316}
1317
1318/**
1319 * lookup_level0_dirty - search for zero-level znode dirtying.
1320 * @c: UBIFS file-system description object
1321 * @key: key to lookup
1322 * @zn: znode is returned here
1323 * @n: znode branch slot number is returned here
1324 *
1325 * This function looks up the TNC tree and search for zero-level znode which
1326 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1327 * cases:
1328 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1329 * is returned and slot number of the matched branch is stored in @n;
1330 * o not exact match, which means that zero-level znode does not contain @key
1331 * then %0 is returned and slot number of the closed branch is stored in
1332 * @n;
1333 * o @key is so small that it is even less than the lowest key of the
1334 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1335 *
1336 * Additionally all znodes in the path from the root to the located zero-level
1337 * znode are marked as dirty.
1338 *
1339 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1340 * function reads corresponding indexing nodes and inserts them to TNC. In
1341 * case of failure, a negative error code is returned.
1342 */
1343static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1344 struct ubifs_znode **zn, int *n)
1345{
1346 int err, exact;
1347 struct ubifs_znode *znode;
1348 time64_t time = ktime_get_seconds();
1349
1350 dbg_tnck(key, "search and dirty key ");
1351
1352 znode = c->zroot.znode;
1353 if (unlikely(!znode)) {
1354 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1355 if (IS_ERR(znode))
1356 return PTR_ERR(znode);
1357 }
1358
1359 znode = dirty_cow_znode(c, &c->zroot);
1360 if (IS_ERR(znode))
1361 return PTR_ERR(znode);
1362
1363 znode->time = time;
1364
1365 while (1) {
1366 struct ubifs_zbranch *zbr;
1367
1368 exact = ubifs_search_zbranch(c, znode, key, n);
1369
1370 if (znode->level == 0)
1371 break;
1372
1373 if (*n < 0)
1374 *n = 0;
1375 zbr = &znode->zbranch[*n];
1376
1377 if (zbr->znode) {
1378 znode->time = time;
1379 znode = dirty_cow_znode(c, zbr);
1380 if (IS_ERR(znode))
1381 return PTR_ERR(znode);
1382 continue;
1383 }
1384
1385 /* znode is not in TNC cache, load it from the media */
1386 znode = ubifs_load_znode(c, zbr, znode, *n);
1387 if (IS_ERR(znode))
1388 return PTR_ERR(znode);
1389 znode = dirty_cow_znode(c, zbr);
1390 if (IS_ERR(znode))
1391 return PTR_ERR(znode);
1392 }
1393
1394 *zn = znode;
1395 if (exact || !is_hash_key(c, key) || *n != -1) {
1396 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1397 return exact;
1398 }
1399
1400 /*
1401 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1402 * code.
1403 */
1404 err = tnc_prev(c, &znode, n);
1405 if (err == -ENOENT) {
1406 *n = -1;
1407 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1408 return 0;
1409 }
1410 if (unlikely(err < 0))
1411 return err;
1412 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1413 *n = -1;
1414 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1415 return 0;
1416 }
1417
1418 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1419 znode = dirty_cow_bottom_up(c, znode);
1420 if (IS_ERR(znode))
1421 return PTR_ERR(znode);
1422 }
1423
1424 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1425 *zn = znode;
1426 return 1;
1427}
1428
1429/**
1430 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1431 * @c: UBIFS file-system description object
1432 * @lnum: LEB number
1433 * @gc_seq1: garbage collection sequence number
1434 *
1435 * This function determines if @lnum may have been garbage collected since
1436 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1437 * %0 is returned.
1438 */
1439static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1440{
1441 int gc_seq2, gced_lnum;
1442
1443 gced_lnum = c->gced_lnum;
1444 smp_rmb();
1445 gc_seq2 = c->gc_seq;
1446 /* Same seq means no GC */
1447 if (gc_seq1 == gc_seq2)
1448 return 0;
1449 /* Different by more than 1 means we don't know */
1450 if (gc_seq1 + 1 != gc_seq2)
1451 return 1;
1452 /*
1453 * We have seen the sequence number has increased by 1. Now we need to
1454 * be sure we read the right LEB number, so read it again.
1455 */
1456 smp_rmb();
1457 if (gced_lnum != c->gced_lnum)
1458 return 1;
1459 /* Finally we can check lnum */
1460 if (gced_lnum == lnum)
1461 return 1;
1462 return 0;
1463}
1464
1465/**
1466 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1467 * @c: UBIFS file-system description object
1468 * @key: node key to lookup
1469 * @node: the node is returned here
1470 * @lnum: LEB number is returned here
1471 * @offs: offset is returned here
1472 *
1473 * This function looks up and reads node with key @key. The caller has to make
1474 * sure the @node buffer is large enough to fit the node. Returns zero in case
1475 * of success, %-ENOENT if the node was not found, and a negative error code in
1476 * case of failure. The node location can be returned in @lnum and @offs.
1477 */
1478int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1479 void *node, int *lnum, int *offs)
1480{
1481 int found, n, err, safely = 0, gc_seq1;
1482 struct ubifs_znode *znode;
1483 struct ubifs_zbranch zbr, *zt;
1484
1485again:
1486 mutex_lock(&c->tnc_mutex);
1487 found = ubifs_lookup_level0(c, key, &znode, &n);
1488 if (!found) {
1489 err = -ENOENT;
1490 goto out;
1491 } else if (found < 0) {
1492 err = found;
1493 goto out;
1494 }
1495 zt = &znode->zbranch[n];
1496 if (lnum) {
1497 *lnum = zt->lnum;
1498 *offs = zt->offs;
1499 }
1500 if (is_hash_key(c, key)) {
1501 /*
1502 * In this case the leaf node cache gets used, so we pass the
1503 * address of the zbranch and keep the mutex locked
1504 */
1505 err = tnc_read_hashed_node(c, zt, node);
1506 goto out;
1507 }
1508 if (safely) {
1509 err = ubifs_tnc_read_node(c, zt, node);
1510 goto out;
1511 }
1512 /* Drop the TNC mutex prematurely and race with garbage collection */
1513 zbr = znode->zbranch[n];
1514 gc_seq1 = c->gc_seq;
1515 mutex_unlock(&c->tnc_mutex);
1516
1517 if (ubifs_get_wbuf(c, zbr.lnum)) {
1518 /* We do not GC journal heads */
1519 err = ubifs_tnc_read_node(c, &zbr, node);
1520 return err;
1521 }
1522
1523 err = fallible_read_node(c, key, &zbr, node);
1524 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1525 /*
1526 * The node may have been GC'ed out from under us so try again
1527 * while keeping the TNC mutex locked.
1528 */
1529 safely = 1;
1530 goto again;
1531 }
1532 return 0;
1533
1534out:
1535 mutex_unlock(&c->tnc_mutex);
1536 return err;
1537}
1538
1539/**
1540 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1541 * @c: UBIFS file-system description object
1542 * @bu: bulk-read parameters and results
1543 *
1544 * Lookup consecutive data node keys for the same inode that reside
1545 * consecutively in the same LEB. This function returns zero in case of success
1546 * and a negative error code in case of failure.
1547 *
1548 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1549 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1550 * maximum possible amount of nodes for bulk-read.
1551 */
1552int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1553{
1554 int n, err = 0, lnum = -1, offs;
1555 int len;
1556 unsigned int block = key_block(c, &bu->key);
1557 struct ubifs_znode *znode;
1558
1559 bu->cnt = 0;
1560 bu->blk_cnt = 0;
1561 bu->eof = 0;
1562
1563 mutex_lock(&c->tnc_mutex);
1564 /* Find first key */
1565 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1566 if (err < 0)
1567 goto out;
1568 if (err) {
1569 /* Key found */
1570 len = znode->zbranch[n].len;
1571 /* The buffer must be big enough for at least 1 node */
1572 if (len > bu->buf_len) {
1573 err = -EINVAL;
1574 goto out;
1575 }
1576 /* Add this key */
1577 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1578 bu->blk_cnt += 1;
1579 lnum = znode->zbranch[n].lnum;
1580 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1581 }
1582 while (1) {
1583 struct ubifs_zbranch *zbr;
1584 union ubifs_key *key;
1585 unsigned int next_block;
1586
1587 /* Find next key */
1588 err = tnc_next(c, &znode, &n);
1589 if (err)
1590 goto out;
1591 zbr = &znode->zbranch[n];
1592 key = &zbr->key;
1593 /* See if there is another data key for this file */
1594 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1595 key_type(c, key) != UBIFS_DATA_KEY) {
1596 err = -ENOENT;
1597 goto out;
1598 }
1599 if (lnum < 0) {
1600 /* First key found */
1601 lnum = zbr->lnum;
1602 offs = ALIGN(zbr->offs + zbr->len, 8);
1603 len = zbr->len;
1604 if (len > bu->buf_len) {
1605 err = -EINVAL;
1606 goto out;
1607 }
1608 } else {
1609 /*
1610 * The data nodes must be in consecutive positions in
1611 * the same LEB.
