blob: 2194915220e56c75bc9894724496dc19c797b908 [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030035#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(!test_bit(OBSOLETE_ZNODE, &znode->flags));
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 (!test_bit(COW_ZNODE, &znode->flags)) {
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);
Hirofumi Nakagawa8d47aef2008-08-21 17:16:40 +0300288 if (IS_ERR(zn))
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300289 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.
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +0200447 *
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->always_chk_crc is true, @c->no_chk_data_crc is ignored and CRC is always
451 * checked.
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300452 */
453static int try_read_node(const struct ubifs_info *c, void *buf, int type,
454 int len, int lnum, int offs)
455{
456 int err, node_len;
457 struct ubifs_ch *ch = buf;
458 uint32_t crc, node_crc;
459
460 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len);
461
462 err = ubi_read(c->ubi, lnum, buf, offs, len);
463 if (err) {
464 ubifs_err("cannot read node type %d from LEB %d:%d, error %d",
465 type, lnum, offs, err);
466 return err;
467 }
468
469 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
470 return 0;
471
472 if (ch->node_type != type)
473 return 0;
474
475 node_len = le32_to_cpu(ch->len);
476 if (node_len != len)
477 return 0;
478
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +0200479 if (type == UBIFS_DATA_NODE && !c->always_chk_crc && c->no_chk_data_crc)
480 return 1;
Adrian Hunter2953e732008-09-04 16:26:00 +0300481
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300482 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8);
483 node_crc = le32_to_cpu(ch->crc);
484 if (crc != node_crc)
485 return 0;
486
487 return 1;
488}
489
490/**
491 * fallible_read_node - try to read a leaf node.
492 * @c: UBIFS file-system description object
493 * @key: key of node to read
494 * @zbr: position of node
495 * @node: node returned
496 *
497 * This function tries to read a node and returns %1 if the node is read, %0
498 * if the node is not present, and a negative error code in the case of error.
499 */
500static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key,
501 struct ubifs_zbranch *zbr, void *node)
502{
503 int ret;
504
505 dbg_tnc("LEB %d:%d, key %s", zbr->lnum, zbr->offs, DBGKEY(key));
506
507 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum,
508 zbr->offs);
509 if (ret == 1) {
510 union ubifs_key node_key;
511 struct ubifs_dent_node *dent = node;
512
513 /* All nodes have key in the same place */
514 key_read(c, &dent->key, &node_key);
515 if (keys_cmp(c, key, &node_key) != 0)
516 ret = 0;
517 }
Adrian Hunter601c0bc2008-08-22 14:23:35 +0300518 if (ret == 0 && c->replaying)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300519 dbg_mnt("dangling branch LEB %d:%d len %d, key %s",
520 zbr->lnum, zbr->offs, zbr->len, DBGKEY(key));
521 return ret;
522}
523
524/**
525 * matches_name - determine if a direntry or xattr entry matches a given name.
526 * @c: UBIFS file-system description object
527 * @zbr: zbranch of dent
528 * @nm: name to match
529 *
530 * This function checks if xentry/direntry referred by zbranch @zbr matches name
531 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by
532 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case
533 * of failure, a negative error code is returned.
534 */
535static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr,
536 const struct qstr *nm)
537{
538 struct ubifs_dent_node *dent;
539 int nlen, err;
540
541 /* If possible, match against the dent in the leaf node cache */
542 if (!zbr->leaf) {
543 dent = kmalloc(zbr->len, GFP_NOFS);
544 if (!dent)
545 return -ENOMEM;
546
547 err = ubifs_tnc_read_node(c, zbr, dent);
548 if (err)
549 goto out_free;
550
551 /* Add the node to the leaf node cache */
552 err = lnc_add_directly(c, zbr, dent);
553 if (err)
554 goto out_free;
555 } else
556 dent = zbr->leaf;
557
558 nlen = le16_to_cpu(dent->nlen);
559 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
560 if (err == 0) {
561 if (nlen == nm->len)
562 return NAME_MATCHES;
563 else if (nlen < nm->len)
564 return NAME_LESS;
565 else
566 return NAME_GREATER;
567 } else if (err < 0)
568 return NAME_LESS;
569 else
570 return NAME_GREATER;
571
572out_free:
573 kfree(dent);
574 return err;
575}
576
577/**
578 * get_znode - get a TNC znode that may not be loaded yet.
579 * @c: UBIFS file-system description object
580 * @znode: parent znode
581 * @n: znode branch slot number
582 *
583 * This function returns the znode or a negative error code.
