Loading...
1/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
21 */
22
23/*
24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of
25 * the UBIFS B-tree.
26 *
27 * At the moment the locking rules of the TNC tree are quite simple and
28 * straightforward. We just have a mutex and lock it when we traverse the
29 * tree. If a znode is not in memory, we read it from flash while still having
30 * the mutex locked.
31 */
32
33#include <linux/crc32.h>
34#include <linux/slab.h>
35#include "ubifs.h"
36
37/*
38 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions.
39 * @NAME_LESS: name corresponding to the first argument is less than second
40 * @NAME_MATCHES: names match
41 * @NAME_GREATER: name corresponding to the second argument is greater than
42 * first
43 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media
44 *
45 * These constants were introduce to improve readability.
46 */
47enum {
48 NAME_LESS = 0,
49 NAME_MATCHES = 1,
50 NAME_GREATER = 2,
51 NOT_ON_MEDIA = 3,
52};
53
54/**
55 * insert_old_idx - record an index node obsoleted since the last commit start.
56 * @c: UBIFS file-system description object
57 * @lnum: LEB number of obsoleted index node
58 * @offs: offset of obsoleted index node
59 *
60 * Returns %0 on success, and a negative error code on failure.
61 *
62 * For recovery, there must always be a complete intact version of the index on
63 * flash at all times. That is called the "old index". It is the index as at the
64 * time of the last successful commit. Many of the index nodes in the old index
65 * may be dirty, but they must not be erased until the next successful commit
66 * (at which point that index becomes the old index).
67 *
68 * That means that the garbage collection and the in-the-gaps method of
69 * committing must be able to determine if an index node is in the old index.
70 * Most of the old index nodes can be found by looking up the TNC using the
71 * 'lookup_znode()' function. However, some of the old index nodes may have
72 * been deleted from the current index or may have been changed so much that
73 * they cannot be easily found. In those cases, an entry is added to an RB-tree.
74 * That is what this function does. The RB-tree is ordered by LEB number and
75 * offset because they uniquely identify the old index node.
76 */
77static int insert_old_idx(struct ubifs_info *c, int lnum, int offs)
78{
79 struct ubifs_old_idx *old_idx, *o;
80 struct rb_node **p, *parent = NULL;
81
82 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS);
83 if (unlikely(!old_idx))
84 return -ENOMEM;
85 old_idx->lnum = lnum;
86 old_idx->offs = offs;
87
88 p = &c->old_idx.rb_node;
89 while (*p) {
90 parent = *p;
91 o = rb_entry(parent, struct ubifs_old_idx, rb);
92 if (lnum < o->lnum)
93 p = &(*p)->rb_left;
94 else if (lnum > o->lnum)
95 p = &(*p)->rb_right;
96 else if (offs < o->offs)
97 p = &(*p)->rb_left;
98 else if (offs > o->offs)
99 p = &(*p)->rb_right;
100 else {
101 ubifs_err("old idx added twice!");
102 kfree(old_idx);
103 return 0;
104 }
105 }
106 rb_link_node(&old_idx->rb, parent, p);
107 rb_insert_color(&old_idx->rb, &c->old_idx);
108 return 0;
109}
110
111/**
112 * insert_old_idx_znode - record a znode obsoleted since last commit start.
113 * @c: UBIFS file-system description object
114 * @znode: znode of obsoleted index node
115 *
116 * Returns %0 on success, and a negative error code on failure.
117 */
118int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode)
119{
120 if (znode->parent) {
121 struct ubifs_zbranch *zbr;
122
123 zbr = &znode->parent->zbranch[znode->iip];
124 if (zbr->len)
125 return insert_old_idx(c, zbr->lnum, zbr->offs);
126 } else
127 if (c->zroot.len)
128 return insert_old_idx(c, c->zroot.lnum,
129 c->zroot.offs);
130 return 0;
131}
132
133/**
134 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start.
135 * @c: UBIFS file-system description object
136 * @znode: znode of obsoleted index node
137 *
138 * Returns %0 on success, and a negative error code on failure.
139 */
140static int ins_clr_old_idx_znode(struct ubifs_info *c,
141 struct ubifs_znode *znode)
142{
143 int err;
144
145 if (znode->parent) {
146 struct ubifs_zbranch *zbr;
147
148 zbr = &znode->parent->zbranch[znode->iip];
149 if (zbr->len) {
150 err = insert_old_idx(c, zbr->lnum, zbr->offs);
151 if (err)
152 return err;
153 zbr->lnum = 0;
154 zbr->offs = 0;
155 zbr->len = 0;
156 }
157 } else
158 if (c->zroot.len) {
159 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs);
160 if (err)
161 return err;
162 c->zroot.lnum = 0;
163 c->zroot.offs = 0;
164 c->zroot.len = 0;
165 }
166 return 0;
167}
168
169/**
170 * destroy_old_idx - destroy the old_idx RB-tree.
171 * @c: UBIFS file-system description object
172 *
173 * During start commit, the old_idx RB-tree is used to avoid overwriting index
174 * nodes that were in the index last commit but have since been deleted. This
175 * is necessary for recovery i.e. the old index must be kept intact until the
176 * new index is successfully written. The old-idx RB-tree is used for the
177 * in-the-gaps method of writing index nodes and is destroyed every commit.
178 */
179void destroy_old_idx(struct ubifs_info *c)
180{
181 struct rb_node *this = c->old_idx.rb_node;
182 struct ubifs_old_idx *old_idx;
183
184 while (this) {
185 if (this->rb_left) {
186 this = this->rb_left;
187 continue;
188 } else if (this->rb_right) {
189 this = this->rb_right;
190 continue;
191 }
192 old_idx = rb_entry(this, struct ubifs_old_idx, rb);
193 this = rb_parent(this);
194 if (this) {
195 if (this->rb_left == &old_idx->rb)
196 this->rb_left = NULL;
197 else
198 this->rb_right = NULL;
199 }
200 kfree(old_idx);
201 }
202 c->old_idx = RB_ROOT;
203}
204
205/**
206 * copy_znode - copy a dirty znode.
207 * @c: UBIFS file-system description object
208 * @znode: znode to copy
209 *
210 * A dirty znode being committed may not be changed, so it is copied.
211 */
212static struct ubifs_znode *copy_znode(struct ubifs_info *c,
213 struct ubifs_znode *znode)
214{
215 struct ubifs_znode *zn;
216
217 zn = kmalloc(c->max_znode_sz, GFP_NOFS);
218 if (unlikely(!zn))
219 return ERR_PTR(-ENOMEM);
220
221 memcpy(zn, znode, c->max_znode_sz);
222 zn->cnext = NULL;
223 __set_bit(DIRTY_ZNODE, &zn->flags);
224 __clear_bit(COW_ZNODE, &zn->flags);
225
226 ubifs_assert(!ubifs_zn_obsolete(znode));
227 __set_bit(OBSOLETE_ZNODE, &znode->flags);
228
229 if (znode->level != 0) {
230 int i;
231 const int n = zn->child_cnt;
232
233 /* The children now have new parent */
234 for (i = 0; i < n; i++) {
235 struct ubifs_zbranch *zbr = &zn->zbranch[i];
236
237 if (zbr->znode)
238 zbr->znode->parent = zn;
239 }
240 }
241
242 atomic_long_inc(&c->dirty_zn_cnt);
243 return zn;
244}
245
246/**
247 * add_idx_dirt - add dirt due to a dirty znode.
248 * @c: UBIFS file-system description object
249 * @lnum: LEB number of index node
250 * @dirt: size of index node
251 *
252 * This function updates lprops dirty space and the new size of the index.
253 */
254static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt)
255{
256 c->calc_idx_sz -= ALIGN(dirt, 8);
257 return ubifs_add_dirt(c, lnum, dirt);
258}
259
260/**
261 * dirty_cow_znode - ensure a znode is not being committed.
262 * @c: UBIFS file-system description object
263 * @zbr: branch of znode to check
264 *
265 * Returns dirtied znode on success or negative error code on failure.
266 */
267static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c,
268 struct ubifs_zbranch *zbr)
269{
270 struct ubifs_znode *znode = zbr->znode;
271 struct ubifs_znode *zn;
272 int err;
273
274 if (!ubifs_zn_cow(znode)) {
275 /* znode is not being committed */
276 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) {
277 atomic_long_inc(&c->dirty_zn_cnt);
278 atomic_long_dec(&c->clean_zn_cnt);
279 atomic_long_dec(&ubifs_clean_zn_cnt);
280 err = add_idx_dirt(c, zbr->lnum, zbr->len);
281 if (unlikely(err))
282 return ERR_PTR(err);
283 }
284 return znode;
285 }
286
287 zn = copy_znode(c, znode);
288 if (IS_ERR(zn))
289 return zn;
290
291 if (zbr->len) {
292 err = insert_old_idx(c, zbr->lnum, zbr->offs);
293 if (unlikely(err))
294 return ERR_PTR(err);
295 err = add_idx_dirt(c, zbr->lnum, zbr->len);
296 } else
297 err = 0;
298
299 zbr->znode = zn;
300 zbr->lnum = 0;
301 zbr->offs = 0;
302 zbr->len = 0;
303
304 if (unlikely(err))
305 return ERR_PTR(err);
306 return zn;
307}
308
309/**
310 * lnc_add - add a leaf node to the leaf node cache.
311 * @c: UBIFS file-system description object
312 * @zbr: zbranch of leaf node
313 * @node: leaf node
314 *
315 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The
316 * purpose of the leaf node cache is to save re-reading the same leaf node over
317 * and over again. Most things are cached by VFS, however the file system must
318 * cache directory entries for readdir and for resolving hash collisions. The
319 * present implementation of the leaf node cache is extremely simple, and
320 * allows for error returns that are not used but that may be needed if a more
321 * complex implementation is created.
322 *
323 * Note, this function does not add the @node object to LNC directly, but
324 * allocates a copy of the object and adds the copy to LNC. The reason for this
325 * is that @node has been allocated outside of the TNC subsystem and will be
326 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC
327 * may be changed at any time, e.g. freed by the shrinker.
328 */
329static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr,
330 const void *node)
331{
332 int err;
333 void *lnc_node;
334 const struct ubifs_dent_node *dent = node;
335
336 ubifs_assert(!zbr->leaf);
337 ubifs_assert(zbr->len != 0);
338 ubifs_assert(is_hash_key(c, &zbr->key));
339
340 err = ubifs_validate_entry(c, dent);
341 if (err) {
342 dbg_dump_stack();
343 dbg_dump_node(c, dent);
344 return err;
345 }
346
347 lnc_node = kmalloc(zbr->len, GFP_NOFS);
348 if (!lnc_node)
349 /* We don't have to have the cache, so no error */
350 return 0;
351
352 memcpy(lnc_node, node, zbr->len);
353 zbr->leaf = lnc_node;
354 return 0;
355}
356
357 /**
358 * lnc_add_directly - add a leaf node to the leaf-node-cache.
359 * @c: UBIFS file-system description object
360 * @zbr: zbranch of leaf node
361 * @node: leaf node
362 *
363 * This function is similar to 'lnc_add()', but it does not create a copy of
364 * @node but inserts @node to TNC directly.
365 */
366static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr,
367 void *node)
368{
369 int err;
370
371 ubifs_assert(!zbr->leaf);
372 ubifs_assert(zbr->len != 0);
373
374 err = ubifs_validate_entry(c, node);
375 if (err) {
376 dbg_dump_stack();
377 dbg_dump_node(c, node);
378 return err;
379 }
380
381 zbr->leaf = node;
382 return 0;
383}
384
385/**
386 * lnc_free - remove a leaf node from the leaf node cache.
387 * @zbr: zbranch of leaf node
388 * @node: leaf node
389 */
390static void lnc_free(struct ubifs_zbranch *zbr)
391{
392 if (!zbr->leaf)
393 return;
394 kfree(zbr->leaf);
395 zbr->leaf = NULL;
396}
397
398/**
399 * tnc_read_node_nm - read a "hashed" leaf node.
400 * @c: UBIFS file-system description object
401 * @zbr: key and position of the node
402 * @node: node is returned here
403 *
404 * This function reads a "hashed" node defined by @zbr from the leaf node cache
405 * (in it is there) or from the hash media, in which case the node is also
406 * added to LNC. Returns zero in case of success or a negative negative error
407 * code in case of failure.
408 */
409static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr,
410 void *node)
411{
412 int err;
413
414 ubifs_assert(is_hash_key(c, &zbr->key));
415
416 if (zbr->leaf) {
417 /* Read from the leaf node cache */
418 ubifs_assert(zbr->len != 0);
419 memcpy(node, zbr->leaf, zbr->len);
420 return 0;
421 }
422
423 err = ubifs_tnc_read_node(c, zbr, node);
424 if (err)
425 return err;
426
427 /* Add the node to the leaf node cache */
428 err = lnc_add(c, zbr, node);
429 return err;
430}
431
432/**
433 * try_read_node - read a node if it is a node.
434 * @c: UBIFS file-system description object
435 * @buf: buffer to read to
436 * @type: node type
437 * @len: node length (not aligned)
438 * @lnum: LEB number of node to read
439 * @offs: offset of node to read
440 *
441 * This function tries to read a node of known type and length, checks it and
442 * stores it in @buf. This function returns %1 if a node is present and %0 if
443 * a node is not present. A negative error code is returned for I/O errors.
444 * This function performs that same function as ubifs_read_node except that
445 * it does not require that there is actually a node present and instead
446 * the return code indicates if a node was read.
447 *
448 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc
449 * is true (it is controlled by corresponding mount option). However, if
450 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to
451 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is
452 * because during mounting or re-mounting from R/O mode to R/W mode we may read
453 * journal nodes (when replying the journal or doing the recovery) and the
454 * journal nodes may potentially be corrupted, so checking is required.
455 */
456static int try_read_node(const struct ubifs_info *c, void *buf, int type,
457 int len, int lnum, int offs)
458{
459 int err, node_len;
460 struct ubifs_ch *ch = buf;
461 uint32_t crc, node_crc;
462
463 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
464
465 err = ubifs_leb_read(c, lnum, buf, offs, len, 1);
466 if (err) {
467 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
468 type, lnum, offs, err);
469 return err;
470 }
471
472 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
473 return 0;
474
475 if (ch->node_type != type)
476 return 0;
477
478 node_len = le32_to_cpu(ch->len);
479 if (node_len != len)
480 return 0;
481
482 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting &&
483 !c->remounting_rw)
484 return 1;
485
486 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
487 node_crc = le32_to_cpu(ch->crc);
488 if (crc != node_crc)
489 return 0;
490
491 return 1;
492}
493
494/**
495 * fallible_read_node - try to read a leaf node.
496 * @c: UBIFS file-system description object
497 * @key: key of node to read
498 * @zbr: position of node
499 * @node: node returned
500 *
501 * This function tries to read a node and returns %1 if the node is read, %0
502 * if the node is not present, and a negative error code in the case of error.