1612 */
1613 if (zbr->lnum != lnum || zbr->offs != offs)
1614 goto out;
1615 offs += ALIGN(zbr->len, 8);
1616 len = ALIGN(len, 8) + zbr->len;
1617 /* Must not exceed buffer length */
1618 if (len > bu->buf_len)
1619 goto out;
1620 }
1621 /* Allow for holes */
1622 next_block = key_block(c, key);
1623 bu->blk_cnt += (next_block - block - 1);
1624 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1625 goto out;
1626 block = next_block;
1627 /* Add this key */
1628 bu->zbranch[bu->cnt++] = *zbr;
1629 bu->blk_cnt += 1;
1630 /* See if we have room for more */
1631 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1632 goto out;
1633 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1634 goto out;
1635 }
1636out:
1637 if (err == -ENOENT) {
1638 bu->eof = 1;
1639 err = 0;
1640 }
1641 bu->gc_seq = c->gc_seq;
1642 mutex_unlock(&c->tnc_mutex);
1643 if (err)
1644 return err;
1645 /*
1646 * An enormous hole could cause bulk-read to encompass too many
1647 * page cache pages, so limit the number here.
1648 */
1649 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1650 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1651 /*
1652 * Ensure that bulk-read covers a whole number of page cache
1653 * pages.
1654 */
1655 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1656 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1657 return 0;
1658 if (bu->eof) {
1659 /* At the end of file we can round up */
1660 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1661 return 0;
1662 }
1663 /* Exclude data nodes that do not make up a whole page cache page */
1664 block = key_block(c, &bu->key) + bu->blk_cnt;
1665 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1666 while (bu->cnt) {
1667 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1668 break;
1669 bu->cnt -= 1;
1670 }
1671 return 0;
1672}
1673
1674/**
1675 * read_wbuf - bulk-read from a LEB with a wbuf.
1676 * @wbuf: wbuf that may overlap the read
1677 * @buf: buffer into which to read
1678 * @len: read length
1679 * @lnum: LEB number from which to read
1680 * @offs: offset from which to read
1681 *
1682 * This functions returns %0 on success or a negative error code on failure.
1683 */
1684static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1685 int offs)
1686{
1687 const struct ubifs_info *c = wbuf->c;
1688 int rlen, overlap;
1689
1690 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1691 ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1692 ubifs_assert(c, !(offs & 7) && offs < c->leb_size);
1693 ubifs_assert(c, offs + len <= c->leb_size);
1694
1695 spin_lock(&wbuf->lock);
1696 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1697 if (!overlap) {
1698 /* We may safely unlock the write-buffer and read the data */
1699 spin_unlock(&wbuf->lock);
1700 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1701 }
1702
1703 /* Don't read under wbuf */
1704 rlen = wbuf->offs - offs;
1705 if (rlen < 0)
1706 rlen = 0;
1707
1708 /* Copy the rest from the write-buffer */
1709 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1710 spin_unlock(&wbuf->lock);
1711
1712 if (rlen > 0)
1713 /* Read everything that goes before write-buffer */
1714 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1715
1716 return 0;
1717}
1718
1719/**
1720 * validate_data_node - validate data nodes for bulk-read.
1721 * @c: UBIFS file-system description object
1722 * @buf: buffer containing data node to validate
1723 * @zbr: zbranch of data node to validate
1724 *
1725 * This functions returns %0 on success or a negative error code on failure.
1726 */
1727static int validate_data_node(struct ubifs_info *c, void *buf,
1728 struct ubifs_zbranch *zbr)
1729{
1730 union ubifs_key key1;
1731 struct ubifs_ch *ch = buf;
1732 int err, len;
1733
1734 if (ch->node_type != UBIFS_DATA_NODE) {
1735 ubifs_err(c, "bad node type (%d but expected %d)",
1736 ch->node_type, UBIFS_DATA_NODE);
1737 goto out_err;
1738 }
1739
1740 err = ubifs_check_node(c, buf, zbr->len, zbr->lnum, zbr->offs, 0, 0);
1741 if (err) {
1742 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE);
1743 goto out;
1744 }
1745
1746 err = ubifs_node_check_hash(c, buf, zbr->hash);
1747 if (err) {
1748 ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs);
1749 return err;
1750 }
1751
1752 len = le32_to_cpu(ch->len);
1753 if (len != zbr->len) {
1754 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len);
1755 goto out_err;
1756 }
1757
1758 /* Make sure the key of the read node is correct */
1759 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1760 if (!keys_eq(c, &zbr->key, &key1)) {
1761 ubifs_err(c, "bad key in node at LEB %d:%d",
1762 zbr->lnum, zbr->offs);
1763 dbg_tnck(&zbr->key, "looked for key ");
1764 dbg_tnck(&key1, "found node's key ");
1765 goto out_err;
1766 }
1767
1768 return 0;
1769
1770out_err:
1771 err = -EINVAL;
1772out:
1773 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1774 ubifs_dump_node(c, buf, zbr->len);
1775 dump_stack();
1776 return err;
1777}
1778
1779/**
1780 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1781 * @c: UBIFS file-system description object
1782 * @bu: bulk-read parameters and results
1783 *
1784 * This functions reads and validates the data nodes that were identified by the
1785 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1786 * -EAGAIN to indicate a race with GC, or another negative error code on
1787 * failure.
1788 */
1789int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1790{
1791 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1792 struct ubifs_wbuf *wbuf;
1793 void *buf;
1794
1795 len = bu->zbranch[bu->cnt - 1].offs;
1796 len += bu->zbranch[bu->cnt - 1].len - offs;
1797 if (len > bu->buf_len) {
1798 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len);
1799 return -EINVAL;
1800 }
1801
1802 /* Do the read */
1803 wbuf = ubifs_get_wbuf(c, lnum);
1804 if (wbuf)
1805 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1806 else
1807 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0);
1808
1809 /* Check for a race with GC */
1810 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1811 return -EAGAIN;
1812
1813 if (err && err != -EBADMSG) {
1814 ubifs_err(c, "failed to read from LEB %d:%d, error %d",
1815 lnum, offs, err);
1816 dump_stack();
1817 dbg_tnck(&bu->key, "key ");
1818 return err;
1819 }
1820
1821 /* Validate the nodes read */
1822 buf = bu->buf;
1823 for (i = 0; i < bu->cnt; i++) {
1824 err = validate_data_node(c, buf, &bu->zbranch[i]);
1825 if (err)
1826 return err;
1827 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1828 }
1829
1830 return 0;
1831}
1832
1833/**
1834 * do_lookup_nm- look up a "hashed" node.
1835 * @c: UBIFS file-system description object
1836 * @key: node key to lookup
1837 * @node: the node is returned here
1838 * @nm: node name
1839 *
1840 * This function looks up and reads a node which contains name hash in the key.
1841 * Since the hash may have collisions, there may be many nodes with the same
1842 * key, so we have to sequentially look to all of them until the needed one is
1843 * found. This function returns zero in case of success, %-ENOENT if the node
1844 * was not found, and a negative error code in case of failure.
1845 */
1846static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1847 void *node, const struct fscrypt_name *nm)
1848{
1849 int found, n, err;
1850 struct ubifs_znode *znode;
1851
1852 dbg_tnck(key, "key ");
1853 mutex_lock(&c->tnc_mutex);
1854 found = ubifs_lookup_level0(c, key, &znode, &n);
1855 if (!found) {
1856 err = -ENOENT;
1857 goto out_unlock;
1858 } else if (found < 0) {
1859 err = found;
1860 goto out_unlock;
1861 }
1862
1863 ubifs_assert(c, n >= 0);
1864
1865 err = resolve_collision(c, key, &znode, &n, nm);
1866 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1867 if (unlikely(err < 0))
1868 goto out_unlock;
1869 if (err == 0) {
1870 err = -ENOENT;
1871 goto out_unlock;
1872 }
1873
1874 err = tnc_read_hashed_node(c, &znode->zbranch[n], node);
1875
1876out_unlock:
1877 mutex_unlock(&c->tnc_mutex);
1878 return err;
1879}
1880
1881/**
1882 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1883 * @c: UBIFS file-system description object
1884 * @key: node key to lookup
1885 * @node: the node is returned here
1886 * @nm: node name
1887 *
1888 * This function looks up and reads a node which contains name hash in the key.
1889 * Since the hash may have collisions, there may be many nodes with the same
1890 * key, so we have to sequentially look to all of them until the needed one is
1891 * found. This function returns zero in case of success, %-ENOENT if the node
1892 * was not found, and a negative error code in case of failure.
1893 */
1894int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1895 void *node, const struct fscrypt_name *nm)
1896{
1897 int err, len;
1898 const struct ubifs_dent_node *dent = node;
1899
1900 /*
1901 * We assume that in most of the cases there are no name collisions and
1902 * 'ubifs_tnc_lookup()' returns us the right direntry.