584 */
585static struct ubifs_znode *get_znode(struct ubifs_info *c,
586 struct ubifs_znode *znode, int n)
587{
588 struct ubifs_zbranch *zbr;
589
590 zbr = &znode->zbranch[n];
591 if (zbr->znode)
592 znode = zbr->znode;
593 else
594 znode = ubifs_load_znode(c, zbr, znode, n);
595 return znode;
596}
597
598/**
599 * tnc_next - find next TNC entry.
600 * @c: UBIFS file-system description object
601 * @zn: znode is passed and returned here
602 * @n: znode branch slot number is passed and returned here
603 *
604 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is
605 * no next entry, or a negative error code otherwise.
606 */
607static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
608{
609 struct ubifs_znode *znode = *zn;
610 int nn = *n;
611
612 nn += 1;
613 if (nn < znode->child_cnt) {
614 *n = nn;
615 return 0;
616 }
617 while (1) {
618 struct ubifs_znode *zp;
619
620 zp = znode->parent;
621 if (!zp)
622 return -ENOENT;
623 nn = znode->iip + 1;
624 znode = zp;
625 if (nn < znode->child_cnt) {
626 znode = get_znode(c, znode, nn);
627 if (IS_ERR(znode))
628 return PTR_ERR(znode);
629 while (znode->level != 0) {
630 znode = get_znode(c, znode, 0);
631 if (IS_ERR(znode))
632 return PTR_ERR(znode);
633 }
634 nn = 0;
635 break;
636 }
637 }
638 *zn = znode;
639 *n = nn;
640 return 0;
641}
642
643/**
644 * tnc_prev - find previous TNC entry.
645 * @c: UBIFS file-system description object
646 * @zn: znode is returned here
647 * @n: znode branch slot number is passed and returned here
648 *
649 * This function returns %0 if the previous TNC entry is found, %-ENOENT if
650 * there is no next entry, or a negative error code otherwise.
651 */
652static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n)
653{
654 struct ubifs_znode *znode = *zn;
655 int nn = *n;
656
657 if (nn > 0) {
658 *n = nn - 1;
659 return 0;
660 }
661 while (1) {
662 struct ubifs_znode *zp;
663
664 zp = znode->parent;
665 if (!zp)
666 return -ENOENT;
667 nn = znode->iip - 1;
668 znode = zp;
669 if (nn >= 0) {
670 znode = get_znode(c, znode, nn);
671 if (IS_ERR(znode))
672 return PTR_ERR(znode);
673 while (znode->level != 0) {
674 nn = znode->child_cnt - 1;
675 znode = get_znode(c, znode, nn);
676 if (IS_ERR(znode))
677 return PTR_ERR(znode);
678 }
679 nn = znode->child_cnt - 1;
680 break;
681 }
682 }
683 *zn = znode;
684 *n = nn;
685 return 0;
686}
687
688/**
689 * resolve_collision - resolve a collision.
690 * @c: UBIFS file-system description object
691 * @key: key of a directory or extended attribute entry
692 * @zn: znode is returned here
693 * @n: zbranch number is passed and returned here
694 * @nm: name of the entry
695 *
696 * This function is called for "hashed" keys to make sure that the found key
697 * really corresponds to the looked up node (directory or extended attribute
698 * entry). It returns %1 and sets @zn and @n if the collision is resolved.
699 * %0 is returned if @nm is not found and @zn and @n are set to the previous
700 * entry, i.e. to the entry after which @nm could follow if it were in TNC.
701 * This means that @n may be set to %-1 if the leftmost key in @zn is the
702 * previous one. A negative error code is returned on failures.
703 */
704static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key,
705 struct ubifs_znode **zn, int *n,
706 const struct qstr *nm)
707{
708 int err;
709
710 err = matches_name(c, &(*zn)->zbranch[*n], nm);
711 if (unlikely(err < 0))
712 return err;
713 if (err == NAME_MATCHES)
714 return 1;
715
716 if (err == NAME_GREATER) {
717 /* Look left */
718 while (1) {
719 err = tnc_prev(c, zn, n);
720 if (err == -ENOENT) {
721 ubifs_assert(*n == 0);
722 *n = -1;
723 return 0;
724 }
725 if (err < 0)
726 return err;
727 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
728 /*
729 * We have found the branch after which we would
730 * like to insert, but inserting in this znode
731 * may still be wrong. Consider the following 3
732 * znodes, in the case where we are resolving a
733 * collision with Key2.