503 */
504static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
505 struct ubifs_zbranch *zbr, void *node)
506{
507 int ret;
508
509 dbg_tnc("LEB %d:%d, key %s", zbr->lnum, zbr->offs, DBGKEY(key));
510
511 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
512 zbr->offs);
513 if (ret == 1) {
514 union ubifs_key node_key;
515 struct ubifs_dent_node *dent = node;
516
517 /* All nodes have key in the same place */
518 key_read(c, &dent->key, &node_key);
519 if (keys_cmp(c, key, &node_key) != 0)
520 ret = 0;
521 }
522 if (ret == 0 && c->replaying)
523 dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
524 zbr->lnum, zbr->offs, zbr->len, DBGKEY(key));
525 return ret;
526}
527
528/**
529 * matches_name - determine if a direntry or xattr entry matches a given name.
530 * @c: UBIFS file-system description object
531 * @zbr: zbranch of dent
532 * @nm: name to match
533 *
534 * This function checks if xentry/direntry referred by zbranch @zbr matches name
535 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
536 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
537 * of failure, a negative error code is returned.
538 */
539static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
540 const struct qstr *nm)
541{
542 struct ubifs_dent_node *dent;
543 int nlen, err;
544
545 /* If possible, match against the dent in the leaf node cache */
546 if (!zbr->leaf) {
547 dent = kmalloc(zbr->len, GFP_NOFS);
548 if (!dent)
549 return -ENOMEM;
550
551 err = ubifs_tnc_read_node(c, zbr, dent);
552 if (err)
553 goto out_free;
554
555 /* Add the node to the leaf node cache */
556 err = lnc_add_directly(c, zbr, dent);
557 if (err)
558 goto out_free;
559 } else
560 dent = zbr->leaf;
561
562 nlen = le16_to_cpu(dent->nlen);
563 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
564 if (err == 0) {
565 if (nlen == nm->len)
566 return NAME_MATCHES;
567 else if (nlen < nm->len)
568 return NAME_LESS;
569 else
570 return NAME_GREATER;
571 } else if (err < 0)
572 return NAME_LESS;
573 else
574 return NAME_GREATER;
575
576out_free:
577 kfree(dent);
578 return err;
579}
580
581/**
582 * get_znode - get a TNC znode that may not be loaded yet.
583 * @c: UBIFS file-system description object
584 * @znode: parent znode
585 * @n: znode branch slot number
586 *
587 * This function returns the znode or a negative error code.
588 */
589static struct ubifs_znode *get_znode(struct ubifs_info *c,
590 struct ubifs_znode *znode, int n)
591{
592 struct ubifs_zbranch *zbr;
593
594 zbr = &znode->zbranch[n];
595 if (zbr->znode)
596 znode = zbr->znode;
597 else
598 znode = ubifs_load_znode(c, zbr, znode, n);
599 return znode;
600}
601
602/**
603 * tnc_next - find next TNC entry.
604 * @c: UBIFS file-system description object
605 * @zn: znode is passed and returned here
606 * @n: znode branch slot number is passed and returned here
607 *
608 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
609 * no next entry, or a negative error code otherwise.
610 */
611static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
612{
613 struct ubifs_znode *znode = *zn;
614 int nn = *n;
615
616 nn += 1;
617 if (nn < znode->child_cnt) {
618 *n = nn;
619 return 0;
620 }
621 while (1) {
622 struct ubifs_znode *zp;
623
624 zp = znode->parent;
625 if (!zp)
626 return -ENOENT;
627 nn = znode->iip + 1;
628 znode = zp;
629 if (nn < znode->child_cnt) {
630 znode = get_znode(c, znode, nn);
631 if (IS_ERR(znode))
632 return PTR_ERR(znode);
633 while (znode->level != 0) {
634 znode = get_znode(c, znode, 0);
635 if (IS_ERR(znode))
636 return PTR_ERR(znode);
637 }
638 nn = 0;
639 break;
640 }
641 }
642 *zn = znode;
643 *n = nn;
644 return 0;
645}
646
647/**
648 * tnc_prev - find previous TNC entry.
649 * @c: UBIFS file-system description object
650 * @zn: znode is returned here
651 * @n: znode branch slot number is passed and returned here
652 *
653 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
654 * there is no next entry, or a negative error code otherwise.
655 */
656static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
657{
658 struct ubifs_znode *znode = *zn;
659 int nn = *n;
660
661 if (nn > 0) {
662 *n = nn - 1;
663 return 0;
664 }
665 while (1) {
666 struct ubifs_znode *zp;
667
668 zp = znode->parent;
669 if (!zp)
670 return -ENOENT;
671 nn = znode->iip - 1;
672 znode = zp;
673 if (nn >= 0) {
674 znode = get_znode(c, znode, nn);
675 if (IS_ERR(znode))
676 return PTR_ERR(znode);
677 while (znode->level != 0) {
678 nn = znode->child_cnt - 1;
679 znode = get_znode(c, znode, nn);
680 if (IS_ERR(znode))
681 return PTR_ERR(znode);
682 }
683 nn = znode->child_cnt - 1;
684 break;
685 }
686 }
687 *zn = znode;
688 *n = nn;
689 return 0;
690}
691
692/**
693 * resolve_collision - resolve a collision.
694 * @c: UBIFS file-system description object
695 * @key: key of a directory or extended attribute entry
696 * @zn: znode is returned here
697 * @n: zbranch number is passed and returned here
698 * @nm: name of the entry
699 *
700 * This function is called for "hashed" keys to make sure that the found key
701 * really corresponds to the looked up node (directory or extended attribute
702 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
703 * %0 is returned if @nm is not found and @zn and @n are set to the previous
704 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
705 * This means that @n may be set to %-1 if the leftmost key in @zn is the
706 * previous one. A negative error code is returned on failures.
707 */
708static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
709 struct ubifs_znode **zn, int *n,
710 const struct qstr *nm)
711{
712 int err;
713
714 err = matches_name(c, &(*zn)->zbranch[*n], nm);
715 if (unlikely(err < 0))
716 return err;
717 if (err == NAME_MATCHES)
718 return 1;
719
720 if (err == NAME_GREATER) {
721 /* Look left */
722 while (1) {
723 err = tnc_prev(c, zn, n);
724 if (err == -ENOENT) {
725 ubifs_assert(*n == 0);
726 *n = -1;
727 return 0;
728 }
729 if (err < 0)
730 return err;
731 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
732 /*
733 * We have found the branch after which we would
734 * like to insert, but inserting in this znode
735 * may still be wrong. Consider the following 3
736 * znodes, in the case where we are resolving a
737 * collision with Key2.
738 *
739 * znode zp
740 * ----------------------
741 * level 1 | Key0 | Key1 |
742 * -----------------------
743 * | |
744 * znode za | | znode zb
745 * ------------ ------------
746 * level 0 | Key0 | | Key2 |
747 * ------------ ------------
748 *
749 * The lookup finds Key2 in znode zb. Lets say
750 * there is no match and the name is greater so
751 * we look left. When we find Key0, we end up
752 * here. If we return now, we will insert into
753 * znode za at slot n = 1. But that is invalid
754 * according to the parent's keys. Key2 must
755 * be inserted into znode zb.
756 *
757 * Note, this problem is not relevant for the
758 * case when we go right, because
759 * 'tnc_insert()' would correct the parent key.
760 */
761 if (*n == (*zn)->child_cnt - 1) {
762 err = tnc_next(c, zn, n);
763 if (err) {
764 /* Should be impossible */
765 ubifs_assert(0);
766 if (err == -ENOENT)
767 err = -EINVAL;
768 return err;
769 }
770 ubifs_assert(*n == 0);
771 *n = -1;
772 }
773 return 0;
774 }
775 err = matches_name(c, &(*zn)->zbranch[*n], nm);
776 if (err < 0)
777 return err;
778 if (err == NAME_LESS)
779 return 0;
780 if (err == NAME_MATCHES)
781 return 1;
782 ubifs_assert(err == NAME_GREATER);
783 }
784 } else {
785 int nn = *n;
786 struct ubifs_znode *znode = *zn;
787
788 /* Look right */
789 while (1) {
790 err = tnc_next(c, &znode, &nn);
791 if (err == -ENOENT)
792 return 0;
793 if (err < 0)
794 return err;
795 if (keys_cmp(c, &znode->zbranch[nn].key, key))
796 return 0;
797 err = matches_name(c, &znode->zbranch[nn], nm);
798 if (err < 0)
799 return err;
800 if (err == NAME_GREATER)
801 return 0;
802 *zn = znode;
803 *n = nn;
804 if (err == NAME_MATCHES)
805 return 1;
806 ubifs_assert(err == NAME_LESS);
807 }
808 }
809}
810
811/**
812 * fallible_matches_name - determine if a dent matches a given name.
813 * @c: UBIFS file-system description object
814 * @zbr: zbranch of dent
815 * @nm: name to match
816 *
817 * This is a "fallible" version of 'matches_name()' function which does not
818 * panic if the direntry/xentry referred by @zbr does not exist on the media.
819 *
820 * This function checks if xentry/direntry referred by zbranch @zbr matches name
821 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
822 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
823 * if xentry/direntry referred by @zbr does not exist on the media. A negative
824 * error code is returned in case of failure.
825 */
826static int fallible_matches_name(struct ubifs_info *c,
827 struct ubifs_zbranch *zbr,
828 const struct qstr *nm)
829{
830 struct ubifs_dent_node *dent;
831 int nlen, err;
832
833 /* If possible, match against the dent in the leaf node cache */
834 if (!zbr->leaf) {
835 dent = kmalloc(zbr->len, GFP_NOFS);
836 if (!dent)
837 return -ENOMEM;
838
839 err = fallible_read_node(c, &zbr->key, zbr, dent);
840 if (err < 0)
841 goto out_free;
842 if (err == 0) {
843 /* The node was not present */
844 err = NOT_ON_MEDIA;
845 goto out_free;
846 }
847 ubifs_assert(err == 1);
848
849 err = lnc_add_directly(c, zbr, dent);
850 if (err)
851 goto out_free;
852 } else
853 dent = zbr->leaf;
854
855 nlen = le16_to_cpu(dent->nlen);
856 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
857 if (err == 0) {
858 if (nlen == nm->len)
859 return NAME_MATCHES;
860 else if (nlen < nm->len)
861 return NAME_LESS;
862 else
863 return NAME_GREATER;
864 } else if (err < 0)
865 return NAME_LESS;
866 else
867 return NAME_GREATER;
868
869out_free:
870 kfree(dent);
871 return err;
872}
873
874/**
875 * fallible_resolve_collision - resolve a collision even if nodes are missing.
876 * @c: UBIFS file-system description object
877 * @key: key
878 * @zn: znode is returned here
879 * @n: branch number is passed and returned here
880 * @nm: name of directory entry
881 * @adding: indicates caller is adding a key to the TNC
882 *
883 * This is a "fallible" version of the 'resolve_collision()' function which
884 * does not panic if one of the nodes referred to by TNC does not exist on the
885 * media. This may happen when replaying the journal if a deleted node was
886 * Garbage-collected and the commit was not done. A branch that refers to a node
887 * that is not present is called a dangling branch. The following are the return
888 * codes for this function:
889 * o if @nm was found, %1 is returned and @zn and @n are set to the found
890 * branch;
891 * o if we are @adding and @nm was not found, %0 is returned;
892 * o if we are not @adding and @nm was not found, but a dangling branch was
893 * found, then %1 is returned and @zn and @n are set to the dangling branch;
894 * o a negative error code is returned in case of failure.
895 */
896static int fallible_resolve_collision(struct ubifs_info *c,
897 const union ubifs_key *key,
898 struct ubifs_znode **zn, int *n,
899 const struct qstr *nm, int adding)
900{
901 struct ubifs_znode *o_znode = NULL, *znode = *zn;
902 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
903
904 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
905 if (unlikely(cmp < 0))
906 return cmp;
907 if (cmp == NAME_MATCHES)
908 return 1;
909 if (cmp == NOT_ON_MEDIA) {
910 o_znode = znode;
911 o_n = nn;
912 /*
913 * We are unlucky and hit a dangling branch straight away.
914 * Now we do not really know where to go to find the needed
915 * branch - to the left or to the right. Well, let's try left.
916 */
917 unsure = 1;
918 } else if (!adding)
919 unsure = 1; /* Remove a dangling branch wherever it is */
920
921 if (cmp == NAME_GREATER || unsure) {
922 /* Look left */
923 while (1) {
924 err = tnc_prev(c, zn, n);
925 if (err == -ENOENT) {
926 ubifs_assert(*n == 0);
927 *n = -1;
928 break;
929 }
930 if (err < 0)
931 return err;
932 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
933 /* See comments in 'resolve_collision()' */
934 if (*n == (*zn)->child_cnt - 1) {
935 err = tnc_next(c, zn, n);
936 if (err) {
937 /* Should be impossible */
938 ubifs_assert(0);
939 if (err == -ENOENT)
940 err = -EINVAL;
941 return err;
942 }
943 ubifs_assert(*n == 0);
944 *n = -1;
945 }
946 break;
947 }
948 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
949 if (err < 0)
950 return err;
951 if (err == NAME_MATCHES)
952 return 1;
953 if (err == NOT_ON_MEDIA) {
954 o_znode = *zn;
955 o_n = *n;
956 continue;
957 }
958 if (!adding)
959 continue;
960 if (err == NAME_LESS)
961 break;
962 else
963 unsure = 0;
964 }
965 }
966
967 if (cmp == NAME_LESS || unsure) {
968 /* Look right */
969 *zn = znode;
970 *n = nn;
971 while (1) {
972 err = tnc_next(c, &znode, &nn);
973 if (err == -ENOENT)
974 break;
975 if (err < 0)
976 return err;
977 if (keys_cmp(c, &znode->zbranch[nn].key, key))
978 break;
979 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
980 if (err < 0)
981 return err;
982 if (err == NAME_GREATER)
983 break;
984 *zn = znode;
985 *n = nn;
986 if (err == NAME_MATCHES)
987 return 1;
988 if (err == NOT_ON_MEDIA) {
989 o_znode = znode;
990 o_n = nn;
991 }
992 }
993 }
994
995 /* Never match a dangling branch when adding */
996 if (adding || !o_znode)
997 return 0;
998
999 dbg_mnt("dangling match LEB %d:%d len %d %s",
1000 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
1001 o_znode->zbranch[o_n].len, DBGKEY(key));
1002 *zn = o_znode;
1003 *n = o_n;
1004 return 1;
1005}
1006
1007/**
1008 * matches_position - determine if a zbranch matches a given position.
1009 * @zbr: zbranch of dent
1010 * @lnum: LEB number of dent to match
1011 * @offs: offset of dent to match
1012 *
1013 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1014 */
1015static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1016{
1017 if (zbr->lnum == lnum && zbr->offs == offs)
1018 return 1;
1019 else
1020 return 0;
1021}
1022
1023/**
1024 * resolve_collision_directly - resolve a collision directly.