1903 */
1904 err = ubifs_tnc_lookup(c, key, node);
1905 if (err)
1906 return err;
1907
1908 len = le16_to_cpu(dent->nlen);
1909 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len))
1910 return 0;
1911
1912 /*
1913 * Unluckily, there are hash collisions and we have to iterate over
1914 * them look at each direntry with colliding name hash sequentially.
1915 */
1916
1917 return do_lookup_nm(c, key, node, nm);
1918}
1919
1920static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key,
1921 struct ubifs_dent_node *dent, uint32_t cookie,
1922 struct ubifs_znode **zn, int *n, int exact)
1923{
1924 int err;
1925 struct ubifs_znode *znode = *zn;
1926 struct ubifs_zbranch *zbr;
1927 union ubifs_key *dkey;
1928
1929 if (!exact) {
1930 err = tnc_next(c, &znode, n);
1931 if (err)
1932 return err;
1933 }
1934
1935 for (;;) {
1936 zbr = &znode->zbranch[*n];
1937 dkey = &zbr->key;
1938
1939 if (key_inum(c, dkey) != key_inum(c, key) ||
1940 key_type(c, dkey) != key_type(c, key)) {
1941 return -ENOENT;
1942 }
1943
1944 err = tnc_read_hashed_node(c, zbr, dent);
1945 if (err)
1946 return err;
1947
1948 if (key_hash(c, key) == key_hash(c, dkey) &&
1949 le32_to_cpu(dent->cookie) == cookie) {
1950 *zn = znode;
1951 return 0;
1952 }
1953
1954 err = tnc_next(c, &znode, n);
1955 if (err)
1956 return err;
1957 }
1958}
1959
1960static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1961 struct ubifs_dent_node *dent, uint32_t cookie)
1962{
1963 int n, err;
1964 struct ubifs_znode *znode;
1965 union ubifs_key start_key;
1966
1967 ubifs_assert(c, is_hash_key(c, key));
1968
1969 lowest_dent_key(c, &start_key, key_inum(c, key));
1970
1971 mutex_lock(&c->tnc_mutex);
1972 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
1973 if (unlikely(err < 0))
1974 goto out_unlock;
1975
1976 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
1977
1978out_unlock:
1979 mutex_unlock(&c->tnc_mutex);
1980 return err;
1981}
1982
1983/**
1984 * ubifs_tnc_lookup_dh - look up a "double hashed" node.
1985 * @c: UBIFS file-system description object
1986 * @key: node key to lookup
1987 * @node: the node is returned here
1988 * @cookie: node cookie for collision resolution
1989 *
1990 * This function looks up and reads a node which contains name hash in the key.
1991 * Since the hash may have collisions, there may be many nodes with the same
1992 * key, so we have to sequentially look to all of them until the needed one
1993 * with the same cookie value is found.
1994 * This function returns zero in case of success, %-ENOENT if the node
1995 * was not found, and a negative error code in case of failure.
1996 */
1997int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key,
1998 void *node, uint32_t cookie)
1999{
2000 int err;
2001 const struct ubifs_dent_node *dent = node;
2002
2003 if (!c->double_hash)
2004 return -EOPNOTSUPP;
2005
2006 /*
2007 * We assume that in most of the cases there are no name collisions and
2008 * 'ubifs_tnc_lookup()' returns us the right direntry.
2009 */
2010 err = ubifs_tnc_lookup(c, key, node);
2011 if (err)
2012 return err;
2013
2014 if (le32_to_cpu(dent->cookie) == cookie)
2015 return 0;
2016
2017 /*
2018 * Unluckily, there are hash collisions and we have to iterate over
2019 * them look at each direntry with colliding name hash sequentially.
2020 */
2021 return do_lookup_dh(c, key, node, cookie);
2022}
2023
2024/**
2025 * correct_parent_keys - correct parent znodes' keys.
2026 * @c: UBIFS file-system description object
2027 * @znode: znode to correct parent znodes for
2028 *
2029 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
2030 * zbranch changes, keys of parent znodes have to be corrected. This helper
2031 * function is called in such situations and corrects the keys if needed.
2032 */
2033static void correct_parent_keys(const struct ubifs_info *c,
2034 struct ubifs_znode *znode)
2035{
2036 union ubifs_key *key, *key1;
2037
2038 ubifs_assert(c, znode->parent);
2039 ubifs_assert(c, znode->iip == 0);
2040
2041 key = &znode->zbranch[0].key;
2042 key1 = &znode->parent->zbranch[0].key;
2043
2044 while (keys_cmp(c, key, key1) < 0) {
2045 key_copy(c, key, key1);
2046 znode = znode->parent;
2047 znode->alt = 1;
2048 if (!znode->parent || znode->iip)
2049 break;
2050 key1 = &znode->parent->zbranch[0].key;
2051 }
2052}
2053
2054/**
2055 * insert_zbranch - insert a zbranch into a znode.
2056 * @c: UBIFS file-system description object
2057 * @znode: znode into which to insert
2058 * @zbr: zbranch to insert
2059 * @n: slot number to insert to
2060 *
2061 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
2062 * znode's array of zbranches and keeps zbranches consolidated, so when a new
2063 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
2064 * slot, zbranches starting from @n have to be moved right.
2065 */
2066static void insert_zbranch(struct ubifs_info *c, struct ubifs_znode *znode,
2067 const struct ubifs_zbranch *zbr, int n)
2068{
2069 int i;
2070
2071 ubifs_assert(c, ubifs_zn_dirty(znode));
2072
2073 if (znode->level) {
2074 for (i = znode->child_cnt; i > n; i--) {
2075 znode->zbranch[i] = znode->zbranch[i - 1];
2076 if (znode->zbranch[i].znode)
2077 znode->zbranch[i].znode->iip = i;
2078 }
2079 if (zbr->znode)
2080 zbr->znode->iip = n;
2081 } else
2082 for (i = znode->child_cnt; i > n; i--)
2083 znode->zbranch[i] = znode->zbranch[i - 1];
2084
2085 znode->zbranch[n] = *zbr;
2086 znode->child_cnt += 1;
2087
2088 /*
2089 * After inserting at slot zero, the lower bound of the key range of
2090 * this znode may have changed. If this znode is subsequently split
2091 * then the upper bound of the key range may change, and furthermore
2092 * it could change to be lower than the original lower bound. If that
2093 * happens, then it will no longer be possible to find this znode in the
2094 * TNC using the key from the index node on flash. That is bad because
2095 * if it is not found, we will assume it is obsolete and may overwrite
2096 * it. Then if there is an unclean unmount, we will start using the
2097 * old index which will be broken.
2098 *
2099 * So we first mark znodes that have insertions at slot zero, and then
2100 * if they are split we add their lnum/offs to the old_idx tree.
2101 */
2102 if (n == 0)
2103 znode->alt = 1;
2104}
2105
2106/**
2107 * tnc_insert - insert a node into TNC.
2108 * @c: UBIFS file-system description object
2109 * @znode: znode to insert into
2110 * @zbr: branch to insert
2111 * @n: slot number to insert new zbranch to
2112 *
2113 * This function inserts a new node described by @zbr into znode @znode. If
2114 * znode does not have a free slot for new zbranch, it is split. Parent znodes
2115 * are splat as well if needed. Returns zero in case of success or a negative
2116 * error code in case of failure.
2117 */
2118static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
2119 struct ubifs_zbranch *zbr, int n)
2120{
2121 struct ubifs_znode *zn, *zi, *zp;
2122 int i, keep, move, appending = 0;
2123 union ubifs_key *key = &zbr->key, *key1;
2124
2125 ubifs_assert(c, n >= 0 && n <= c->fanout);
2126
2127 /* Implement naive insert for now */
2128again:
2129 zp = znode->parent;
2130 if (znode->child_cnt < c->fanout) {
2131 ubifs_assert(c, n != c->fanout);
2132 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level);
2133
2134 insert_zbranch(c, znode, zbr, n);
2135
2136 /* Ensure parent's key is correct */
2137 if (n == 0 && zp && znode->iip == 0)
2138 correct_parent_keys(c, znode);
2139
2140 return 0;
2141 }
2142
2143 /*
2144 * Unfortunately, @znode does not have more empty slots and we have to
2145 * split it.
2146 */
2147 dbg_tnck(key, "splitting level %d, key ", znode->level);
2148
2149 if (znode->alt)
2150 /*
2151 * We can no longer be sure of finding this znode by key, so we
2152 * record it in the old_idx tree.