1025 * @c: UBIFS file-system description object
1026 * @key: key of directory entry
1027 * @zn: znode is passed and returned here
1028 * @n: zbranch number is passed and returned here
1029 * @lnum: LEB number of dent node to match
1030 * @offs: offset of dent node to match
1031 *
1032 * This function is used for "hashed" keys to make sure the found directory or
1033 * extended attribute entry node is what was looked for. It is used when the
1034 * flash address of the right node is known (@lnum:@offs) which makes it much
1035 * easier to resolve collisions (no need to read entries and match full
1036 * names). This function returns %1 and sets @zn and @n if the collision is
1037 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1038 * previous directory entry. Otherwise a negative error code is returned.
1039 */
1040static int resolve_collision_directly(struct ubifs_info *c,
1041 const union ubifs_key *key,
1042 struct ubifs_znode **zn, int *n,
1043 int lnum, int offs)
1044{
1045 struct ubifs_znode *znode;
1046 int nn, err;
1047
1048 znode = *zn;
1049 nn = *n;
1050 if (matches_position(&znode->zbranch[nn], lnum, offs))
1051 return 1;
1052
1053 /* Look left */
1054 while (1) {
1055 err = tnc_prev(c, &znode, &nn);
1056 if (err == -ENOENT)
1057 break;
1058 if (err < 0)
1059 return err;
1060 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1061 break;
1062 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1063 *zn = znode;
1064 *n = nn;
1065 return 1;
1066 }
1067 }
1068
1069 /* Look right */
1070 znode = *zn;
1071 nn = *n;
1072 while (1) {
1073 err = tnc_next(c, &znode, &nn);
1074 if (err == -ENOENT)
1075 return 0;
1076 if (err < 0)
1077 return err;
1078 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1079 return 0;
1080 *zn = znode;
1081 *n = nn;
1082 if (matches_position(&znode->zbranch[nn], lnum, offs))
1083 return 1;
1084 }
1085}
1086
1087/**
1088 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1089 * @c: UBIFS file-system description object
1090 * @znode: znode to dirty
1091 *
1092 * If we do not have a unique key that resides in a znode, then we cannot
1093 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1094 * This function records the path back to the last dirty ancestor, and then
1095 * dirties the znodes on that path.
1096 */
1097static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1098 struct ubifs_znode *znode)
1099{
1100 struct ubifs_znode *zp;
1101 int *path = c->bottom_up_buf, p = 0;
1102
1103 ubifs_assert(c->zroot.znode);
1104 ubifs_assert(znode);
1105 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1106 kfree(c->bottom_up_buf);
1107 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1108 GFP_NOFS);
1109 if (!c->bottom_up_buf)
1110 return ERR_PTR(-ENOMEM);
1111 path = c->bottom_up_buf;
1112 }
1113 if (c->zroot.znode->level) {
1114 /* Go up until parent is dirty */
1115 while (1) {
1116 int n;
1117
1118 zp = znode->parent;
1119 if (!zp)
1120 break;
1121 n = znode->iip;
1122 ubifs_assert(p < c->zroot.znode->level);
1123 path[p++] = n;
1124 if (!zp->cnext && ubifs_zn_dirty(znode))
1125 break;
1126 znode = zp;
1127 }
1128 }
1129
1130 /* Come back down, dirtying as we go */
1131 while (1) {
1132 struct ubifs_zbranch *zbr;
1133
1134 zp = znode->parent;
1135 if (zp) {
1136 ubifs_assert(path[p - 1] >= 0);
1137 ubifs_assert(path[p - 1] < zp->child_cnt);
1138 zbr = &zp->zbranch[path[--p]];
1139 znode = dirty_cow_znode(c, zbr);
1140 } else {
1141 ubifs_assert(znode == c->zroot.znode);
1142 znode = dirty_cow_znode(c, &c->zroot);
1143 }
1144 if (IS_ERR(znode) || !p)
1145 break;
1146 ubifs_assert(path[p - 1] >= 0);
1147 ubifs_assert(path[p - 1] < znode->child_cnt);
1148 znode = znode->zbranch[path[p - 1]].znode;
1149 }
1150
1151 return znode;
1152}
1153
1154/**
1155 * ubifs_lookup_level0 - search for zero-level znode.
1156 * @c: UBIFS file-system description object
1157 * @key: key to lookup
1158 * @zn: znode is returned here
1159 * @n: znode branch slot number is returned here
1160 *
1161 * This function looks up the TNC tree and search for zero-level znode which
1162 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1163 * cases:
1164 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1165 * is returned and slot number of the matched branch is stored in @n;
1166 * o not exact match, which means that zero-level znode does not contain
1167 * @key, then %0 is returned and slot number of the closest branch is stored
1168 * in @n;
1169 * o @key is so small that it is even less than the lowest key of the
1170 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1171 *
1172 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1173 * function reads corresponding indexing nodes and inserts them to TNC. In
1174 * case of failure, a negative error code is returned.
1175 */
1176int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1177 struct ubifs_znode **zn, int *n)
1178{
1179 int err, exact;
1180 struct ubifs_znode *znode;
1181 unsigned long time = get_seconds();
1182
1183 dbg_tnc("search key %s", DBGKEY(key));
1184 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
1185
1186 znode = c->zroot.znode;
1187 if (unlikely(!znode)) {
1188 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1189 if (IS_ERR(znode))
1190 return PTR_ERR(znode);
1191 }
1192
1193 znode->time = time;
1194
1195 while (1) {
1196 struct ubifs_zbranch *zbr;
1197
1198 exact = ubifs_search_zbranch(c, znode, key, n);
1199
1200 if (znode->level == 0)
1201 break;
1202
1203 if (*n < 0)
1204 *n = 0;
1205 zbr = &znode->zbranch[*n];
1206
1207 if (zbr->znode) {
1208 znode->time = time;
1209 znode = zbr->znode;
1210 continue;
1211 }
1212
1213 /* znode is not in TNC cache, load it from the media */
1214 znode = ubifs_load_znode(c, zbr, znode, *n);
1215 if (IS_ERR(znode))
1216 return PTR_ERR(znode);
1217 }
1218
1219 *zn = znode;
1220 if (exact || !is_hash_key(c, key) || *n != -1) {
1221 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1222 return exact;
1223 }
1224
1225 /*
1226 * Here is a tricky place. We have not found the key and this is a
1227 * "hashed" key, which may collide. The rest of the code deals with
1228 * situations like this:
1229 *
1230 * | 3 | 5 |
1231 * / \
1232 * | 3 | 5 | | 6 | 7 | (x)
1233 *
1234 * Or more a complex example:
1235 *
1236 * | 1 | 5 |
1237 * / \
1238 * | 1 | 3 | | 5 | 8 |
1239 * \ /
1240 * | 5 | 5 | | 6 | 7 | (x)
1241 *
1242 * In the examples, if we are looking for key "5", we may reach nodes
1243 * marked with "(x)". In this case what we have do is to look at the
1244 * left and see if there is "5" key there. If there is, we have to
1245 * return it.
1246 *
1247 * Note, this whole situation is possible because we allow to have
1248 * elements which are equivalent to the next key in the parent in the
1249 * children of current znode. For example, this happens if we split a
1250 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1251 * like this:
1252 * | 3 | 5 |
1253 * / \
1254 * | 3 | 5 | | 5 | 6 | 7 |
1255 * ^
1256 * And this becomes what is at the first "picture" after key "5" marked
1257 * with "^" is removed. What could be done is we could prohibit
1258 * splitting in the middle of the colliding sequence. Also, when
1259 * removing the leftmost key, we would have to correct the key of the
1260 * parent node, which would introduce additional complications. Namely,
1261 * if we changed the leftmost key of the parent znode, the garbage
1262 * collector would be unable to find it (GC is doing this when GC'ing
1263 * indexing LEBs). Although we already have an additional RB-tree where
1264 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1265 * after the commit. But anyway, this does not look easy to implement
1266 * so we did not try this.
1267 */
1268 err = tnc_prev(c, &znode, n);
1269 if (err == -ENOENT) {
1270 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1271 *n = -1;
1272 return 0;
1273 }
1274 if (unlikely(err < 0))
1275 return err;
1276 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1277 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1278 *n = -1;
1279 return 0;
1280 }
1281
1282 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1283 *zn = znode;
1284 return 1;
1285}
1286
1287/**
1288 * lookup_level0_dirty - search for zero-level znode dirtying.
1289 * @c: UBIFS file-system description object
1290 * @key: key to lookup
1291 * @zn: znode is returned here
1292 * @n: znode branch slot number is returned here
1293 *
1294 * This function looks up the TNC tree and search for zero-level znode which
1295 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1296 * cases:
1297 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1298 * is returned and slot number of the matched branch is stored in @n;
1299 * o not exact match, which means that zero-level znode does not contain @key
1300 * then %0 is returned and slot number of the closed branch is stored in
1301 * @n;
1302 * o @key is so small that it is even less than the lowest key of the
1303 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1304 *
1305 * Additionally all znodes in the path from the root to the located zero-level
1306 * znode are marked as dirty.
1307 *
1308 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1309 * function reads corresponding indexing nodes and inserts them to TNC. In
1310 * case of failure, a negative error code is returned.
1311 */
1312static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1313 struct ubifs_znode **zn, int *n)
1314{
1315 int err, exact;
1316 struct ubifs_znode *znode;
1317 unsigned long time = get_seconds();
1318
1319 dbg_tnc("search and dirty key %s", DBGKEY(key));
1320
1321 znode = c->zroot.znode;
1322 if (unlikely(!znode)) {
1323 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1324 if (IS_ERR(znode))
1325 return PTR_ERR(znode);
1326 }
1327
1328 znode = dirty_cow_znode(c, &c->zroot);
1329 if (IS_ERR(znode))
1330 return PTR_ERR(znode);
1331
1332 znode->time = time;
1333
1334 while (1) {
1335 struct ubifs_zbranch *zbr;
1336
1337 exact = ubifs_search_zbranch(c, znode, key, n);
1338
1339 if (znode->level == 0)
1340 break;
1341
1342 if (*n < 0)
1343 *n = 0;
1344 zbr = &znode->zbranch[*n];
1345
1346 if (zbr->znode) {
1347 znode->time = time;
1348 znode = dirty_cow_znode(c, zbr);
1349 if (IS_ERR(znode))
1350 return PTR_ERR(znode);
1351 continue;
1352 }
1353
1354 /* znode is not in TNC cache, load it from the media */
1355 znode = ubifs_load_znode(c, zbr, znode, *n);
1356 if (IS_ERR(znode))
1357 return PTR_ERR(znode);
1358 znode = dirty_cow_znode(c, zbr);
1359 if (IS_ERR(znode))
1360 return PTR_ERR(znode);
1361 }
1362
1363 *zn = znode;
1364 if (exact || !is_hash_key(c, key) || *n != -1) {
1365 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1366 return exact;
1367 }
1368
1369 /*
1370 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1371 * code.
1372 */
1373 err = tnc_prev(c, &znode, n);
1374 if (err == -ENOENT) {
1375 *n = -1;
1376 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1377 return 0;
1378 }
1379 if (unlikely(err < 0))
1380 return err;
1381 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1382 *n = -1;
1383 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1384 return 0;
1385 }
1386
1387 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1388 znode = dirty_cow_bottom_up(c, znode);
1389 if (IS_ERR(znode))
1390 return PTR_ERR(znode);
1391 }
1392
1393 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1394 *zn = znode;
1395 return 1;
1396}
1397
1398/**
1399 * maybe_leb_gced - determine if a LEB may have been garbage collected.
1400 * @c: UBIFS file-system description object
1401 * @lnum: LEB number
1402 * @gc_seq1: garbage collection sequence number
1403 *
1404 * This function determines if @lnum may have been garbage collected since
1405 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1406 * %0 is returned.
1407 */
1408static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
1409{
1410 int gc_seq2, gced_lnum;
1411
1412 gced_lnum = c->gced_lnum;
1413 smp_rmb();
1414 gc_seq2 = c->gc_seq;
1415 /* Same seq means no GC */
1416 if (gc_seq1 == gc_seq2)
1417 return 0;
1418 /* Different by more than 1 means we don't know */
1419 if (gc_seq1 + 1 != gc_seq2)
1420 return 1;
1421 /*
1422 * We have seen the sequence number has increased by 1. Now we need to
1423 * be sure we read the right LEB number, so read it again.
1424 */
1425 smp_rmb();
1426 if (gced_lnum != c->gced_lnum)
1427 return 1;
1428 /* Finally we can check lnum */
1429 if (gced_lnum == lnum)
1430 return 1;
1431 return 0;
1432}
1433
1434/**
1435 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1436 * @c: UBIFS file-system description object
1437 * @key: node key to lookup
1438 * @node: the node is returned here
1439 * @lnum: LEB number is returned here
1440 * @offs: offset is returned here
1441 *
1442 * This function looks up and reads node with key @key. The caller has to make
1443 * sure the @node buffer is large enough to fit the node. Returns zero in case
1444 * of success, %-ENOENT if the node was not found, and a negative error code in
1445 * case of failure. The node location can be returned in @lnum and @offs.
1446 */
1447int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1448 void *node, int *lnum, int *offs)
1449{
1450 int found, n, err, safely = 0, gc_seq1;
1451 struct ubifs_znode *znode;
1452 struct ubifs_zbranch zbr, *zt;
1453
1454again:
1455 mutex_lock(&c->tnc_mutex);
1456 found = ubifs_lookup_level0(c, key, &znode, &n);
1457 if (!found) {
1458 err = -ENOENT;
1459 goto out;
1460 } else if (found < 0) {
1461 err = found;
1462 goto out;
1463 }
1464 zt = &znode->zbranch[n];
1465 if (lnum) {
1466 *lnum = zt->lnum;
1467 *offs = zt->offs;
1468 }
1469 if (is_hash_key(c, key)) {
1470 /*
1471 * In this case the leaf node cache gets used, so we pass the
1472 * address of the zbranch and keep the mutex locked
1473 */
1474 err = tnc_read_node_nm(c, zt, node);
1475 goto out;
1476 }
1477 if (safely) {
1478 err = ubifs_tnc_read_node(c, zt, node);
1479 goto out;
1480 }
1481 /* Drop the TNC mutex prematurely and race with garbage collection */
1482 zbr = znode->zbranch[n];
1483 gc_seq1 = c->gc_seq;
1484 mutex_unlock(&c->tnc_mutex);
1485
1486 if (ubifs_get_wbuf(c, zbr.lnum)) {
1487 /* We do not GC journal heads */
1488 err = ubifs_tnc_read_node(c, &zbr, node);
1489 return err;
1490 }
1491
1492 err = fallible_read_node(c, key, &zbr, node);
1493 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
1494 /*
1495 * The node may have been GC'ed out from under us so try again
1496 * while keeping the TNC mutex locked.
1497 */
1498 safely = 1;
1499 goto again;
1500 }
1501 return 0;
1502
1503out:
1504 mutex_unlock(&c->tnc_mutex);
1505 return err;
1506}
1507
1508/**
1509 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1510 * @c: UBIFS file-system description object
1511 * @bu: bulk-read parameters and results
1512 *
1513 * Lookup consecutive data node keys for the same inode that reside
1514 * consecutively in the same LEB. This function returns zero in case of success
1515 * and a negative error code in case of failure.
1516 *
1517 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1518 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
1519 * maximum possible amount of nodes for bulk-read.
1520 */
1521int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1522{
1523 int n, err = 0, lnum = -1, uninitialized_var(offs);
1524 int uninitialized_var(len);
1525 unsigned int block = key_block(c, &bu->key);
1526 struct ubifs_znode *znode;
1527
1528 bu->cnt = 0;
1529 bu->blk_cnt = 0;
1530 bu->eof = 0;
1531
1532 mutex_lock(&c->tnc_mutex);
1533 /* Find first key */
1534 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1535 if (err < 0)
1536 goto out;
1537 if (err) {
1538 /* Key found */
1539 len = znode->zbranch[n].len;
1540 /* The buffer must be big enough for at least 1 node */
1541 if (len > bu->buf_len) {
1542 err = -EINVAL;
1543 goto out;
1544 }
1545 /* Add this key */
1546 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1547 bu->blk_cnt += 1;
1548 lnum = znode->zbranch[n].lnum;
1549 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1550 }
1551 while (1) {
1552 struct ubifs_zbranch *zbr;
1553 union ubifs_key *key;
1554 unsigned int next_block;
1555
1556 /* Find next key */
1557 err = tnc_next(c, &znode, &n);
1558 if (err)
1559 goto out;
1560 zbr = &znode->zbranch[n];
1561 key = &zbr->key;
1562 /* See if there is another data key for this file */
1563 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1564 key_type(c, key) != UBIFS_DATA_KEY) {
1565 err = -ENOENT;
1566 goto out;
1567 }
1568 if (lnum < 0) {
1569 /* First key found */
1570 lnum = zbr->lnum;
1571 offs = ALIGN(zbr->offs + zbr->len, 8);
1572 len = zbr->len;
1573 if (len > bu->buf_len) {
1574 err = -EINVAL;
1575 goto out;
1576 }
1577 } else {
1578 /*
1579 * The data nodes must be in consecutive positions in
1580 * the same LEB.
1581 */
1582 if (zbr->lnum != lnum || zbr->offs != offs)
1583 goto out;
1584 offs += ALIGN(zbr->len, 8);
1585 len = ALIGN(len, 8) + zbr->len;
1586 /* Must not exceed buffer length */
1587 if (len > bu->buf_len)
1588 goto out;
1589 }
1590 /* Allow for holes */
1591 next_block = key_block(c, key);
1592 bu->blk_cnt += (next_block - block - 1);
1593 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1594 goto out;
1595 block = next_block;
1596 /* Add this key */
1597 bu->zbranch[bu->cnt++] = *zbr;
1598 bu->blk_cnt += 1;
1599 /* See if we have room for more */
1600 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1601 goto out;
1602 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1603 goto out;
1604 }
1605out:
1606 if (err == -ENOENT) {
1607 bu->eof = 1;
1608 err = 0;
1609 }
1610 bu->gc_seq = c->gc_seq;
1611 mutex_unlock(&c->tnc_mutex);
1612 if (err)
1613 return err;
1614 /*
1615 * An enormous hole could cause bulk-read to encompass too many
1616 * page cache pages, so limit the number here.
1617 */
1618 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
1619 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1620 /*
1621 * Ensure that bulk-read covers a whole number of page cache
1622 * pages.
1623 */
1624 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1625 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1626 return 0;
1627 if (bu->eof) {
1628 /* At the end of file we can round up */
1629 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1630 return 0;
1631 }
1632 /* Exclude data nodes that do not make up a whole page cache page */
1633 block = key_block(c, &bu->key) + bu->blk_cnt;
1634 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1635 while (bu->cnt) {
1636 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1637 break;
1638 bu->cnt -= 1;
1639 }
1640 return 0;
1641}
1642
1643/**
1644 * read_wbuf - bulk-read from a LEB with a wbuf.
1645 * @wbuf: wbuf that may overlap the read
1646 * @buf: buffer into which to read
1647 * @len: read length
1648 * @lnum: LEB number from which to read
1649 * @offs: offset from which to read
1650 *
1651 * This functions returns %0 on success or a negative error code on failure.
1652 */
1653static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1654 int offs)
1655{
1656 const struct ubifs_info *c = wbuf->c;
1657 int rlen, overlap;
1658
1659 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1660 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1661 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1662 ubifs_assert(offs + len <= c->leb_size);
1663
1664 spin_lock(&wbuf->lock);
1665 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1666 if (!overlap) {
1667 /* We may safely unlock the write-buffer and read the data */
1668 spin_unlock(&wbuf->lock);
1669 return ubifs_leb_read(c, lnum, buf, offs, len, 0);
1670 }
1671
1672 /* Don't read under wbuf */
1673 rlen = wbuf->offs - offs;
1674 if (rlen < 0)
1675 rlen = 0;
1676
1677 /* Copy the rest from the write-buffer */
1678 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1679 spin_unlock(&wbuf->lock);
1680
1681 if (rlen > 0)
1682 /* Read everything that goes before write-buffer */
1683 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0);
1684
1685 return 0;
1686}
1687
1688/**
1689 * validate_data_node - validate data nodes for bulk-read.
1690 * @c: UBIFS file-system description object
1691 * @buf: buffer containing data node to validate
1692 * @zbr: zbranch of data node to validate
1693 *
1694 * This functions returns %0 on success or a negative error code on failure.
1695 */
1696static int validate_data_node(struct ubifs_info *c, void *buf,
1697 struct ubifs_zbranch *zbr)
1698{
1699 union ubifs_key key1;
1700 struct ubifs_ch *ch = buf;
1701 int err, len;
1702
1703 if (ch->node_type != UBIFS_DATA_NODE) {
1704 ubifs_err("bad node type (%d but expected %d)",
1705 ch->node_type, UBIFS_DATA_NODE);
1706 goto out_err;
1707 }
1708
1709 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
1710 if (err) {
1711 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1712 goto out;
1713 }
1714
1715 len = le32_to_cpu(ch->len);
1716 if (len != zbr->len) {
1717 ubifs_err("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("bad key in node at LEB %d:%d",
1725 zbr->lnum, zbr->offs);
1726 dbg_tnc("looked for key %s found node's key %s",
1727 DBGKEY(&zbr->key), DBGKEY1(&key1));
1728 goto out_err;
1729 }
1730
1731 return 0;
1732
1733out_err:
1734 err = -EINVAL;
1735out:
1736 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1737 dbg_dump_node(c, buf);
1738 dbg_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("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("failed to read from LEB %d:%d, error %d",
1778 lnum, offs, err);
1779 dbg_dump_stack();
1780 dbg_tnc("key %s", DBGKEY(&bu->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 look 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 qstr *nm)
1811{
1812 int found, n, err;
1813 struct ubifs_znode *znode;
1814
1815 dbg_tnc("name '%.*s' key %s", nm->len, nm->name, DBGKEY(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(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_node_nm(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 look 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 qstr *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 (nm->len == len && !memcmp(dent->name, nm->name, 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 return do_lookup_nm(c, key, node, nm);
1880}
1881
1882/**
1883 * correct_parent_keys - correct parent znodes' keys.
1884 * @c: UBIFS file-system description object
1885 * @znode: znode to correct parent znodes for
1886 *
1887 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1888 * zbranch changes, keys of parent znodes have to be corrected. This helper
1889 * function is called in such situations and corrects the keys if needed.
1890 */
1891static void correct_parent_keys(const struct ubifs_info *c,
1892 struct ubifs_znode *znode)
1893{
1894 union ubifs_key *key, *key1;
1895
1896 ubifs_assert(znode->parent);
1897 ubifs_assert(znode->iip == 0);
1898
1899 key = &znode->zbranch[0].key;
1900 key1 = &znode->parent->zbranch[0].key;
1901
1902 while (keys_cmp(c, key, key1) < 0) {
1903 key_copy(c, key, key1);
1904 znode = znode->parent;
1905 znode->alt = 1;
1906 if (!znode->parent || znode->iip)
1907 break;
1908 key1 = &znode->parent->zbranch[0].key;
1909 }
1910}
1911
1912/**
1913 * insert_zbranch - insert a zbranch into a znode.
1914 * @znode: znode into which to insert
1915 * @zbr: zbranch to insert
1916 * @n: slot number to insert to
1917 *
1918 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1919 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1920 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1921 * slot, zbranches starting from @n have to be moved right.
1922 */
1923static void insert_zbranch(struct ubifs_znode *znode,
1924 const struct ubifs_zbranch *zbr, int n)
1925{
1926 int i;
1927
1928 ubifs_assert(ubifs_zn_dirty(znode));
1929
1930 if (znode->level) {
1931 for (i = znode->child_cnt; i > n; i--) {
1932 znode->zbranch[i] = znode->zbranch[i - 1];
1933 if (znode->zbranch[i].znode)
1934 znode->zbranch[i].znode->iip = i;
1935 }
1936 if (zbr->znode)
1937 zbr->znode->iip = n;
1938 } else
1939 for (i = znode->child_cnt; i > n; i--)
1940 znode->zbranch[i] = znode->zbranch[i - 1];
1941
1942 znode->zbranch[n] = *zbr;
1943 znode->child_cnt += 1;
1944
1945 /*
1946 * After inserting at slot zero, the lower bound of the key range of
1947 * this znode may have changed. If this znode is subsequently split
1948 * then the upper bound of the key range may change, and furthermore
1949 * it could change to be lower than the original lower bound. If that
1950 * happens, then it will no longer be possible to find this znode in the
1951 * TNC using the key from the index node on flash. That is bad because
1952 * if it is not found, we will assume it is obsolete and may overwrite
1953 * it. Then if there is an unclean unmount, we will start using the
1954 * old index which will be broken.
1955 *
1956 * So we first mark znodes that have insertions at slot zero, and then
1957 * if they are split we add their lnum/offs to the old_idx tree.
1958 */
1959 if (n == 0)
1960 znode->alt = 1;
1961}
1962
1963/**
1964 * tnc_insert - insert a node into TNC.
1965 * @c: UBIFS file-system description object
1966 * @znode: znode to insert into
1967 * @zbr: branch to insert
1968 * @n: slot number to insert new zbranch to
1969 *
1970 * This function inserts a new node described by @zbr into znode @znode. If
1971 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1972 * are splat as well if needed. Returns zero in case of success or a negative
1973 * error code in case of failure.
1974 */
1975static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1976 struct ubifs_zbranch *zbr, int n)
1977{
1978 struct ubifs_znode *zn, *zi, *zp;
1979 int i, keep, move, appending = 0;
1980 union ubifs_key *key = &zbr->key, *key1;
1981
1982 ubifs_assert(n >= 0 && n <= c->fanout);
1983
1984 /* Implement naive insert for now */
1985again:
1986 zp = znode->parent;
1987 if (znode->child_cnt < c->fanout) {
1988 ubifs_assert(n != c->fanout);
1989 dbg_tnc("inserted at %d level %d, key %s", n, znode->level,
1990 DBGKEY(key));
1991
1992 insert_zbranch(znode, zbr, n);
1993
1994 /* Ensure parent's key is correct */
1995 if (n == 0 && zp && znode->iip == 0)
1996 correct_parent_keys(c, znode);
1997
1998 return 0;
1999 }
2000
2001 /*
2002 * Unfortunately, @znode does not have more empty slots and we have to
2003 * split it.
2004 */
2005 dbg_tnc("splitting level %d, key %s", znode->level, DBGKEY(key));
2006
2007 if (znode->alt)
2008 /*
2009 * We can no longer be sure of finding this znode by key, so we
2010 * record it in the old_idx tree.
2011 */
2012 ins_clr_old_idx_znode(c, znode);
2013
2014 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2015 if (!zn)
2016 return -ENOMEM;
2017 zn->parent = zp;
2018 zn->level = znode->level;
2019
2020 /* Decide where to split */
2021 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2022 /* Try not to split consecutive data keys */
2023 if (n == c->fanout) {
2024 key1 = &znode->zbranch[n - 1].key;
2025 if (key_inum(c, key1) == key_inum(c, key) &&
2026 key_type(c, key1) == UBIFS_DATA_KEY)
2027 appending = 1;
2028 } else
2029 goto check_split;
2030 } else if (appending && n != c->fanout) {
2031 /* Try not to split consecutive data keys */
2032 appending = 0;
2033check_split:
2034 if (n >= (c->fanout + 1) / 2) {
2035 key1 = &znode->zbranch[0].key;
2036 if (key_inum(c, key1) == key_inum(c, key) &&
2037 key_type(c, key1) == UBIFS_DATA_KEY) {
2038 key1 = &znode->zbranch[n].key;
2039 if (key_inum(c, key1) != key_inum(c, key) ||
2040 key_type(c, key1) != UBIFS_DATA_KEY) {
2041 keep = n;
2042 move = c->fanout - keep;
2043 zi = znode;
2044 goto do_split;
2045 }
2046 }
2047 }
2048 }
2049
2050 if (appending) {
2051 keep = c->fanout;
2052 move = 0;
2053 } else {
2054 keep = (c->fanout + 1) / 2;
2055 move = c->fanout - keep;
2056 }
2057
2058 /*
2059 * Although we don't at present, we could look at the neighbors and see
2060 * if we can move some zbranches there.