2153 */
2154 ins_clr_old_idx_znode(c, znode);
2155
2156 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2157 if (!zn)
2158 return -ENOMEM;
2159 zn->parent = zp;
2160 zn->level = znode->level;
2161
2162 /* Decide where to split */
2163 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2164 /* Try not to split consecutive data keys */
2165 if (n == c->fanout) {
2166 key1 = &znode->zbranch[n - 1].key;
2167 if (key_inum(c, key1) == key_inum(c, key) &&
2168 key_type(c, key1) == UBIFS_DATA_KEY)
2169 appending = 1;
2170 } else
2171 goto check_split;
2172 } else if (appending && n != c->fanout) {
2173 /* Try not to split consecutive data keys */
2174 appending = 0;
2175check_split:
2176 if (n >= (c->fanout + 1) / 2) {
2177 key1 = &znode->zbranch[0].key;
2178 if (key_inum(c, key1) == key_inum(c, key) &&
2179 key_type(c, key1) == UBIFS_DATA_KEY) {
2180 key1 = &znode->zbranch[n].key;
2181 if (key_inum(c, key1) != key_inum(c, key) ||
2182 key_type(c, key1) != UBIFS_DATA_KEY) {
2183 keep = n;
2184 move = c->fanout - keep;
2185 zi = znode;
2186 goto do_split;
2187 }
2188 }
2189 }
2190 }
2191
2192 if (appending) {
2193 keep = c->fanout;
2194 move = 0;
2195 } else {
2196 keep = (c->fanout + 1) / 2;
2197 move = c->fanout - keep;
2198 }
2199
2200 /*
2201 * Although we don't at present, we could look at the neighbors and see
2202 * if we can move some zbranches there.
2203 */
2204
2205 if (n < keep) {
2206 /* Insert into existing znode */
2207 zi = znode;
2208 move += 1;
2209 keep -= 1;
2210 } else {
2211 /* Insert into new znode */
2212 zi = zn;
2213 n -= keep;
2214 /* Re-parent */
2215 if (zn->level != 0)
2216 zbr->znode->parent = zn;
2217 }
2218
2219do_split:
2220
2221 __set_bit(DIRTY_ZNODE, &zn->flags);
2222 atomic_long_inc(&c->dirty_zn_cnt);
2223
2224 zn->child_cnt = move;
2225 znode->child_cnt = keep;
2226
2227 dbg_tnc("moving %d, keeping %d", move, keep);
2228
2229 /* Move zbranch */
2230 for (i = 0; i < move; i++) {
2231 zn->zbranch[i] = znode->zbranch[keep + i];
2232 /* Re-parent */
2233 if (zn->level != 0)
2234 if (zn->zbranch[i].znode) {
2235 zn->zbranch[i].znode->parent = zn;
2236 zn->zbranch[i].znode->iip = i;
2237 }
2238 }
2239
2240 /* Insert new key and branch */
2241 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level);
2242
2243 insert_zbranch(c, zi, zbr, n);
2244
2245 /* Insert new znode (produced by spitting) into the parent */
2246 if (zp) {
2247 if (n == 0 && zi == znode && znode->iip == 0)
2248 correct_parent_keys(c, znode);
2249
2250 /* Locate insertion point */
2251 n = znode->iip + 1;
2252
2253 /* Tail recursion */
2254 zbr->key = zn->zbranch[0].key;
2255 zbr->znode = zn;
2256 zbr->lnum = 0;
2257 zbr->offs = 0;
2258 zbr->len = 0;
2259 znode = zp;
2260
2261 goto again;
2262 }
2263
2264 /* We have to split root znode */
2265 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2266
2267 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2268 if (!zi)
2269 return -ENOMEM;
2270
2271 zi->child_cnt = 2;
2272 zi->level = znode->level + 1;
2273
2274 __set_bit(DIRTY_ZNODE, &zi->flags);
2275 atomic_long_inc(&c->dirty_zn_cnt);
2276
2277 zi->zbranch[0].key = znode->zbranch[0].key;
2278 zi->zbranch[0].znode = znode;
2279 zi->zbranch[0].lnum = c->zroot.lnum;
2280 zi->zbranch[0].offs = c->zroot.offs;
2281 zi->zbranch[0].len = c->zroot.len;
2282 zi->zbranch[1].key = zn->zbranch[0].key;
2283 zi->zbranch[1].znode = zn;
2284
2285 c->zroot.lnum = 0;
2286 c->zroot.offs = 0;
2287 c->zroot.len = 0;
2288 c->zroot.znode = zi;
2289
2290 zn->parent = zi;
2291 zn->iip = 1;
2292 znode->parent = zi;
2293 znode->iip = 0;
2294
2295 return 0;
2296}
2297
2298/**
2299 * ubifs_tnc_add - add a node to TNC.
2300 * @c: UBIFS file-system description object
2301 * @key: key to add
2302 * @lnum: LEB number of node
2303 * @offs: node offset
2304 * @len: node length
2305 * @hash: The hash over the node
2306 *
2307 * This function adds a node with key @key to TNC. The node may be new or it may
2308 * obsolete some existing one. Returns %0 on success or negative error code on
2309 * failure.
2310 */
2311int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2312 int offs, int len, const u8 *hash)
2313{
2314 int found, n, err = 0;
2315 struct ubifs_znode *znode;
2316
2317 mutex_lock(&c->tnc_mutex);
2318 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len);
2319 found = lookup_level0_dirty(c, key, &znode, &n);
2320 if (!found) {
2321 struct ubifs_zbranch zbr;
2322
2323 zbr.znode = NULL;
2324 zbr.lnum = lnum;
2325 zbr.offs = offs;
2326 zbr.len = len;
2327 ubifs_copy_hash(c, hash, zbr.hash);
2328 key_copy(c, key, &zbr.key);
2329 err = tnc_insert(c, znode, &zbr, n + 1);
2330 } else if (found == 1) {
2331 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2332
2333 lnc_free(zbr);
2334 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2335 zbr->lnum = lnum;
2336 zbr->offs = offs;
2337 zbr->len = len;
2338 ubifs_copy_hash(c, hash, zbr->hash);
2339 } else
2340 err = found;
2341 if (!err)
2342 err = dbg_check_tnc(c, 0);
2343 mutex_unlock(&c->tnc_mutex);
2344
2345 return err;
2346}
2347
2348/**
2349 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2350 * @c: UBIFS file-system description object
2351 * @key: key to add
2352 * @old_lnum: LEB number of old node
2353 * @old_offs: old node offset
2354 * @lnum: LEB number of node
2355 * @offs: node offset
2356 * @len: node length
2357 *
2358 * This function replaces a node with key @key in the TNC only if the old node
2359 * is found. This function is called by garbage collection when node are moved.
2360 * Returns %0 on success or negative error code on failure.
2361 */
2362int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2363 int old_lnum, int old_offs, int lnum, int offs, int len)
2364{
2365 int found, n, err = 0;
2366 struct ubifs_znode *znode;
2367
2368 mutex_lock(&c->tnc_mutex);
2369 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum,
2370 old_offs, lnum, offs, len);
2371 found = lookup_level0_dirty(c, key, &znode, &n);
2372 if (found < 0) {
2373 err = found;
2374 goto out_unlock;
2375 }
2376
2377 if (found == 1) {
2378 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2379
2380 found = 0;
2381 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2382 lnc_free(zbr);
2383 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2384 if (err)
2385 goto out_unlock;
2386 zbr->lnum = lnum;
2387 zbr->offs = offs;
2388 zbr->len = len;
2389 found = 1;
2390 } else if (is_hash_key(c, key)) {
2391 found = resolve_collision_directly(c, key, &znode, &n,
2392 old_lnum, old_offs);
2393 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2394 found, znode, n, old_lnum, old_offs);
2395 if (found < 0) {
2396 err = found;
2397 goto out_unlock;
2398 }
2399
2400 if (found) {
2401 /* Ensure the znode is dirtied */
2402 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2403 znode = dirty_cow_bottom_up(c, znode);
2404 if (IS_ERR(znode)) {
2405 err = PTR_ERR(znode);
2406 goto out_unlock;
2407 }
2408 }
2409 zbr = &znode->zbranch[n];
2410 lnc_free(zbr);
2411 err = ubifs_add_dirt(c, zbr->lnum,
2412 zbr->len);
2413 if (err)
2414 goto out_unlock;
2415 zbr->lnum = lnum;
2416 zbr->offs = offs;
2417 zbr->len = len;
2418 }
2419 }
2420 }
2421
2422 if (!found)
2423 err = ubifs_add_dirt(c, lnum, len);
2424
2425 if (!err)
2426 err = dbg_check_tnc(c, 0);
2427
2428out_unlock:
2429 mutex_unlock(&c->tnc_mutex);
2430 return err;
2431}
2432
2433/**
2434 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2435 * @c: UBIFS file-system description object
2436 * @key: key to add
2437 * @lnum: LEB number of node
2438 * @offs: node offset
2439 * @len: node length
2440 * @hash: The hash over the node
2441 * @nm: node name
2442 *
2443 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2444 * may have collisions, like directory entry keys.