2061 */
2062
2063 if (n < keep) {
2064 /* Insert into existing znode */
2065 zi = znode;
2066 move += 1;
2067 keep -= 1;
2068 } else {
2069 /* Insert into new znode */
2070 zi = zn;
2071 n -= keep;
2072 /* Re-parent */
2073 if (zn->level != 0)
2074 zbr->znode->parent = zn;
2075 }
2076
2077do_split:
2078
2079 __set_bit(DIRTY_ZNODE, &zn->flags);
2080 atomic_long_inc(&c->dirty_zn_cnt);
2081
2082 zn->child_cnt = move;
2083 znode->child_cnt = keep;
2084
2085 dbg_tnc("moving %d, keeping %d", move, keep);
2086
2087 /* Move zbranch */
2088 for (i = 0; i < move; i++) {
2089 zn->zbranch[i] = znode->zbranch[keep + i];
2090 /* Re-parent */
2091 if (zn->level != 0)
2092 if (zn->zbranch[i].znode) {
2093 zn->zbranch[i].znode->parent = zn;
2094 zn->zbranch[i].znode->iip = i;
2095 }
2096 }
2097
2098 /* Insert new key and branch */
2099 dbg_tnc("inserting at %d level %d, key %s", n, zn->level, DBGKEY(key));
2100
2101 insert_zbranch(zi, zbr, n);
2102
2103 /* Insert new znode (produced by spitting) into the parent */
2104 if (zp) {
2105 if (n == 0 && zi == znode && znode->iip == 0)
2106 correct_parent_keys(c, znode);
2107
2108 /* Locate insertion point */
2109 n = znode->iip + 1;
2110
2111 /* Tail recursion */
2112 zbr->key = zn->zbranch[0].key;
2113 zbr->znode = zn;
2114 zbr->lnum = 0;
2115 zbr->offs = 0;
2116 zbr->len = 0;
2117 znode = zp;
2118
2119 goto again;
2120 }
2121
2122 /* We have to split root znode */
2123 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2124
2125 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2126 if (!zi)
2127 return -ENOMEM;
2128
2129 zi->child_cnt = 2;
2130 zi->level = znode->level + 1;
2131
2132 __set_bit(DIRTY_ZNODE, &zi->flags);
2133 atomic_long_inc(&c->dirty_zn_cnt);
2134
2135 zi->zbranch[0].key = znode->zbranch[0].key;
2136 zi->zbranch[0].znode = znode;
2137 zi->zbranch[0].lnum = c->zroot.lnum;
2138 zi->zbranch[0].offs = c->zroot.offs;
2139 zi->zbranch[0].len = c->zroot.len;
2140 zi->zbranch[1].key = zn->zbranch[0].key;
2141 zi->zbranch[1].znode = zn;
2142
2143 c->zroot.lnum = 0;
2144 c->zroot.offs = 0;
2145 c->zroot.len = 0;
2146 c->zroot.znode = zi;
2147
2148 zn->parent = zi;
2149 zn->iip = 1;
2150 znode->parent = zi;
2151 znode->iip = 0;
2152
2153 return 0;
2154}
2155
2156/**
2157 * ubifs_tnc_add - add a node to TNC.
2158 * @c: UBIFS file-system description object
2159 * @key: key to add
2160 * @lnum: LEB number of node
2161 * @offs: node offset
2162 * @len: node length
2163 *
2164 * This function adds a node with key @key to TNC. The node may be new or it may
2165 * obsolete some existing one. Returns %0 on success or negative error code on
2166 * failure.
2167 */
2168int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2169 int offs, int len)
2170{
2171 int found, n, err = 0;
2172 struct ubifs_znode *znode;
2173
2174 mutex_lock(&c->tnc_mutex);
2175 dbg_tnc("%d:%d, len %d, key %s", lnum, offs, len, DBGKEY(key));
2176 found = lookup_level0_dirty(c, key, &znode, &n);
2177 if (!found) {
2178 struct ubifs_zbranch zbr;
2179
2180 zbr.znode = NULL;
2181 zbr.lnum = lnum;
2182 zbr.offs = offs;
2183 zbr.len = len;
2184 key_copy(c, key, &zbr.key);
2185 err = tnc_insert(c, znode, &zbr, n + 1);
2186 } else if (found == 1) {
2187 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2188
2189 lnc_free(zbr);
2190 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2191 zbr->lnum = lnum;
2192 zbr->offs = offs;
2193 zbr->len = len;
2194 } else
2195 err = found;
2196 if (!err)
2197 err = dbg_check_tnc(c, 0);
2198 mutex_unlock(&c->tnc_mutex);
2199
2200 return err;
2201}
2202
2203/**
2204 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2205 * @c: UBIFS file-system description object
2206 * @key: key to add
2207 * @old_lnum: LEB number of old node
2208 * @old_offs: old node offset
2209 * @lnum: LEB number of node
2210 * @offs: node offset
2211 * @len: node length
2212 *
2213 * This function replaces a node with key @key in the TNC only if the old node
2214 * is found. This function is called by garbage collection when node are moved.
2215 * Returns %0 on success or negative error code on failure.
2216 */
2217int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2218 int old_lnum, int old_offs, int lnum, int offs, int len)
2219{
2220 int found, n, err = 0;
2221 struct ubifs_znode *znode;
2222
2223 mutex_lock(&c->tnc_mutex);
2224 dbg_tnc("old LEB %d:%d, new LEB %d:%d, len %d, key %s", old_lnum,
2225 old_offs, lnum, offs, len, DBGKEY(key));
2226 found = lookup_level0_dirty(c, key, &znode, &n);
2227 if (found < 0) {
2228 err = found;
2229 goto out_unlock;
2230 }
2231
2232 if (found == 1) {
2233 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2234
2235 found = 0;
2236 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2237 lnc_free(zbr);
2238 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2239 if (err)
2240 goto out_unlock;
2241 zbr->lnum = lnum;
2242 zbr->offs = offs;
2243 zbr->len = len;
2244 found = 1;
2245 } else if (is_hash_key(c, key)) {
2246 found = resolve_collision_directly(c, key, &znode, &n,
2247 old_lnum, old_offs);
2248 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2249 found, znode, n, old_lnum, old_offs);
2250 if (found < 0) {
2251 err = found;
2252 goto out_unlock;
2253 }
2254
2255 if (found) {
2256 /* Ensure the znode is dirtied */
2257 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2258 znode = dirty_cow_bottom_up(c, znode);
2259 if (IS_ERR(znode)) {
2260 err = PTR_ERR(znode);
2261 goto out_unlock;
2262 }
2263 }
2264 zbr = &znode->zbranch[n];
2265 lnc_free(zbr);
2266 err = ubifs_add_dirt(c, zbr->lnum,
2267 zbr->len);
2268 if (err)
2269 goto out_unlock;
2270 zbr->lnum = lnum;
2271 zbr->offs = offs;
2272 zbr->len = len;
2273 }
2274 }
2275 }
2276
2277 if (!found)
2278 err = ubifs_add_dirt(c, lnum, len);
2279
2280 if (!err)
2281 err = dbg_check_tnc(c, 0);
2282
2283out_unlock:
2284 mutex_unlock(&c->tnc_mutex);
2285 return err;
2286}
2287
2288/**
2289 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2290 * @c: UBIFS file-system description object
2291 * @key: key to add
2292 * @lnum: LEB number of node
2293 * @offs: node offset
2294 * @len: node length
2295 * @nm: node name
2296 *
2297 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2298 * may have collisions, like directory entry keys.
2299 */
2300int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2301 int lnum, int offs, int len, const struct qstr *nm)
2302{
2303 int found, n, err = 0;
2304 struct ubifs_znode *znode;
2305
2306 mutex_lock(&c->tnc_mutex);
2307 dbg_tnc("LEB %d:%d, name '%.*s', key %s", lnum, offs, nm->len, nm->name,
2308 DBGKEY(key));
2309 found = lookup_level0_dirty(c, key, &znode, &n);
2310 if (found < 0) {
2311 err = found;
2312 goto out_unlock;
2313 }
2314
2315 if (found == 1) {
2316 if (c->replaying)
2317 found = fallible_resolve_collision(c, key, &znode, &n,
2318 nm, 1);
2319 else
2320 found = resolve_collision(c, key, &znode, &n, nm);
2321 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2322 if (found < 0) {
2323 err = found;
2324 goto out_unlock;
2325 }
2326
2327 /* Ensure the znode is dirtied */
2328 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2329 znode = dirty_cow_bottom_up(c, znode);
2330 if (IS_ERR(znode)) {
2331 err = PTR_ERR(znode);
2332 goto out_unlock;
2333 }
2334 }
2335
2336 if (found == 1) {
2337 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2338
2339 lnc_free(zbr);
2340 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2341 zbr->lnum = lnum;
2342 zbr->offs = offs;
2343 zbr->len = len;
2344 goto out_unlock;
2345 }
2346 }
2347
2348 if (!found) {
2349 struct ubifs_zbranch zbr;
2350
2351 zbr.znode = NULL;
2352 zbr.lnum = lnum;
2353 zbr.offs = offs;
2354 zbr.len = len;
2355 key_copy(c, key, &zbr.key);
2356 err = tnc_insert(c, znode, &zbr, n + 1);
2357 if (err)
2358 goto out_unlock;
2359 if (c->replaying) {
2360 /*
2361 * We did not find it in the index so there may be a
2362 * dangling branch still in the index. So we remove it
2363 * by passing 'ubifs_tnc_remove_nm()' the same key but
2364 * an unmatchable name.
2365 */
2366 struct qstr noname = { .len = 0, .name = "" };
2367
2368 err = dbg_check_tnc(c, 0);
2369 mutex_unlock(&c->tnc_mutex);
2370 if (err)
2371 return err;
2372 return ubifs_tnc_remove_nm(c, key, &noname);
2373 }
2374 }
2375
2376out_unlock:
2377 if (!err)
2378 err = dbg_check_tnc(c, 0);
2379 mutex_unlock(&c->tnc_mutex);
2380 return err;
2381}
2382
2383/**
2384 * tnc_delete - delete a znode form TNC.
2385 * @c: UBIFS file-system description object
2386 * @znode: znode to delete from
2387 * @n: zbranch slot number to delete
2388 *
2389 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2390 * case of success and a negative error code in case of failure.
2391 */
2392static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2393{
2394 struct ubifs_zbranch *zbr;
2395 struct ubifs_znode *zp;
2396 int i, err;
2397
2398 /* Delete without merge for now */
2399 ubifs_assert(znode->level == 0);
2400 ubifs_assert(n >= 0 && n < c->fanout);
2401 dbg_tnc("deleting %s", DBGKEY(&znode->zbranch[n].key));
2402
2403 zbr = &znode->zbranch[n];
2404 lnc_free(zbr);
2405
2406 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2407 if (err) {
2408 dbg_dump_znode(c, znode);
2409 return err;
2410 }
2411
2412 /* We do not "gap" zbranch slots */
2413 for (i = n; i < znode->child_cnt - 1; i++)
2414 znode->zbranch[i] = znode->zbranch[i + 1];
2415 znode->child_cnt -= 1;
2416
2417 if (znode->child_cnt > 0)
2418 return 0;
2419
2420 /*
2421 * This was the last zbranch, we have to delete this znode from the
2422 * parent.
2423 */
2424
2425 do {
2426 ubifs_assert(!ubifs_zn_obsolete(znode));
2427 ubifs_assert(ubifs_zn_dirty(znode));
2428
2429 zp = znode->parent;
2430 n = znode->iip;
2431
2432 atomic_long_dec(&c->dirty_zn_cnt);
2433
2434 err = insert_old_idx_znode(c, znode);
2435 if (err)
2436 return err;
2437
2438 if (znode->cnext) {
2439 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2440 atomic_long_inc(&c->clean_zn_cnt);
2441 atomic_long_inc(&ubifs_clean_zn_cnt);
2442 } else
2443 kfree(znode);
2444 znode = zp;
2445 } while (znode->child_cnt == 1); /* while removing last child */
2446
2447 /* Remove from znode, entry n - 1 */
2448 znode->child_cnt -= 1;
2449 ubifs_assert(znode->level != 0);
2450 for (i = n; i < znode->child_cnt; i++) {
2451 znode->zbranch[i] = znode->zbranch[i + 1];
2452 if (znode->zbranch[i].znode)
2453 znode->zbranch[i].znode->iip = i;
2454 }
2455
2456 /*
2457 * If this is the root and it has only 1 child then
2458 * collapse the tree.
2459 */
2460 if (!znode->parent) {
2461 while (znode->child_cnt == 1 && znode->level != 0) {
2462 zp = znode;
2463 zbr = &znode->zbranch[0];
2464 znode = get_znode(c, znode, 0);
2465 if (IS_ERR(znode))
2466 return PTR_ERR(znode);
2467 znode = dirty_cow_znode(c, zbr);
2468 if (IS_ERR(znode))
2469 return PTR_ERR(znode);
2470 znode->parent = NULL;
2471 znode->iip = 0;
2472 if (c->zroot.len) {
2473 err = insert_old_idx(c, c->zroot.lnum,
2474 c->zroot.offs);
2475 if (err)
2476 return err;
2477 }
2478 c->zroot.lnum = zbr->lnum;
2479 c->zroot.offs = zbr->offs;
2480 c->zroot.len = zbr->len;
2481 c->zroot.znode = znode;
2482 ubifs_assert(!ubifs_zn_obsolete(zp));
2483 ubifs_assert(ubifs_zn_dirty(zp));
2484 atomic_long_dec(&c->dirty_zn_cnt);
2485
2486 if (zp->cnext) {
2487 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2488 atomic_long_inc(&c->clean_zn_cnt);
2489 atomic_long_inc(&ubifs_clean_zn_cnt);
2490 } else
2491 kfree(zp);
2492 }
2493 }
2494
2495 return 0;
2496}
2497
2498/**
2499 * ubifs_tnc_remove - remove an index entry of a node.
2500 * @c: UBIFS file-system description object
2501 * @key: key of node
2502 *
2503 * Returns %0 on success or negative error code on failure.
2504 */
2505int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2506{
2507 int found, n, err = 0;
2508 struct ubifs_znode *znode;
2509
2510 mutex_lock(&c->tnc_mutex);
2511 dbg_tnc("key %s", DBGKEY(key));
2512 found = lookup_level0_dirty(c, key, &znode, &n);
2513 if (found < 0) {
2514 err = found;
2515 goto out_unlock;
2516 }
2517 if (found == 1)
2518 err = tnc_delete(c, znode, n);
2519 if (!err)
2520 err = dbg_check_tnc(c, 0);
2521
2522out_unlock:
2523 mutex_unlock(&c->tnc_mutex);
2524 return err;
2525}
2526
2527/**
2528 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2529 * @c: UBIFS file-system description object
2530 * @key: key of node
2531 * @nm: directory entry name
2532 *
2533 * Returns %0 on success or negative error code on failure.
2534 */
2535int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2536 const struct qstr *nm)
2537{
2538 int n, err;
2539 struct ubifs_znode *znode;
2540
2541 mutex_lock(&c->tnc_mutex);
2542 dbg_tnc("%.*s, key %s", nm->len, nm->name, DBGKEY(key));
2543 err = lookup_level0_dirty(c, key, &znode, &n);
2544 if (err < 0)
2545 goto out_unlock;
2546
2547 if (err) {
2548 if (c->replaying)
2549 err = fallible_resolve_collision(c, key, &znode, &n,
2550 nm, 0);
2551 else
2552 err = resolve_collision(c, key, &znode, &n, nm);
2553 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2554 if (err < 0)
2555 goto out_unlock;
2556 if (err) {
2557 /* Ensure the znode is dirtied */
2558 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2559 znode = dirty_cow_bottom_up(c, znode);
2560 if (IS_ERR(znode)) {
2561 err = PTR_ERR(znode);
2562 goto out_unlock;
2563 }
2564 }
2565 err = tnc_delete(c, znode, n);
2566 }
2567 }
2568
2569out_unlock:
2570 if (!err)
2571 err = dbg_check_tnc(c, 0);
2572 mutex_unlock(&c->tnc_mutex);
2573 return err;
2574}
2575
2576/**
2577 * key_in_range - determine if a key falls within a range of keys.