2445 */
2446int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2447 int lnum, int offs, int len, const u8 *hash,
2448 const struct fscrypt_name *nm)
2449{
2450 int found, n, err = 0;
2451 struct ubifs_znode *znode;
2452
2453 mutex_lock(&c->tnc_mutex);
2454 dbg_tnck(key, "LEB %d:%d, key ", lnum, offs);
2455 found = lookup_level0_dirty(c, key, &znode, &n);
2456 if (found < 0) {
2457 err = found;
2458 goto out_unlock;
2459 }
2460
2461 if (found == 1) {
2462 if (c->replaying)
2463 found = fallible_resolve_collision(c, key, &znode, &n,
2464 nm, 1);
2465 else
2466 found = resolve_collision(c, key, &znode, &n, nm);
2467 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2468 if (found < 0) {
2469 err = found;
2470 goto out_unlock;
2471 }
2472
2473 /* Ensure the znode is dirtied */
2474 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2475 znode = dirty_cow_bottom_up(c, znode);
2476 if (IS_ERR(znode)) {
2477 err = PTR_ERR(znode);
2478 goto out_unlock;
2479 }
2480 }
2481
2482 if (found == 1) {
2483 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2484
2485 lnc_free(zbr);
2486 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2487 zbr->lnum = lnum;
2488 zbr->offs = offs;
2489 zbr->len = len;
2490 ubifs_copy_hash(c, hash, zbr->hash);
2491 goto out_unlock;
2492 }
2493 }
2494
2495 if (!found) {
2496 struct ubifs_zbranch zbr;
2497
2498 zbr.znode = NULL;
2499 zbr.lnum = lnum;
2500 zbr.offs = offs;
2501 zbr.len = len;
2502 ubifs_copy_hash(c, hash, zbr.hash);
2503 key_copy(c, key, &zbr.key);
2504 err = tnc_insert(c, znode, &zbr, n + 1);
2505 if (err)
2506 goto out_unlock;
2507 if (c->replaying) {
2508 /*
2509 * We did not find it in the index so there may be a
2510 * dangling branch still in the index. So we remove it
2511 * by passing 'ubifs_tnc_remove_nm()' the same key but
2512 * an unmatchable name.
2513 */
2514 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } };
2515
2516 err = dbg_check_tnc(c, 0);
2517 mutex_unlock(&c->tnc_mutex);
2518 if (err)
2519 return err;
2520 return ubifs_tnc_remove_nm(c, key, &noname);
2521 }
2522 }
2523
2524out_unlock:
2525 if (!err)
2526 err = dbg_check_tnc(c, 0);
2527 mutex_unlock(&c->tnc_mutex);
2528 return err;
2529}
2530
2531/**
2532 * tnc_delete - delete a znode form TNC.
2533 * @c: UBIFS file-system description object
2534 * @znode: znode to delete from
2535 * @n: zbranch slot number to delete
2536 *
2537 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2538 * case of success and a negative error code in case of failure.
2539 */
2540static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2541{
2542 struct ubifs_zbranch *zbr;
2543 struct ubifs_znode *zp;
2544 int i, err;
2545
2546 /* Delete without merge for now */
2547 ubifs_assert(c, znode->level == 0);
2548 ubifs_assert(c, n >= 0 && n < c->fanout);
2549 dbg_tnck(&znode->zbranch[n].key, "deleting key ");
2550
2551 zbr = &znode->zbranch[n];
2552 lnc_free(zbr);
2553
2554 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2555 if (err) {
2556 ubifs_dump_znode(c, znode);
2557 return err;
2558 }
2559
2560 /* We do not "gap" zbranch slots */
2561 for (i = n; i < znode->child_cnt - 1; i++)
2562 znode->zbranch[i] = znode->zbranch[i + 1];
2563 znode->child_cnt -= 1;
2564
2565 if (znode->child_cnt > 0)
2566 return 0;
2567
2568 /*
2569 * This was the last zbranch, we have to delete this znode from the
2570 * parent.
2571 */
2572
2573 do {
2574 ubifs_assert(c, !ubifs_zn_obsolete(znode));
2575 ubifs_assert(c, ubifs_zn_dirty(znode));
2576
2577 zp = znode->parent;
2578 n = znode->iip;
2579
2580 atomic_long_dec(&c->dirty_zn_cnt);
2581
2582 err = insert_old_idx_znode(c, znode);
2583 if (err)
2584 return err;
2585
2586 if (znode->cnext) {
2587 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2588 atomic_long_inc(&c->clean_zn_cnt);
2589 atomic_long_inc(&ubifs_clean_zn_cnt);
2590 } else
2591 kfree(znode);
2592 znode = zp;
2593 } while (znode->child_cnt == 1); /* while removing last child */
2594
2595 /* Remove from znode, entry n - 1 */
2596 znode->child_cnt -= 1;
2597 ubifs_assert(c, znode->level != 0);
2598 for (i = n; i < znode->child_cnt; i++) {
2599 znode->zbranch[i] = znode->zbranch[i + 1];
2600 if (znode->zbranch[i].znode)
2601 znode->zbranch[i].znode->iip = i;
2602 }
2603
2604 /*
2605 * If this is the root and it has only 1 child then
2606 * collapse the tree.
2607 */
2608 if (!znode->parent) {
2609 while (znode->child_cnt == 1 && znode->level != 0) {
2610 zp = znode;
2611 zbr = &znode->zbranch[0];
2612 znode = get_znode(c, znode, 0);
2613 if (IS_ERR(znode))
2614 return PTR_ERR(znode);
2615 znode = dirty_cow_znode(c, zbr);
2616 if (IS_ERR(znode))
2617 return PTR_ERR(znode);
2618 znode->parent = NULL;
2619 znode->iip = 0;
2620 if (c->zroot.len) {
2621 err = insert_old_idx(c, c->zroot.lnum,
2622 c->zroot.offs);
2623 if (err)
2624 return err;
2625 }
2626 c->zroot.lnum = zbr->lnum;
2627 c->zroot.offs = zbr->offs;
2628 c->zroot.len = zbr->len;
2629 c->zroot.znode = znode;
2630 ubifs_assert(c, !ubifs_zn_obsolete(zp));
2631 ubifs_assert(c, ubifs_zn_dirty(zp));
2632 atomic_long_dec(&c->dirty_zn_cnt);
2633
2634 if (zp->cnext) {
2635 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2636 atomic_long_inc(&c->clean_zn_cnt);
2637 atomic_long_inc(&ubifs_clean_zn_cnt);
2638 } else
2639 kfree(zp);
2640 }
2641 }
2642
2643 return 0;
2644}
2645
2646/**
2647 * ubifs_tnc_remove - remove an index entry of a node.
2648 * @c: UBIFS file-system description object
2649 * @key: key of node
2650 *
2651 * Returns %0 on success or negative error code on failure.
2652 */
2653int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2654{
2655 int found, n, err = 0;
2656 struct ubifs_znode *znode;
2657
2658 mutex_lock(&c->tnc_mutex);
2659 dbg_tnck(key, "key ");
2660 found = lookup_level0_dirty(c, key, &znode, &n);
2661 if (found < 0) {
2662 err = found;
2663 goto out_unlock;
2664 }
2665 if (found == 1)
2666 err = tnc_delete(c, znode, n);
2667 if (!err)
2668 err = dbg_check_tnc(c, 0);
2669
2670out_unlock:
2671 mutex_unlock(&c->tnc_mutex);
2672 return err;
2673}
2674
2675/**
2676 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2677 * @c: UBIFS file-system description object
2678 * @key: key of node
2679 * @nm: directory entry name
2680 *
2681 * Returns %0 on success or negative error code on failure.
2682 */
2683int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2684 const struct fscrypt_name *nm)
2685{
2686 int n, err;
2687 struct ubifs_znode *znode;
2688
2689 mutex_lock(&c->tnc_mutex);
2690 dbg_tnck(key, "key ");
2691 err = lookup_level0_dirty(c, key, &znode, &n);
2692 if (err < 0)
2693 goto out_unlock;
2694
2695 if (err) {
2696 if (c->replaying)
2697 err = fallible_resolve_collision(c, key, &znode, &n,
2698 nm, 0);
2699 else
2700 err = resolve_collision(c, key, &znode, &n, nm);
2701 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2702 if (err < 0)
2703 goto out_unlock;
2704 if (err) {
2705 /* Ensure the znode is dirtied */
2706 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2707 znode = dirty_cow_bottom_up(c, znode);
2708 if (IS_ERR(znode)) {
2709 err = PTR_ERR(znode);
2710 goto out_unlock;
2711 }
2712 }
2713 err = tnc_delete(c, znode, n);
2714 }
2715 }
2716
2717out_unlock:
2718 if (!err)
2719 err = dbg_check_tnc(c, 0);
2720 mutex_unlock(&c->tnc_mutex);
2721 return err;
2722}
2723
2724/**
2725 * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node.
2726 * @c: UBIFS file-system description object
2727 * @key: key of node
2728 * @cookie: node cookie for collision resolution
2729 *
2730 * Returns %0 on success or negative error code on failure.
2731 */
2732int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key,
2733 uint32_t cookie)
2734{
2735 int n, err;
2736 struct ubifs_znode *znode;
2737 struct ubifs_dent_node *dent;
2738 struct ubifs_zbranch *zbr;
2739
2740 if (!c->double_hash)
2741 return -EOPNOTSUPP;
2742
2743 mutex_lock(&c->tnc_mutex);
2744 err = lookup_level0_dirty(c, key, &znode, &n);
2745 if (err <= 0)
2746 goto out_unlock;
2747
2748 zbr = &znode->zbranch[n];
2749 dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
2750 if (!dent) {
2751 err = -ENOMEM;
2752 goto out_unlock;
2753 }
2754
2755 err = tnc_read_hashed_node(c, zbr, dent);
2756 if (err)
2757 goto out_free;
2758
2759 /* If the cookie does not match, we're facing a hash collision. */
2760 if (le32_to_cpu(dent->cookie) != cookie) {
2761 union ubifs_key start_key;
2762
2763 lowest_dent_key(c, &start_key, key_inum(c, key));
2764
2765 err = ubifs_lookup_level0(c, &start_key, &znode, &n);
2766 if (unlikely(err < 0))
2767 goto out_free;
2768
2769 err = search_dh_cookie(c, key, dent, cookie, &znode, &n, err);
2770 if (err)
2771 goto out_free;
2772 }
2773
2774 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2775 znode = dirty_cow_bottom_up(c, znode);
2776 if (IS_ERR(znode)) {
2777 err = PTR_ERR(znode);
2778 goto out_free;
2779 }
2780 }
2781 err = tnc_delete(c, znode, n);
2782
2783out_free:
2784 kfree(dent);
2785out_unlock:
2786 if (!err)
2787 err = dbg_check_tnc(c, 0);
2788 mutex_unlock(&c->tnc_mutex);
2789 return err;
2790}
2791
2792/**
2793 * key_in_range - determine if a key falls within a range of keys.
2794 * @c: UBIFS file-system description object
2795 * @key: key to check
2796 * @from_key: lowest key in range
2797 * @to_key: highest key in range
2798 *
2799 * This function returns %1 if the key is in range and %0 otherwise.
2800 */
2801static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2802 union ubifs_key *from_key, union ubifs_key *to_key)
2803{
2804 if (keys_cmp(c, key, from_key) < 0)
2805 return 0;
2806 if (keys_cmp(c, key, to_key) > 0)
2807 return 0;
2808 return 1;
2809}
2810
2811/**
2812 * ubifs_tnc_remove_range - remove index entries in range.
2813 * @c: UBIFS file-system description object
2814 * @from_key: lowest key to remove
2815 * @to_key: highest key to remove
2816 *
2817 * This function removes index entries starting at @from_key and ending at
2818 * @to_key. This function returns zero in case of success and a negative error
2819 * code in case of failure.
2820 */
2821int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2822 union ubifs_key *to_key)
2823{
2824 int i, n, k, err = 0;
2825 struct ubifs_znode *znode;
2826 union ubifs_key *key;
2827
2828 mutex_lock(&c->tnc_mutex);
2829 while (1) {
2830 /* Find first level 0 znode that contains keys to remove */
2831 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2832 if (err < 0)
2833 goto out_unlock;
2834
2835 if (err)
2836 key = from_key;
2837 else {
2838 err = tnc_next(c, &znode, &n);
2839 if (err == -ENOENT) {
2840 err = 0;
2841 goto out_unlock;
2842 }
2843 if (err < 0)
2844 goto out_unlock;
2845 key = &znode->zbranch[n].key;
2846 if (!key_in_range(c, key, from_key, to_key)) {
2847 err = 0;
2848 goto out_unlock;
2849 }
2850 }
2851
2852 /* Ensure the znode is dirtied */
2853 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2854 znode = dirty_cow_bottom_up(c, znode);
2855 if (IS_ERR(znode)) {
2856 err = PTR_ERR(znode);
2857 goto out_unlock;
2858 }
2859 }
2860
2861 /* Remove all keys in range except the first */
2862 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2863 key = &znode->zbranch[i].key;
2864 if (!key_in_range(c, key, from_key, to_key))
2865 break;
2866 lnc_free(&znode->zbranch[i]);
2867 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2868 znode->zbranch[i].len);
2869 if (err) {
2870 ubifs_dump_znode(c, znode);
2871 goto out_unlock;
2872 }
2873 dbg_tnck(key, "removing key ");
2874 }
2875 if (k) {
2876 for (i = n + 1 + k; i < znode->child_cnt; i++)
2877 znode->zbranch[i - k] = znode->zbranch[i];
2878 znode->child_cnt -= k;
2879 }
2880
2881 /* Now delete the first */
2882 err = tnc_delete(c, znode, n);
2883 if (err)
2884 goto out_unlock;
2885 }
2886
2887out_unlock:
2888 if (!err)
2889 err = dbg_check_tnc(c, 0);
2890 mutex_unlock(&c->tnc_mutex);
2891 return err;
2892}
2893
2894/**
2895 * ubifs_tnc_remove_ino - remove an inode from TNC.
2896 * @c: UBIFS file-system description object
2897 * @inum: inode number to remove
2898 *
2899 * This function remove inode @inum and all the extended attributes associated
2900 * with the anode from TNC and returns zero in case of success or a negative
2901 * error code in case of failure.
2902 */
2903int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2904{
2905 union ubifs_key key1, key2;
2906 struct ubifs_dent_node *xent, *pxent = NULL;
2907 struct fscrypt_name nm = {0};
2908
2909 dbg_tnc("ino %lu", (unsigned long)inum);
2910
2911 /*
2912 * Walk all extended attribute entries and remove them together with
2913 * corresponding extended attribute inodes.
2914 */
2915 lowest_xent_key(c, &key1, inum);
2916 while (1) {
2917 ino_t xattr_inum;
2918 int err;
2919
2920 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2921 if (IS_ERR(xent)) {
2922 err = PTR_ERR(xent);
2923 if (err == -ENOENT)
2924 break;
2925 kfree(pxent);
2926 return err;
2927 }
2928
2929 xattr_inum = le64_to_cpu(xent->inum);
2930 dbg_tnc("xent '%s', ino %lu", xent->name,
2931 (unsigned long)xattr_inum);
2932
2933 ubifs_evict_xattr_inode(c, xattr_inum);
2934
2935 fname_name(&nm) = xent->name;
2936 fname_len(&nm) = le16_to_cpu(xent->nlen);
2937 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2938 if (err) {
2939 kfree(pxent);
2940 kfree(xent);
2941 return err;
2942 }
2943
2944 lowest_ino_key(c, &key1, xattr_inum);
2945 highest_ino_key(c, &key2, xattr_inum);
2946 err = ubifs_tnc_remove_range(c, &key1, &key2);
2947 if (err) {
2948 kfree(pxent);
2949 kfree(xent);
2950 return err;
2951 }
2952
2953 kfree(pxent);
2954 pxent = xent;
2955 key_read(c, &xent->key, &key1);
2956 }
2957
2958 kfree(pxent);
2959 lowest_ino_key(c, &key1, inum);
2960 highest_ino_key(c, &key2, inum);
2961
2962 return ubifs_tnc_remove_range(c, &key1, &key2);
2963}
2964
2965/**
2966 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2967 * @c: UBIFS file-system description object
2968 * @key: key of last entry
2969 * @nm: name of last entry found or %NULL
2970 *
2971 * This function finds and reads the next directory or extended attribute entry
2972 * after the given key (@key) if there is one. @nm is used to resolve
2973 * collisions.
2974 *
2975 * If the name of the current entry is not known and only the key is known,
2976 * @nm->name has to be %NULL. In this case the semantics of this function is a
2977 * little bit different and it returns the entry corresponding to this key, not
2978 * the next one. If the key was not found, the closest "right" entry is
2979 * returned.
2980 *
2981 * If the fist entry has to be found, @key has to contain the lowest possible
2982 * key value for this inode and @name has to be %NULL.
2983 *
2984 * This function returns the found directory or extended attribute entry node
2985 * in case of success, %-ENOENT is returned if no entry was found, and a
2986 * negative error code is returned in case of failure.
2987 */
2988struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2989 union ubifs_key *key,
2990 const struct fscrypt_name *nm)
2991{
2992 int n, err, type = key_type(c, key);
2993 struct ubifs_znode *znode;
2994 struct ubifs_dent_node *dent;
2995 struct ubifs_zbranch *zbr;
2996 union ubifs_key *dkey;
2997
2998 dbg_tnck(key, "key ");
2999 ubifs_assert(c, is_hash_key(c, key));
3000
3001 mutex_lock(&c->tnc_mutex);
3002 err = ubifs_lookup_level0(c, key, &znode, &n);
3003 if (unlikely(err < 0))
3004 goto out_unlock;
3005
3006 if (fname_len(nm) > 0) {
3007 if (err) {
3008 /* Handle collisions */
3009 if (c->replaying)
3010 err = fallible_resolve_collision(c, key, &znode, &n,
3011 nm, 0);
3012 else
3013 err = resolve_collision(c, key, &znode, &n, nm);
3014 dbg_tnc("rc returned %d, znode %p, n %d",
3015 err, znode, n);
3016 if (unlikely(err < 0))
3017 goto out_unlock;
3018 }
3019
3020 /* Now find next entry */
3021 err = tnc_next(c, &znode, &n);
3022 if (unlikely(err))
3023 goto out_unlock;
3024 } else {
3025 /*
3026 * The full name of the entry was not given, in which case the
3027 * behavior of this function is a little different and it
3028 * returns current entry, not the next one.