2578 * @c: UBIFS file-system description object
2579 * @key: key to check
2580 * @from_key: lowest key in range
2581 * @to_key: highest key in range
2582 *
2583 * This function returns %1 if the key is in range and %0 otherwise.
2584 */
2585static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2586 union ubifs_key *from_key, union ubifs_key *to_key)
2587{
2588 if (keys_cmp(c, key, from_key) < 0)
2589 return 0;
2590 if (keys_cmp(c, key, to_key) > 0)
2591 return 0;
2592 return 1;
2593}
2594
2595/**
2596 * ubifs_tnc_remove_range - remove index entries in range.
2597 * @c: UBIFS file-system description object
2598 * @from_key: lowest key to remove
2599 * @to_key: highest key to remove
2600 *
2601 * This function removes index entries starting at @from_key and ending at
2602 * @to_key. This function returns zero in case of success and a negative error
2603 * code in case of failure.
2604 */
2605int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2606 union ubifs_key *to_key)
2607{
2608 int i, n, k, err = 0;
2609 struct ubifs_znode *znode;
2610 union ubifs_key *key;
2611
2612 mutex_lock(&c->tnc_mutex);
2613 while (1) {
2614 /* Find first level 0 znode that contains keys to remove */
2615 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2616 if (err < 0)
2617 goto out_unlock;
2618
2619 if (err)
2620 key = from_key;
2621 else {
2622 err = tnc_next(c, &znode, &n);
2623 if (err == -ENOENT) {
2624 err = 0;
2625 goto out_unlock;
2626 }
2627 if (err < 0)
2628 goto out_unlock;
2629 key = &znode->zbranch[n].key;
2630 if (!key_in_range(c, key, from_key, to_key)) {
2631 err = 0;
2632 goto out_unlock;
2633 }
2634 }
2635
2636 /* Ensure the znode is dirtied */
2637 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2638 znode = dirty_cow_bottom_up(c, znode);
2639 if (IS_ERR(znode)) {
2640 err = PTR_ERR(znode);
2641 goto out_unlock;
2642 }
2643 }
2644
2645 /* Remove all keys in range except the first */
2646 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2647 key = &znode->zbranch[i].key;
2648 if (!key_in_range(c, key, from_key, to_key))
2649 break;
2650 lnc_free(&znode->zbranch[i]);
2651 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2652 znode->zbranch[i].len);
2653 if (err) {
2654 dbg_dump_znode(c, znode);
2655 goto out_unlock;
2656 }
2657 dbg_tnc("removing %s", DBGKEY(key));
2658 }
2659 if (k) {
2660 for (i = n + 1 + k; i < znode->child_cnt; i++)
2661 znode->zbranch[i - k] = znode->zbranch[i];
2662 znode->child_cnt -= k;
2663 }
2664
2665 /* Now delete the first */
2666 err = tnc_delete(c, znode, n);
2667 if (err)
2668 goto out_unlock;
2669 }
2670
2671out_unlock:
2672 if (!err)
2673 err = dbg_check_tnc(c, 0);
2674 mutex_unlock(&c->tnc_mutex);
2675 return err;
2676}
2677
2678/**
2679 * ubifs_tnc_remove_ino - remove an inode from TNC.
2680 * @c: UBIFS file-system description object
2681 * @inum: inode number to remove
2682 *
2683 * This function remove inode @inum and all the extended attributes associated
2684 * with the anode from TNC and returns zero in case of success or a negative
2685 * error code in case of failure.
2686 */
2687int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2688{
2689 union ubifs_key key1, key2;
2690 struct ubifs_dent_node *xent, *pxent = NULL;
2691 struct qstr nm = { .name = NULL };
2692
2693 dbg_tnc("ino %lu", (unsigned long)inum);
2694
2695 /*
2696 * Walk all extended attribute entries and remove them together with
2697 * corresponding extended attribute inodes.
2698 */
2699 lowest_xent_key(c, &key1, inum);
2700 while (1) {
2701 ino_t xattr_inum;
2702 int err;
2703
2704 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2705 if (IS_ERR(xent)) {
2706 err = PTR_ERR(xent);
2707 if (err == -ENOENT)
2708 break;
2709 return err;
2710 }
2711
2712 xattr_inum = le64_to_cpu(xent->inum);
2713 dbg_tnc("xent '%s', ino %lu", xent->name,
2714 (unsigned long)xattr_inum);
2715
2716 nm.name = xent->name;
2717 nm.len = le16_to_cpu(xent->nlen);
2718 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2719 if (err) {
2720 kfree(xent);
2721 return err;
2722 }
2723
2724 lowest_ino_key(c, &key1, xattr_inum);
2725 highest_ino_key(c, &key2, xattr_inum);
2726 err = ubifs_tnc_remove_range(c, &key1, &key2);
2727 if (err) {
2728 kfree(xent);
2729 return err;
2730 }
2731
2732 kfree(pxent);
2733 pxent = xent;
2734 key_read(c, &xent->key, &key1);
2735 }
2736
2737 kfree(pxent);
2738 lowest_ino_key(c, &key1, inum);
2739 highest_ino_key(c, &key2, inum);
2740
2741 return ubifs_tnc_remove_range(c, &key1, &key2);
2742}
2743
2744/**
2745 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2746 * @c: UBIFS file-system description object
2747 * @key: key of last entry
2748 * @nm: name of last entry found or %NULL
2749 *
2750 * This function finds and reads the next directory or extended attribute entry
2751 * after the given key (@key) if there is one. @nm is used to resolve
2752 * collisions.
2753 *
2754 * If the name of the current entry is not known and only the key is known,
2755 * @nm->name has to be %NULL. In this case the semantics of this function is a
2756 * little bit different and it returns the entry corresponding to this key, not
2757 * the next one. If the key was not found, the closest "right" entry is
2758 * returned.
2759 *
2760 * If the fist entry has to be found, @key has to contain the lowest possible
2761 * key value for this inode and @name has to be %NULL.
2762 *
2763 * This function returns the found directory or extended attribute entry node
2764 * in case of success, %-ENOENT is returned if no entry was found, and a
2765 * negative error code is returned in case of failure.
2766 */
2767struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2768 union ubifs_key *key,
2769 const struct qstr *nm)
2770{
2771 int n, err, type = key_type(c, key);
2772 struct ubifs_znode *znode;
2773 struct ubifs_dent_node *dent;
2774 struct ubifs_zbranch *zbr;
2775 union ubifs_key *dkey;
2776
2777 dbg_tnc("%s %s", nm->name ? (char *)nm->name : "(lowest)", DBGKEY(key));
2778 ubifs_assert(is_hash_key(c, key));
2779
2780 mutex_lock(&c->tnc_mutex);
2781 err = ubifs_lookup_level0(c, key, &znode, &n);
2782 if (unlikely(err < 0))
2783 goto out_unlock;
2784
2785 if (nm->name) {
2786 if (err) {
2787 /* Handle collisions */
2788 err = resolve_collision(c, key, &znode, &n, nm);
2789 dbg_tnc("rc returned %d, znode %p, n %d",
2790 err, znode, n);
2791 if (unlikely(err < 0))
2792 goto out_unlock;
2793 }
2794
2795 /* Now find next entry */
2796 err = tnc_next(c, &znode, &n);
2797 if (unlikely(err))
2798 goto out_unlock;
2799 } else {
2800 /*
2801 * The full name of the entry was not given, in which case the
2802 * behavior of this function is a little different and it
2803 * returns current entry, not the next one.
2804 */
2805 if (!err) {
2806 /*
2807 * However, the given key does not exist in the TNC
2808 * tree and @znode/@n variables contain the closest
2809 * "preceding" element. Switch to the next one.
2810 */
2811 err = tnc_next(c, &znode, &n);
2812 if (err)
2813 goto out_unlock;
2814 }
2815 }
2816
2817 zbr = &znode->zbranch[n];
2818 dent = kmalloc(zbr->len, GFP_NOFS);
2819 if (unlikely(!dent)) {
2820 err = -ENOMEM;
2821 goto out_unlock;
2822 }
2823
2824 /*
2825 * The above 'tnc_next()' call could lead us to the next inode, check
2826 * this.
2827 */
2828 dkey = &zbr->key;
2829 if (key_inum(c, dkey) != key_inum(c, key) ||
2830 key_type(c, dkey) != type) {
2831 err = -ENOENT;
2832 goto out_free;
2833 }
2834
2835 err = tnc_read_node_nm(c, zbr, dent);
2836 if (unlikely(err))
2837 goto out_free;
2838
2839 mutex_unlock(&c->tnc_mutex);
2840 return dent;
2841
2842out_free:
2843 kfree(dent);
2844out_unlock:
2845 mutex_unlock(&c->tnc_mutex);
2846 return ERR_PTR(err);
2847}
2848
2849/**
2850 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2851 * @c: UBIFS file-system description object
2852 *
2853 * Destroy left-over obsolete znodes from a failed commit.
2854 */
2855static void tnc_destroy_cnext(struct ubifs_info *c)
2856{
2857 struct ubifs_znode *cnext;
2858
2859 if (!c->cnext)
2860 return;
2861 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2862 cnext = c->cnext;
2863 do {
2864 struct ubifs_znode *znode = cnext;
2865
2866 cnext = cnext->cnext;
2867 if (ubifs_zn_obsolete(znode))
2868 kfree(znode);
2869 } while (cnext && cnext != c->cnext);
2870}
2871
2872/**
2873 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2874 * @c: UBIFS file-system description object
2875 */
2876void ubifs_tnc_close(struct ubifs_info *c)
2877{
2878 tnc_destroy_cnext(c);
2879 if (c->zroot.znode) {
2880 long n;
2881
2882 ubifs_destroy_tnc_subtree(c->zroot.znode);
2883 n = atomic_long_read(&c->clean_zn_cnt);
2884 atomic_long_sub(n, &ubifs_clean_zn_cnt);
2885 }
2886 kfree(c->gap_lebs);
2887 kfree(c->ilebs);
2888 destroy_old_idx(c);
2889}
2890
2891/**
2892 * left_znode - get the znode to the left.
2893 * @c: UBIFS file-system description object
2894 * @znode: znode
2895 *
2896 * This function returns a pointer to the znode to the left of @znode or NULL if
2897 * there is not one. A negative error code is returned on failure.
2898 */
2899static struct ubifs_znode *left_znode(struct ubifs_info *c,
2900 struct ubifs_znode *znode)
2901{
2902 int level = znode->level;
2903
2904 while (1) {
2905 int n = znode->iip - 1;
2906
2907 /* Go up until we can go left */
2908 znode = znode->parent;
2909 if (!znode)
2910 return NULL;
2911 if (n >= 0) {
2912 /* Now go down the rightmost branch to 'level' */
2913 znode = get_znode(c, znode, n);
2914 if (IS_ERR(znode))
2915 return znode;
2916 while (znode->level != level) {
2917 n = znode->child_cnt - 1;
2918 znode = get_znode(c, znode, n);
2919 if (IS_ERR(znode))
2920 return znode;
2921 }
2922 break;
2923 }
2924 }
2925 return znode;
2926}
2927
2928/**
2929 * right_znode - get the znode to the right.
2930 * @c: UBIFS file-system description object
2931 * @znode: znode
2932 *
2933 * This function returns a pointer to the znode to the right of @znode or NULL
2934 * if there is not one. A negative error code is returned on failure.
2935 */
2936static struct ubifs_znode *right_znode(struct ubifs_info *c,
2937 struct ubifs_znode *znode)
2938{
2939 int level = znode->level;
2940
2941 while (1) {
2942 int n = znode->iip + 1;
2943
2944 /* Go up until we can go right */
2945 znode = znode->parent;
2946 if (!znode)
2947 return NULL;
2948 if (n < znode->child_cnt) {
2949 /* Now go down the leftmost branch to 'level' */
2950 znode = get_znode(c, znode, n);
2951 if (IS_ERR(znode))
2952 return znode;
2953 while (znode->level != level) {
2954 znode = get_znode(c, znode, 0);
2955 if (IS_ERR(znode))
2956 return znode;
2957 }
2958 break;
2959 }
2960 }
2961 return znode;
2962}
2963
2964/**
2965 * lookup_znode - find a particular indexing node from TNC.
2966 * @c: UBIFS file-system description object
2967 * @key: index node key to lookup
2968 * @level: index node level
2969 * @lnum: index node LEB number
2970 * @offs: index node offset
2971 *
2972 * This function searches an indexing node by its first key @key and its
2973 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2974 * nodes it traverses to TNC. This function is called for indexing nodes which
2975 * were found on the media by scanning, for example when garbage-collecting or
2976 * when doing in-the-gaps commit. This means that the indexing node which is
2977 * looked for does not have to have exactly the same leftmost key @key, because
2978 * the leftmost key may have been changed, in which case TNC will contain a
2979 * dirty znode which still refers the same @lnum:@offs. This function is clever
2980 * enough to recognize such indexing nodes.
2981 *
2982 * Note, if a znode was deleted or changed too much, then this function will
2983 * not find it. For situations like this UBIFS has the old index RB-tree
2984 * (indexed by @lnum:@offs).
2985 *
2986 * This function returns a pointer to the znode found or %NULL if it is not
2987 * found. A negative error code is returned on failure.
2988 */
2989static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2990 union ubifs_key *key, int level,
2991 int lnum, int offs)
2992{
2993 struct ubifs_znode *znode, *zn;
2994 int n, nn;
2995
2996 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY);
2997
2998 /*
2999 * The arguments have probably been read off flash, so don't assume
3000 * they are valid.
3001 */
3002 if (level < 0)
3003 return ERR_PTR(-EINVAL);
3004
3005 /* Get the root znode */
3006 znode = c->zroot.znode;
3007 if (!znode) {
3008 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3009 if (IS_ERR(znode))
3010 return znode;
3011 }
3012 /* Check if it is the one we are looking for */
3013 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3014 return znode;
3015 /* Descend to the parent level i.e. (level + 1) */
3016 if (level >= znode->level)
3017 return NULL;
3018 while (1) {
3019 ubifs_search_zbranch(c, znode, key, &n);
3020 if (n < 0) {
3021 /*
3022 * We reached a znode where the leftmost key is greater
3023 * than the key we are searching for. This is the same
3024 * situation as the one described in a huge comment at
3025 * the end of the 'ubifs_lookup_level0()' function. And
3026 * for exactly the same reasons we have to try to look
3027 * left before giving up.