3029 */
3030 if (!err) {
3031 /*
3032 * However, the given key does not exist in the TNC
3033 * tree and @znode/@n variables contain the closest
3034 * "preceding" element. Switch to the next one.
3035 */
3036 err = tnc_next(c, &znode, &n);
3037 if (err)
3038 goto out_unlock;
3039 }
3040 }
3041
3042 zbr = &znode->zbranch[n];
3043 dent = kmalloc(zbr->len, GFP_NOFS);
3044 if (unlikely(!dent)) {
3045 err = -ENOMEM;
3046 goto out_unlock;
3047 }
3048
3049 /*
3050 * The above 'tnc_next()' call could lead us to the next inode, check
3051 * this.
3052 */
3053 dkey = &zbr->key;
3054 if (key_inum(c, dkey) != key_inum(c, key) ||
3055 key_type(c, dkey) != type) {
3056 err = -ENOENT;
3057 goto out_free;
3058 }
3059
3060 err = tnc_read_hashed_node(c, zbr, dent);
3061 if (unlikely(err))
3062 goto out_free;
3063
3064 mutex_unlock(&c->tnc_mutex);
3065 return dent;
3066
3067out_free:
3068 kfree(dent);
3069out_unlock:
3070 mutex_unlock(&c->tnc_mutex);
3071 return ERR_PTR(err);
3072}
3073
3074/**
3075 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3076 * @c: UBIFS file-system description object
3077 *
3078 * Destroy left-over obsolete znodes from a failed commit.
3079 */
3080static void tnc_destroy_cnext(struct ubifs_info *c)
3081{
3082 struct ubifs_znode *cnext;
3083
3084 if (!c->cnext)
3085 return;
3086 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3087 cnext = c->cnext;
3088 do {
3089 struct ubifs_znode *znode = cnext;
3090
3091 cnext = cnext->cnext;
3092 if (ubifs_zn_obsolete(znode))
3093 kfree(znode);
3094 else if (!ubifs_zn_cow(znode)) {
3095 /*
3096 * Don't forget to update clean znode count after
3097 * committing failed, because ubifs will check this
3098 * count while closing tnc. Non-obsolete znode could
3099 * be re-dirtied during committing process, so dirty
3100 * flag is untrustable. The flag 'COW_ZNODE' is set
3101 * for each dirty znode before committing, and it is
3102 * cleared as long as the znode become clean, so we
3103 * can statistic clean znode count according to this
3104 * flag.
3105 */
3106 atomic_long_inc(&c->clean_zn_cnt);
3107 atomic_long_inc(&ubifs_clean_zn_cnt);
3108 }
3109 } while (cnext && cnext != c->cnext);
3110}
3111
3112/**
3113 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3114 * @c: UBIFS file-system description object
3115 */
3116void ubifs_tnc_close(struct ubifs_info *c)
3117{
3118 tnc_destroy_cnext(c);
3119 if (c->zroot.znode) {
3120 long n, freed;
3121
3122 n = atomic_long_read(&c->clean_zn_cnt);
3123 freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode);
3124 ubifs_assert(c, freed == n);
3125 atomic_long_sub(n, &ubifs_clean_zn_cnt);
3126 }
3127 kfree(c->gap_lebs);
3128 kfree(c->ilebs);
3129 destroy_old_idx(c);
3130}
3131
3132/**
3133 * left_znode - get the znode to the left.
3134 * @c: UBIFS file-system description object
3135 * @znode: znode
3136 *
3137 * This function returns a pointer to the znode to the left of @znode or NULL if
3138 * there is not one. A negative error code is returned on failure.
3139 */
3140static struct ubifs_znode *left_znode(struct ubifs_info *c,
3141 struct ubifs_znode *znode)
3142{
3143 int level = znode->level;
3144
3145 while (1) {
3146 int n = znode->iip - 1;
3147
3148 /* Go up until we can go left */
3149 znode = znode->parent;
3150 if (!znode)
3151 return NULL;
3152 if (n >= 0) {
3153 /* Now go down the rightmost branch to 'level' */
3154 znode = get_znode(c, znode, n);
3155 if (IS_ERR(znode))
3156 return znode;
3157 while (znode->level != level) {
3158 n = znode->child_cnt - 1;
3159 znode = get_znode(c, znode, n);
3160 if (IS_ERR(znode))
3161 return znode;
3162 }
3163 break;
3164 }
3165 }
3166 return znode;
3167}
3168
3169/**
3170 * right_znode - get the znode to the right.
3171 * @c: UBIFS file-system description object
3172 * @znode: znode
3173 *
3174 * This function returns a pointer to the znode to the right of @znode or NULL
3175 * if there is not one. A negative error code is returned on failure.
3176 */
3177static struct ubifs_znode *right_znode(struct ubifs_info *c,
3178 struct ubifs_znode *znode)
3179{
3180 int level = znode->level;
3181
3182 while (1) {
3183 int n = znode->iip + 1;
3184
3185 /* Go up until we can go right */
3186 znode = znode->parent;
3187 if (!znode)
3188 return NULL;
3189 if (n < znode->child_cnt) {
3190 /* Now go down the leftmost branch to 'level' */
3191 znode = get_znode(c, znode, n);
3192 if (IS_ERR(znode))
3193 return znode;
3194 while (znode->level != level) {
3195 znode = get_znode(c, znode, 0);
3196 if (IS_ERR(znode))
3197 return znode;
3198 }
3199 break;
3200 }
3201 }
3202 return znode;
3203}
3204
3205/**
3206 * lookup_znode - find a particular indexing node from TNC.
3207 * @c: UBIFS file-system description object
3208 * @key: index node key to lookup
3209 * @level: index node level
3210 * @lnum: index node LEB number
3211 * @offs: index node offset
3212 *
3213 * This function searches an indexing node by its first key @key and its
3214 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3215 * nodes it traverses to TNC. This function is called for indexing nodes which
3216 * were found on the media by scanning, for example when garbage-collecting or
3217 * when doing in-the-gaps commit. This means that the indexing node which is
3218 * looked for does not have to have exactly the same leftmost key @key, because
3219 * the leftmost key may have been changed, in which case TNC will contain a
3220 * dirty znode which still refers the same @lnum:@offs. This function is clever
3221 * enough to recognize such indexing nodes.
3222 *
3223 * Note, if a znode was deleted or changed too much, then this function will
3224 * not find it. For situations like this UBIFS has the old index RB-tree
3225 * (indexed by @lnum:@offs).
3226 *
3227 * This function returns a pointer to the znode found or %NULL if it is not
3228 * found. A negative error code is returned on failure.
3229 */
3230static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3231 union ubifs_key *key, int level,
3232 int lnum, int offs)
3233{
3234 struct ubifs_znode *znode, *zn;
3235 int n, nn;
3236
3237 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3238
3239 /*
3240 * The arguments have probably been read off flash, so don't assume
3241 * they are valid.
3242 */
3243 if (level < 0)
3244 return ERR_PTR(-EINVAL);
3245
3246 /* Get the root znode */
3247 znode = c->zroot.znode;
3248 if (!znode) {
3249 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3250 if (IS_ERR(znode))
3251 return znode;
3252 }
3253 /* Check if it is the one we are looking for */
3254 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3255 return znode;
3256 /* Descend to the parent level i.e. (level + 1) */
3257 if (level >= znode->level)
3258 return NULL;
3259 while (1) {
3260 ubifs_search_zbranch(c, znode, key, &n);
3261 if (n < 0) {
3262 /*
3263 * We reached a znode where the leftmost key is greater
3264 * than the key we are searching for. This is the same
3265 * situation as the one described in a huge comment at
3266 * the end of the 'ubifs_lookup_level0()' function. And
3267 * for exactly the same reasons we have to try to look
3268 * left before giving up.
3269 */
3270 znode = left_znode(c, znode);
3271 if (!znode)
3272 return NULL;
3273 if (IS_ERR(znode))
3274 return znode;
3275 ubifs_search_zbranch(c, znode, key, &n);
3276 ubifs_assert(c, n >= 0);
3277 }
3278 if (znode->level == level + 1)
3279 break;
3280 znode = get_znode(c, znode, n);
3281 if (IS_ERR(znode))
3282 return znode;
3283 }
3284 /* Check if the child is the one we are looking for */
3285 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3286 return get_znode(c, znode, n);
3287 /* If the key is unique, there is nowhere else to look */
3288 if (!is_hash_key(c, key))
3289 return NULL;
3290 /*
3291 * The key is not unique and so may be also in the znodes to either
3292 * side.
3293 */
3294 zn = znode;
3295 nn = n;
3296 /* Look left */
3297 while (1) {
3298 /* Move one branch to the left */
3299 if (n)
3300 n -= 1;
3301 else {
3302 znode = left_znode(c, znode);
3303 if (!znode)
3304 break;
3305 if (IS_ERR(znode))
3306 return znode;
3307 n = znode->child_cnt - 1;
3308 }
3309 /* Check it */
3310 if (znode->zbranch[n].lnum == lnum &&
3311 znode->zbranch[n].offs == offs)
3312 return get_znode(c, znode, n);
3313 /* Stop if the key is less than the one we are looking for */
3314 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3315 break;
3316 }
3317 /* Back to the middle */
3318 znode = zn;
3319 n = nn;
3320 /* Look right */
3321 while (1) {
3322 /* Move one branch to the right */
3323 if (++n >= znode->child_cnt) {
3324 znode = right_znode(c, znode);
3325 if (!znode)
3326 break;
3327 if (IS_ERR(znode))
3328 return znode;
3329 n = 0;
3330 }
3331 /* Check it */
3332 if (znode->zbranch[n].lnum == lnum &&
3333 znode->zbranch[n].offs == offs)
3334 return get_znode(c, znode, n);
3335 /* Stop if the key is greater than the one we are looking for */
3336 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3337 break;
3338 }
3339 return NULL;
3340}
3341
3342/**
3343 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3344 * @c: UBIFS file-system description object
3345 * @key: key of index node
3346 * @level: index node level
3347 * @lnum: LEB number of index node
3348 * @offs: offset of index node
3349 *
3350 * This function returns %0 if the index node is not referred to in the TNC, %1
3351 * if the index node is referred to in the TNC and the corresponding znode is
3352 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3353 * znode is clean, and a negative error code in case of failure.
3354 *
3355 * Note, the @key argument has to be the key of the first child. Also note,
3356 * this function relies on the fact that 0:0 is never a valid LEB number and
3357 * offset for a main-area node.
3358 */
3359int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3360 int lnum, int offs)
3361{
3362 struct ubifs_znode *znode;
3363
3364 znode = lookup_znode(c, key, level, lnum, offs);
3365 if (!znode)
3366 return 0;
3367 if (IS_ERR(znode))
3368 return PTR_ERR(znode);
3369
3370 return ubifs_zn_dirty(znode) ? 1 : 2;
3371}
3372
3373/**
3374 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3375 * @c: UBIFS file-system description object
3376 * @key: node key
3377 * @lnum: node LEB number
3378 * @offs: node offset
3379 *
3380 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3381 * not, and a negative error code in case of failure.
3382 *
3383 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3384 * and offset for a main-area node.
3385 */
3386static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3387 int lnum, int offs)
3388{
3389 struct ubifs_zbranch *zbr;
3390 struct ubifs_znode *znode, *zn;
3391 int n, found, err, nn;
3392 const int unique = !is_hash_key(c, key);
3393
3394 found = ubifs_lookup_level0(c, key, &znode, &n);
3395 if (found < 0)
3396 return found; /* Error code */
3397 if (!found)
3398 return 0;
3399 zbr = &znode->zbranch[n];
3400 if (lnum == zbr->lnum && offs == zbr->offs)
3401 return 1; /* Found it */
3402 if (unique)
3403 return 0;
3404 /*
3405 * Because the key is not unique, we have to look left
3406 * and right as well
3407 */
3408 zn = znode;
3409 nn = n;
3410 /* Look left */
3411 while (1) {
3412 err = tnc_prev(c, &znode, &n);
3413 if (err == -ENOENT)
3414 break;
3415 if (err)
3416 return err;
3417 if (keys_cmp(c, key, &znode->zbranch[n].key))
3418 break;
3419 zbr = &znode->zbranch[n];
3420 if (lnum == zbr->lnum && offs == zbr->offs)
3421 return 1; /* Found it */
3422 }
3423 /* Look right */
3424 znode = zn;
3425 n = nn;
3426 while (1) {
3427 err = tnc_next(c, &znode, &n);
3428 if (err) {
3429 if (err == -ENOENT)
3430 return 0;
3431 return err;
3432 }
3433 if (keys_cmp(c, key, &znode->zbranch[n].key))
3434 break;
3435 zbr = &znode->zbranch[n];
3436 if (lnum == zbr->lnum && offs == zbr->offs)
3437 return 1; /* Found it */
3438 }
3439 return 0;
3440}
3441
3442/**
3443 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3444 * @c: UBIFS file-system description object
3445 * @key: node key
3446 * @level: index node level (if it is an index node)
3447 * @lnum: node LEB number
3448 * @offs: node offset
3449 * @is_idx: non-zero if the node is an index node
3450 *
3451 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3452 * negative error code in case of failure. For index nodes, @key has to be the
3453 * key of the first child. An index node is considered to be in the TNC only if
3454 * the corresponding znode is clean or has not been loaded.
3455 */
3456int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3457 int lnum, int offs, int is_idx)
3458{
3459 int err;
3460
3461 mutex_lock(&c->tnc_mutex);
3462 if (is_idx) {
3463 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3464 if (err < 0)
3465 goto out_unlock;
3466 if (err == 1)
3467 /* The index node was found but it was dirty */
3468 err = 0;
3469 else if (err == 2)
3470 /* The index node was found and it was clean */
3471 err = 1;
3472 else
3473 BUG_ON(err != 0);
3474 } else
3475 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3476
3477out_unlock:
3478 mutex_unlock(&c->tnc_mutex);
3479 return err;
3480}
3481
3482/**
3483 * ubifs_dirty_idx_node - dirty an index node.
3484 * @c: UBIFS file-system description object
3485 * @key: index node key
3486 * @level: index node level
3487 * @lnum: index node LEB number
3488 * @offs: index node offset
3489 *
3490 * This function loads and dirties an index node so that it can be garbage
3491 * collected. The @key argument has to be the key of the first child. This
3492 * function relies on the fact that 0:0 is never a valid LEB number and offset
3493 * for a main-area node. Returns %0 on success and a negative error code on
3494 * failure.
3495 */
3496int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3497 int lnum, int offs)
3498{
3499 struct ubifs_znode *znode;
3500 int err = 0;
3501
3502 mutex_lock(&c->tnc_mutex);
3503 znode = lookup_znode(c, key, level, lnum, offs);
3504 if (!znode)
3505 goto out_unlock;
3506 if (IS_ERR(znode)) {
3507 err = PTR_ERR(znode);
3508 goto out_unlock;
3509 }
3510 znode = dirty_cow_bottom_up(c, znode);
3511 if (IS_ERR(znode)) {
3512 err = PTR_ERR(znode);
3513 goto out_unlock;
3514 }
3515
3516out_unlock:
3517 mutex_unlock(&c->tnc_mutex);
3518 return err;
3519}
3520
3521/**
3522 * dbg_check_inode_size - check if inode size is correct.
3523 * @c: UBIFS file-system description object
3524 * @inode: inode to check
3525 * @size: inode size
3526 *
3527 * This function makes sure that the inode size (@size) is correct and it does
3528 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3529 * if it has a data page beyond @size, and other negative error code in case of
3530 * other errors.
3531 */
3532int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3533 loff_t size)
3534{
3535 int err, n;
3536 union ubifs_key from_key, to_key, *key;
3537 struct ubifs_znode *znode;
3538 unsigned int block;
3539
3540 if (!S_ISREG(inode->i_mode))
3541 return 0;
3542 if (!dbg_is_chk_gen(c))
3543 return 0;
3544
3545 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3546 data_key_init(c, &from_key, inode->i_ino, block);
3547 highest_data_key(c, &to_key, inode->i_ino);
3548
3549 mutex_lock(&c->tnc_mutex);
3550 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3551 if (err < 0)
3552 goto out_unlock;
3553
3554 if (err) {
3555 key = &from_key;
3556 goto out_dump;
3557 }
3558
3559 err = tnc_next(c, &znode, &n);
3560 if (err == -ENOENT) {
3561 err = 0;
3562 goto out_unlock;
3563 }
3564 if (err < 0)
3565 goto out_unlock;
3566
3567 ubifs_assert(c, err == 0);
3568 key = &znode->zbranch[n].key;
3569 if (!key_in_range(c, key, &from_key, &to_key))
3570 goto out_unlock;
3571
3572out_dump:
3573 block = key_block(c, key);
3574 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3575 (unsigned long)inode->i_ino, size,
3576 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3577 mutex_unlock(&c->tnc_mutex);
3578 ubifs_dump_inode(c, inode);
3579 dump_stack();
3580 return -EINVAL;
3581
3582out_unlock:
3583 mutex_unlock(&c->tnc_mutex);
3584 return err;
3585}