3028 */
3029 znode = left_znode(c, znode);
3030 if (!znode)
3031 return NULL;
3032 if (IS_ERR(znode))
3033 return znode;
3034 ubifs_search_zbranch(c, znode, key, &n);
3035 ubifs_assert(n >= 0);
3036 }
3037 if (znode->level == level + 1)
3038 break;
3039 znode = get_znode(c, znode, n);
3040 if (IS_ERR(znode))
3041 return znode;
3042 }
3043 /* Check if the child is the one we are looking for */
3044 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3045 return get_znode(c, znode, n);
3046 /* If the key is unique, there is nowhere else to look */
3047 if (!is_hash_key(c, key))
3048 return NULL;
3049 /*
3050 * The key is not unique and so may be also in the znodes to either
3051 * side.
3052 */
3053 zn = znode;
3054 nn = n;
3055 /* Look left */
3056 while (1) {
3057 /* Move one branch to the left */
3058 if (n)
3059 n -= 1;
3060 else {
3061 znode = left_znode(c, znode);
3062 if (!znode)
3063 break;
3064 if (IS_ERR(znode))
3065 return znode;
3066 n = znode->child_cnt - 1;
3067 }
3068 /* Check it */
3069 if (znode->zbranch[n].lnum == lnum &&
3070 znode->zbranch[n].offs == offs)
3071 return get_znode(c, znode, n);
3072 /* Stop if the key is less than the one we are looking for */
3073 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3074 break;
3075 }
3076 /* Back to the middle */
3077 znode = zn;
3078 n = nn;
3079 /* Look right */
3080 while (1) {
3081 /* Move one branch to the right */
3082 if (++n >= znode->child_cnt) {
3083 znode = right_znode(c, znode);
3084 if (!znode)
3085 break;
3086 if (IS_ERR(znode))
3087 return znode;
3088 n = 0;
3089 }
3090 /* Check it */
3091 if (znode->zbranch[n].lnum == lnum &&
3092 znode->zbranch[n].offs == offs)
3093 return get_znode(c, znode, n);
3094 /* Stop if the key is greater than the one we are looking for */
3095 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3096 break;
3097 }
3098 return NULL;
3099}
3100
3101/**
3102 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3103 * @c: UBIFS file-system description object
3104 * @key: key of index node
3105 * @level: index node level
3106 * @lnum: LEB number of index node
3107 * @offs: offset of index node
3108 *
3109 * This function returns %0 if the index node is not referred to in the TNC, %1
3110 * if the index node is referred to in the TNC and the corresponding znode is
3111 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3112 * znode is clean, and a negative error code in case of failure.
3113 *
3114 * Note, the @key argument has to be the key of the first child. Also note,
3115 * this function relies on the fact that 0:0 is never a valid LEB number and
3116 * offset for a main-area node.
3117 */
3118int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3119 int lnum, int offs)
3120{
3121 struct ubifs_znode *znode;
3122
3123 znode = lookup_znode(c, key, level, lnum, offs);
3124 if (!znode)
3125 return 0;
3126 if (IS_ERR(znode))
3127 return PTR_ERR(znode);
3128
3129 return ubifs_zn_dirty(znode) ? 1 : 2;
3130}
3131
3132/**
3133 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3134 * @c: UBIFS file-system description object
3135 * @key: node key
3136 * @lnum: node LEB number
3137 * @offs: node offset
3138 *
3139 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3140 * not, and a negative error code in case of failure.
3141 *
3142 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3143 * and offset for a main-area node.
3144 */
3145static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3146 int lnum, int offs)
3147{
3148 struct ubifs_zbranch *zbr;
3149 struct ubifs_znode *znode, *zn;
3150 int n, found, err, nn;
3151 const int unique = !is_hash_key(c, key);
3152
3153 found = ubifs_lookup_level0(c, key, &znode, &n);
3154 if (found < 0)
3155 return found; /* Error code */
3156 if (!found)
3157 return 0;
3158 zbr = &znode->zbranch[n];
3159 if (lnum == zbr->lnum && offs == zbr->offs)
3160 return 1; /* Found it */
3161 if (unique)
3162 return 0;
3163 /*
3164 * Because the key is not unique, we have to look left
3165 * and right as well
3166 */
3167 zn = znode;
3168 nn = n;
3169 /* Look left */
3170 while (1) {
3171 err = tnc_prev(c, &znode, &n);
3172 if (err == -ENOENT)
3173 break;
3174 if (err)
3175 return err;
3176 if (keys_cmp(c, key, &znode->zbranch[n].key))
3177 break;
3178 zbr = &znode->zbranch[n];
3179 if (lnum == zbr->lnum && offs == zbr->offs)
3180 return 1; /* Found it */
3181 }
3182 /* Look right */
3183 znode = zn;
3184 n = nn;
3185 while (1) {
3186 err = tnc_next(c, &znode, &n);
3187 if (err) {
3188 if (err == -ENOENT)
3189 return 0;
3190 return err;
3191 }
3192 if (keys_cmp(c, key, &znode->zbranch[n].key))
3193 break;
3194 zbr = &znode->zbranch[n];
3195 if (lnum == zbr->lnum && offs == zbr->offs)
3196 return 1; /* Found it */
3197 }
3198 return 0;
3199}
3200
3201/**
3202 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3203 * @c: UBIFS file-system description object
3204 * @key: node key
3205 * @level: index node level (if it is an index node)
3206 * @lnum: node LEB number
3207 * @offs: node offset
3208 * @is_idx: non-zero if the node is an index node
3209 *
3210 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3211 * negative error code in case of failure. For index nodes, @key has to be the
3212 * key of the first child. An index node is considered to be in the TNC only if
3213 * the corresponding znode is clean or has not been loaded.
3214 */
3215int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3216 int lnum, int offs, int is_idx)
3217{
3218 int err;
3219
3220 mutex_lock(&c->tnc_mutex);
3221 if (is_idx) {
3222 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3223 if (err < 0)
3224 goto out_unlock;
3225 if (err == 1)
3226 /* The index node was found but it was dirty */
3227 err = 0;
3228 else if (err == 2)
3229 /* The index node was found and it was clean */
3230 err = 1;
3231 else
3232 BUG_ON(err != 0);
3233 } else
3234 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3235
3236out_unlock:
3237 mutex_unlock(&c->tnc_mutex);
3238 return err;
3239}
3240
3241/**
3242 * ubifs_dirty_idx_node - dirty an index node.
3243 * @c: UBIFS file-system description object
3244 * @key: index node key
3245 * @level: index node level
3246 * @lnum: index node LEB number
3247 * @offs: index node offset
3248 *
3249 * This function loads and dirties an index node so that it can be garbage
3250 * collected. The @key argument has to be the key of the first child. This
3251 * function relies on the fact that 0:0 is never a valid LEB number and offset
3252 * for a main-area node. Returns %0 on success and a negative error code on
3253 * failure.
3254 */
3255int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3256 int lnum, int offs)
3257{
3258 struct ubifs_znode *znode;
3259 int err = 0;
3260
3261 mutex_lock(&c->tnc_mutex);
3262 znode = lookup_znode(c, key, level, lnum, offs);
3263 if (!znode)
3264 goto out_unlock;
3265 if (IS_ERR(znode)) {
3266 err = PTR_ERR(znode);
3267 goto out_unlock;
3268 }
3269 znode = dirty_cow_bottom_up(c, znode);
3270 if (IS_ERR(znode)) {
3271 err = PTR_ERR(znode);
3272 goto out_unlock;
3273 }
3274
3275out_unlock:
3276 mutex_unlock(&c->tnc_mutex);
3277 return err;
3278}
3279
3280#ifdef CONFIG_UBIFS_FS_DEBUG
3281
3282/**
3283 * dbg_check_inode_size - check if inode size is correct.
3284 * @c: UBIFS file-system description object
3285 * @inum: inode number
3286 * @size: inode size
3287 *
3288 * This function makes sure that the inode size (@size) is correct and it does
3289 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3290 * if it has a data page beyond @size, and other negative error code in case of
3291 * other errors.
3292 */
3293int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3294 loff_t size)
3295{
3296 int err, n;
3297 union ubifs_key from_key, to_key, *key;
3298 struct ubifs_znode *znode;
3299 unsigned int block;
3300
3301 if (!S_ISREG(inode->i_mode))
3302 return 0;
3303 if (!dbg_is_chk_gen(c))
3304 return 0;
3305
3306 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3307 data_key_init(c, &from_key, inode->i_ino, block);
3308 highest_data_key(c, &to_key, inode->i_ino);
3309
3310 mutex_lock(&c->tnc_mutex);
3311 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3312 if (err < 0)
3313 goto out_unlock;
3314
3315 if (err) {
3316 err = -EINVAL;
3317 key = &from_key;
3318 goto out_dump;
3319 }
3320
3321 err = tnc_next(c, &znode, &n);
3322 if (err == -ENOENT) {
3323 err = 0;
3324 goto out_unlock;
3325 }
3326 if (err < 0)
3327 goto out_unlock;
3328
3329 ubifs_assert(err == 0);
3330 key = &znode->zbranch[n].key;
3331 if (!key_in_range(c, key, &from_key, &to_key))
3332 goto out_unlock;
3333
3334out_dump:
3335 block = key_block(c, key);
3336 ubifs_err("inode %lu has size %lld, but there are data at offset %lld "
3337 "(data key %s)", (unsigned long)inode->i_ino, size,
3338 ((loff_t)block) << UBIFS_BLOCK_SHIFT, DBGKEY(key));
3339 mutex_unlock(&c->tnc_mutex);
3340 dbg_dump_inode(c, inode);
3341 dbg_dump_stack();
3342 return -EINVAL;
3343
3344out_unlock:
3345 mutex_unlock(&c->tnc_mutex);
3346 return err;
3347}
3348
3349#endif /* CONFIG_UBIFS_FS_DEBUG */
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 fname_name(&nm) = xent->name;
2934 fname_len(&nm) = le16_to_cpu(xent->nlen);
2935 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2936 if (err) {
2937 kfree(pxent);
2938 kfree(xent);
2939 return err;
2940 }
2941
2942 lowest_ino_key(c, &key1, xattr_inum);
2943 highest_ino_key(c, &key2, xattr_inum);
2944 err = ubifs_tnc_remove_range(c, &key1, &key2);
2945 if (err) {
2946 kfree(pxent);
2947 kfree(xent);
2948 return err;
2949 }
2950
2951 kfree(pxent);
2952 pxent = xent;
2953 key_read(c, &xent->key, &key1);
2954 }
2955
2956 kfree(pxent);
2957 lowest_ino_key(c, &key1, inum);
2958 highest_ino_key(c, &key2, inum);
2959
2960 return ubifs_tnc_remove_range(c, &key1, &key2);
2961}
2962
2963/**
2964 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2965 * @c: UBIFS file-system description object
2966 * @key: key of last entry
2967 * @nm: name of last entry found or %NULL
2968 *
2969 * This function finds and reads the next directory or extended attribute entry
2970 * after the given key (@key) if there is one. @nm is used to resolve
2971 * collisions.
2972 *
2973 * If the name of the current entry is not known and only the key is known,
2974 * @nm->name has to be %NULL. In this case the semantics of this function is a
2975 * little bit different and it returns the entry corresponding to this key, not
2976 * the next one. If the key was not found, the closest "right" entry is
2977 * returned.
2978 *
2979 * If the fist entry has to be found, @key has to contain the lowest possible
2980 * key value for this inode and @name has to be %NULL.
2981 *
2982 * This function returns the found directory or extended attribute entry node
2983 * in case of success, %-ENOENT is returned if no entry was found, and a
2984 * negative error code is returned in case of failure.
2985 */
2986struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2987 union ubifs_key *key,
2988 const struct fscrypt_name *nm)
2989{
2990 int n, err, type = key_type(c, key);
2991 struct ubifs_znode *znode;
2992 struct ubifs_dent_node *dent;
2993 struct ubifs_zbranch *zbr;
2994 union ubifs_key *dkey;
2995
2996 dbg_tnck(key, "key ");
2997 ubifs_assert(c, is_hash_key(c, key));
2998
2999 mutex_lock(&c->tnc_mutex);
3000 err = ubifs_lookup_level0(c, key, &znode, &n);
3001 if (unlikely(err < 0))
3002 goto out_unlock;
3003
3004 if (fname_len(nm) > 0) {
3005 if (err) {
3006 /* Handle collisions */
3007 if (c->replaying)
3008 err = fallible_resolve_collision(c, key, &znode, &n,
3009 nm, 0);
3010 else
3011 err = resolve_collision(c, key, &znode, &n, nm);
3012 dbg_tnc("rc returned %d, znode %p, n %d",
3013 err, znode, n);
3014 if (unlikely(err < 0))
3015 goto out_unlock;
3016 }
3017
3018 /* Now find next entry */
3019 err = tnc_next(c, &znode, &n);
3020 if (unlikely(err))
3021 goto out_unlock;
3022 } else {
3023 /*
3024 * The full name of the entry was not given, in which case the
3025 * behavior of this function is a little different and it
3026 * returns current entry, not the next one.
3027 */
3028 if (!err) {
3029 /*
3030 * However, the given key does not exist in the TNC
3031 * tree and @znode/@n variables contain the closest
3032 * "preceding" element. Switch to the next one.
3033 */
3034 err = tnc_next(c, &znode, &n);
3035 if (err)
3036 goto out_unlock;
3037 }
3038 }
3039
3040 zbr = &znode->zbranch[n];
3041 dent = kmalloc(zbr->len, GFP_NOFS);
3042 if (unlikely(!dent)) {
3043 err = -ENOMEM;
3044 goto out_unlock;
3045 }
3046
3047 /*
3048 * The above 'tnc_next()' call could lead us to the next inode, check
3049 * this.
3050 */
3051 dkey = &zbr->key;
3052 if (key_inum(c, dkey) != key_inum(c, key) ||
3053 key_type(c, dkey) != type) {
3054 err = -ENOENT;
3055 goto out_free;
3056 }
3057
3058 err = tnc_read_hashed_node(c, zbr, dent);
3059 if (unlikely(err))
3060 goto out_free;
3061
3062 mutex_unlock(&c->tnc_mutex);
3063 return dent;
3064
3065out_free:
3066 kfree(dent);
3067out_unlock:
3068 mutex_unlock(&c->tnc_mutex);
3069 return ERR_PTR(err);
3070}
3071
3072/**
3073 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
3074 * @c: UBIFS file-system description object
3075 *
3076 * Destroy left-over obsolete znodes from a failed commit.
3077 */
3078static void tnc_destroy_cnext(struct ubifs_info *c)
3079{
3080 struct ubifs_znode *cnext;
3081
3082 if (!c->cnext)
3083 return;
3084 ubifs_assert(c, c->cmt_state == COMMIT_BROKEN);
3085 cnext = c->cnext;
3086 do {
3087 struct ubifs_znode *znode = cnext;
3088
3089 cnext = cnext->cnext;
3090 if (ubifs_zn_obsolete(znode))
3091 kfree(znode);
3092 else if (!ubifs_zn_cow(znode)) {
3093 /*
3094 * Don't forget to update clean znode count after
3095 * committing failed, because ubifs will check this
3096 * count while closing tnc. Non-obsolete znode could
3097 * be re-dirtied during committing process, so dirty
3098 * flag is untrustable. The flag 'COW_ZNODE' is set
3099 * for each dirty znode before committing, and it is
3100 * cleared as long as the znode become clean, so we
3101 * can statistic clean znode count according to this
3102 * flag.
3103 */
3104 atomic_long_inc(&c->clean_zn_cnt);
3105 atomic_long_inc(&ubifs_clean_zn_cnt);
3106 }
3107 } while (cnext && cnext != c->cnext);
3108}
3109
3110/**
3111 * ubifs_tnc_close - close TNC subsystem and free all related resources.
3112 * @c: UBIFS file-system description object
3113 */
3114void ubifs_tnc_close(struct ubifs_info *c)
3115{
3116 tnc_destroy_cnext(c);
3117 ubifs_destroy_tnc_tree(c);
3118 kfree(c->gap_lebs);
3119 kfree(c->ilebs);
3120 destroy_old_idx(c);
3121}
3122
3123/**
3124 * left_znode - get the znode to the left.
3125 * @c: UBIFS file-system description object
3126 * @znode: znode
3127 *
3128 * This function returns a pointer to the znode to the left of @znode or NULL if
3129 * there is not one. A negative error code is returned on failure.
3130 */
3131static struct ubifs_znode *left_znode(struct ubifs_info *c,
3132 struct ubifs_znode *znode)
3133{
3134 int level = znode->level;
3135
3136 while (1) {
3137 int n = znode->iip - 1;
3138
3139 /* Go up until we can go left */
3140 znode = znode->parent;
3141 if (!znode)
3142 return NULL;
3143 if (n >= 0) {
3144 /* Now go down the rightmost branch to 'level' */
3145 znode = get_znode(c, znode, n);
3146 if (IS_ERR(znode))
3147 return znode;
3148 while (znode->level != level) {
3149 n = znode->child_cnt - 1;
3150 znode = get_znode(c, znode, n);
3151 if (IS_ERR(znode))
3152 return znode;
3153 }
3154 break;
3155 }
3156 }
3157 return znode;
3158}
3159
3160/**
3161 * right_znode - get the znode to the right.
3162 * @c: UBIFS file-system description object
3163 * @znode: znode
3164 *
3165 * This function returns a pointer to the znode to the right of @znode or NULL
3166 * if there is not one. A negative error code is returned on failure.
3167 */
3168static struct ubifs_znode *right_znode(struct ubifs_info *c,
3169 struct ubifs_znode *znode)
3170{
3171 int level = znode->level;
3172
3173 while (1) {
3174 int n = znode->iip + 1;
3175
3176 /* Go up until we can go right */
3177 znode = znode->parent;
3178 if (!znode)
3179 return NULL;
3180 if (n < znode->child_cnt) {
3181 /* Now go down the leftmost branch to 'level' */
3182 znode = get_znode(c, znode, n);
3183 if (IS_ERR(znode))
3184 return znode;
3185 while (znode->level != level) {
3186 znode = get_znode(c, znode, 0);
3187 if (IS_ERR(znode))
3188 return znode;
3189 }
3190 break;
3191 }
3192 }
3193 return znode;
3194}
3195
3196/**
3197 * lookup_znode - find a particular indexing node from TNC.
3198 * @c: UBIFS file-system description object
3199 * @key: index node key to lookup
3200 * @level: index node level
3201 * @lnum: index node LEB number
3202 * @offs: index node offset
3203 *
3204 * This function searches an indexing node by its first key @key and its
3205 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
3206 * nodes it traverses to TNC. This function is called for indexing nodes which
3207 * were found on the media by scanning, for example when garbage-collecting or
3208 * when doing in-the-gaps commit. This means that the indexing node which is
3209 * looked for does not have to have exactly the same leftmost key @key, because
3210 * the leftmost key may have been changed, in which case TNC will contain a
3211 * dirty znode which still refers the same @lnum:@offs. This function is clever
3212 * enough to recognize such indexing nodes.
3213 *
3214 * Note, if a znode was deleted or changed too much, then this function will
3215 * not find it. For situations like this UBIFS has the old index RB-tree
3216 * (indexed by @lnum:@offs).
3217 *
3218 * This function returns a pointer to the znode found or %NULL if it is not
3219 * found. A negative error code is returned on failure.
3220 */
3221static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
3222 union ubifs_key *key, int level,
3223 int lnum, int offs)
3224{
3225 struct ubifs_znode *znode, *zn;
3226 int n, nn;
3227
3228 ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY);
3229
3230 /*
3231 * The arguments have probably been read off flash, so don't assume
3232 * they are valid.
3233 */
3234 if (level < 0)
3235 return ERR_PTR(-EINVAL);
3236
3237 /* Get the root znode */
3238 znode = c->zroot.znode;
3239 if (!znode) {
3240 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3241 if (IS_ERR(znode))
3242 return znode;
3243 }
3244 /* Check if it is the one we are looking for */
3245 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3246 return znode;
3247 /* Descend to the parent level i.e. (level + 1) */
3248 if (level >= znode->level)
3249 return NULL;
3250 while (1) {
3251 ubifs_search_zbranch(c, znode, key, &n);
3252 if (n < 0) {
3253 /*
3254 * We reached a znode where the leftmost key is greater
3255 * than the key we are searching for. This is the same
3256 * situation as the one described in a huge comment at
3257 * the end of the 'ubifs_lookup_level0()' function. And
3258 * for exactly the same reasons we have to try to look
3259 * left before giving up.
3260 */
3261 znode = left_znode(c, znode);
3262 if (!znode)
3263 return NULL;
3264 if (IS_ERR(znode))
3265 return znode;
3266 ubifs_search_zbranch(c, znode, key, &n);
3267 ubifs_assert(c, n >= 0);
3268 }
3269 if (znode->level == level + 1)
3270 break;
3271 znode = get_znode(c, znode, n);
3272 if (IS_ERR(znode))
3273 return znode;
3274 }
3275 /* Check if the child is the one we are looking for */
3276 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3277 return get_znode(c, znode, n);
3278 /* If the key is unique, there is nowhere else to look */
3279 if (!is_hash_key(c, key))
3280 return NULL;
3281 /*
3282 * The key is not unique and so may be also in the znodes to either
3283 * side.
3284 */
3285 zn = znode;
3286 nn = n;
3287 /* Look left */
3288 while (1) {
3289 /* Move one branch to the left */
3290 if (n)
3291 n -= 1;
3292 else {
3293 znode = left_znode(c, znode);
3294 if (!znode)
3295 break;
3296 if (IS_ERR(znode))
3297 return znode;
3298 n = znode->child_cnt - 1;
3299 }
3300 /* Check it */
3301 if (znode->zbranch[n].lnum == lnum &&
3302 znode->zbranch[n].offs == offs)
3303 return get_znode(c, znode, n);
3304 /* Stop if the key is less than the one we are looking for */
3305 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3306 break;
3307 }
3308 /* Back to the middle */
3309 znode = zn;
3310 n = nn;
3311 /* Look right */
3312 while (1) {
3313 /* Move one branch to the right */
3314 if (++n >= znode->child_cnt) {
3315 znode = right_znode(c, znode);
3316 if (!znode)
3317 break;
3318 if (IS_ERR(znode))
3319 return znode;
3320 n = 0;
3321 }
3322 /* Check it */
3323 if (znode->zbranch[n].lnum == lnum &&
3324 znode->zbranch[n].offs == offs)
3325 return get_znode(c, znode, n);
3326 /* Stop if the key is greater than the one we are looking for */
3327 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3328 break;
3329 }
3330 return NULL;
3331}
3332
3333/**
3334 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3335 * @c: UBIFS file-system description object
3336 * @key: key of index node
3337 * @level: index node level
3338 * @lnum: LEB number of index node
3339 * @offs: offset of index node
3340 *
3341 * This function returns %0 if the index node is not referred to in the TNC, %1
3342 * if the index node is referred to in the TNC and the corresponding znode is
3343 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3344 * znode is clean, and a negative error code in case of failure.
3345 *
3346 * Note, the @key argument has to be the key of the first child. Also note,
3347 * this function relies on the fact that 0:0 is never a valid LEB number and
3348 * offset for a main-area node.
3349 */
3350int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3351 int lnum, int offs)
3352{
3353 struct ubifs_znode *znode;
3354
3355 znode = lookup_znode(c, key, level, lnum, offs);
3356 if (!znode)
3357 return 0;
3358 if (IS_ERR(znode))
3359 return PTR_ERR(znode);
3360
3361 return ubifs_zn_dirty(znode) ? 1 : 2;
3362}
3363
3364/**
3365 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3366 * @c: UBIFS file-system description object
3367 * @key: node key
3368 * @lnum: node LEB number
3369 * @offs: node offset
3370 *
3371 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3372 * not, and a negative error code in case of failure.
3373 *
3374 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3375 * and offset for a main-area node.
3376 */
3377static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3378 int lnum, int offs)
3379{
3380 struct ubifs_zbranch *zbr;
3381 struct ubifs_znode *znode, *zn;
3382 int n, found, err, nn;
3383 const int unique = !is_hash_key(c, key);
3384
3385 found = ubifs_lookup_level0(c, key, &znode, &n);
3386 if (found < 0)
3387 return found; /* Error code */
3388 if (!found)
3389 return 0;
3390 zbr = &znode->zbranch[n];
3391 if (lnum == zbr->lnum && offs == zbr->offs)
3392 return 1; /* Found it */
3393 if (unique)
3394 return 0;
3395 /*
3396 * Because the key is not unique, we have to look left
3397 * and right as well
3398 */
3399 zn = znode;
3400 nn = n;
3401 /* Look left */
3402 while (1) {
3403 err = tnc_prev(c, &znode, &n);
3404 if (err == -ENOENT)
3405 break;
3406 if (err)
3407 return err;
3408 if (keys_cmp(c, key, &znode->zbranch[n].key))
3409 break;
3410 zbr = &znode->zbranch[n];
3411 if (lnum == zbr->lnum && offs == zbr->offs)
3412 return 1; /* Found it */
3413 }
3414 /* Look right */
3415 znode = zn;
3416 n = nn;
3417 while (1) {
3418 err = tnc_next(c, &znode, &n);
3419 if (err) {
3420 if (err == -ENOENT)
3421 return 0;
3422 return err;
3423 }
3424 if (keys_cmp(c, key, &znode->zbranch[n].key))
3425 break;
3426 zbr = &znode->zbranch[n];
3427 if (lnum == zbr->lnum && offs == zbr->offs)
3428 return 1; /* Found it */
3429 }
3430 return 0;
3431}
3432
3433/**
3434 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3435 * @c: UBIFS file-system description object
3436 * @key: node key
3437 * @level: index node level (if it is an index node)
3438 * @lnum: node LEB number
3439 * @offs: node offset
3440 * @is_idx: non-zero if the node is an index node
3441 *
3442 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3443 * negative error code in case of failure. For index nodes, @key has to be the
3444 * key of the first child. An index node is considered to be in the TNC only if
3445 * the corresponding znode is clean or has not been loaded.
3446 */
3447int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3448 int lnum, int offs, int is_idx)
3449{
3450 int err;
3451
3452 mutex_lock(&c->tnc_mutex);
3453 if (is_idx) {
3454 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3455 if (err < 0)
3456 goto out_unlock;
3457 if (err == 1)
3458 /* The index node was found but it was dirty */
3459 err = 0;
3460 else if (err == 2)
3461 /* The index node was found and it was clean */
3462 err = 1;
3463 else
3464 BUG_ON(err != 0);
3465 } else
3466 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3467
3468out_unlock:
3469 mutex_unlock(&c->tnc_mutex);
3470 return err;
3471}
3472
3473/**
3474 * ubifs_dirty_idx_node - dirty an index node.
3475 * @c: UBIFS file-system description object
3476 * @key: index node key
3477 * @level: index node level
3478 * @lnum: index node LEB number
3479 * @offs: index node offset
3480 *
3481 * This function loads and dirties an index node so that it can be garbage
3482 * collected. The @key argument has to be the key of the first child. This
3483 * function relies on the fact that 0:0 is never a valid LEB number and offset
3484 * for a main-area node. Returns %0 on success and a negative error code on
3485 * failure.
3486 */
3487int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3488 int lnum, int offs)
3489{
3490 struct ubifs_znode *znode;
3491 int err = 0;
3492
3493 mutex_lock(&c->tnc_mutex);
3494 znode = lookup_znode(c, key, level, lnum, offs);
3495 if (!znode)
3496 goto out_unlock;
3497 if (IS_ERR(znode)) {
3498 err = PTR_ERR(znode);
3499 goto out_unlock;
3500 }
3501 znode = dirty_cow_bottom_up(c, znode);
3502 if (IS_ERR(znode)) {
3503 err = PTR_ERR(znode);
3504 goto out_unlock;
3505 }
3506
3507out_unlock:
3508 mutex_unlock(&c->tnc_mutex);
3509 return err;
3510}
3511
3512/**
3513 * dbg_check_inode_size - check if inode size is correct.
3514 * @c: UBIFS file-system description object
3515 * @inode: inode to check
3516 * @size: inode size
3517 *
3518 * This function makes sure that the inode size (@size) is correct and it does
3519 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3520 * if it has a data page beyond @size, and other negative error code in case of
3521 * other errors.
3522 */
3523int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3524 loff_t size)
3525{
3526 int err, n;
3527 union ubifs_key from_key, to_key, *key;
3528 struct ubifs_znode *znode;
3529 unsigned int block;
3530
3531 if (!S_ISREG(inode->i_mode))
3532 return 0;
3533 if (!dbg_is_chk_gen(c))
3534 return 0;
3535
3536 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3537 data_key_init(c, &from_key, inode->i_ino, block);
3538 highest_data_key(c, &to_key, inode->i_ino);
3539
3540 mutex_lock(&c->tnc_mutex);
3541 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3542 if (err < 0)
3543 goto out_unlock;
3544
3545 if (err) {
3546 key = &from_key;
3547 goto out_dump;
3548 }
3549
3550 err = tnc_next(c, &znode, &n);
3551 if (err == -ENOENT) {
3552 err = 0;
3553 goto out_unlock;
3554 }
3555 if (err < 0)
3556 goto out_unlock;
3557
3558 ubifs_assert(c, err == 0);
3559 key = &znode->zbranch[n].key;
3560 if (!key_in_range(c, key, &from_key, &to_key))
3561 goto out_unlock;
3562
3563out_dump:
3564 block = key_block(c, key);
3565 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld",
3566 (unsigned long)inode->i_ino, size,
3567 ((loff_t)block) << UBIFS_BLOCK_SHIFT);
3568 mutex_unlock(&c->tnc_mutex);
3569 ubifs_dump_inode(c, inode);
3570 dump_stack();
3571 return -EINVAL;
3572
3573out_unlock:
3574 mutex_unlock(&c->tnc_mutex);
3575 return err;
3576}