734 *
735 * znode zp
736 * ----------------------
737 * level 1 | Key0 | Key1 |
738 * -----------------------
739 * | |
740 * znode za | | znode zb
741 * ------------ ------------
742 * level 0 | Key0 | | Key2 |
743 * ------------ ------------
744 *
745 * The lookup finds Key2 in znode zb. Lets say
746 * there is no match and the name is greater so
747 * we look left. When we find Key0, we end up
748 * here. If we return now, we will insert into
749 * znode za at slot n = 1. But that is invalid
750 * according to the parent's keys. Key2 must
751 * be inserted into znode zb.
752 *
753 * Note, this problem is not relevant for the
754 * case when we go right, because
755 * 'tnc_insert()' would correct the parent key.
756 */
757 if (*n == (*zn)->child_cnt - 1) {
758 err = tnc_next(c, zn, n);
759 if (err) {
760 /* Should be impossible */
761 ubifs_assert(0);
762 if (err == -ENOENT)
763 err = -EINVAL;
764 return err;
765 }
766 ubifs_assert(*n == 0);
767 *n = -1;
768 }
769 return 0;
770 }
771 err = matches_name(c, &(*zn)->zbranch[*n], nm);
772 if (err < 0)
773 return err;
774 if (err == NAME_LESS)
775 return 0;
776 if (err == NAME_MATCHES)
777 return 1;
778 ubifs_assert(err == NAME_GREATER);
779 }
780 } else {
781 int nn = *n;
782 struct ubifs_znode *znode = *zn;
783
784 /* Look right */
785 while (1) {
786 err = tnc_next(c, &znode, &nn);
787 if (err == -ENOENT)
788 return 0;
789 if (err < 0)
790 return err;
791 if (keys_cmp(c, &znode->zbranch[nn].key, key))
792 return 0;
793 err = matches_name(c, &znode->zbranch[nn], nm);
794 if (err < 0)
795 return err;
796 if (err == NAME_GREATER)
797 return 0;
798 *zn = znode;
799 *n = nn;
800 if (err == NAME_MATCHES)
801 return 1;
802 ubifs_assert(err == NAME_LESS);
803 }
804 }
805}
806
807/**
808 * fallible_matches_name - determine if a dent matches a given name.
809 * @c: UBIFS file-system description object
810 * @zbr: zbranch of dent
811 * @nm: name to match
812 *
813 * This is a "fallible" version of 'matches_name()' function which does not
814 * panic if the direntry/xentry referred by @zbr does not exist on the media.
815 *
816 * This function checks if xentry/direntry referred by zbranch @zbr matches name
817 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr
818 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA
819 * if xentry/direntry referred by @zbr does not exist on the media. A negative
820 * error code is returned in case of failure.
821 */
822static int fallible_matches_name(struct ubifs_info *c,
823 struct ubifs_zbranch *zbr,
824 const struct qstr *nm)
825{
826 struct ubifs_dent_node *dent;
827 int nlen, err;
828
829 /* If possible, match against the dent in the leaf node cache */
830 if (!zbr->leaf) {
831 dent = kmalloc(zbr->len, GFP_NOFS);
832 if (!dent)
833 return -ENOMEM;
834
835 err = fallible_read_node(c, &zbr->key, zbr, dent);
836 if (err < 0)
837 goto out_free;
838 if (err == 0) {
839 /* The node was not present */
840 err = NOT_ON_MEDIA;
841 goto out_free;
842 }
843 ubifs_assert(err == 1);
844
845 err = lnc_add_directly(c, zbr, dent);
846 if (err)
847 goto out_free;
848 } else
849 dent = zbr->leaf;
850
851 nlen = le16_to_cpu(dent->nlen);
852 err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len));
853 if (err == 0) {
854 if (nlen == nm->len)
855 return NAME_MATCHES;
856 else if (nlen < nm->len)
857 return NAME_LESS;
858 else
859 return NAME_GREATER;
860 } else if (err < 0)
861 return NAME_LESS;
862 else
863 return NAME_GREATER;
864
865out_free:
866 kfree(dent);
867 return err;
868}
869
870/**
871 * fallible_resolve_collision - resolve a collision even if nodes are missing.
872 * @c: UBIFS file-system description object
873 * @key: key
874 * @zn: znode is returned here
875 * @n: branch number is passed and returned here
876 * @nm: name of directory entry
877 * @adding: indicates caller is adding a key to the TNC
878 *
879 * This is a "fallible" version of the 'resolve_collision()' function which
880 * does not panic if one of the nodes referred to by TNC does not exist on the
881 * media. This may happen when replaying the journal if a deleted node was
882 * Garbage-collected and the commit was not done. A branch that refers to a node
883 * that is not present is called a dangling branch. The following are the return
884 * codes for this function:
885 * o if @nm was found, %1 is returned and @zn and @n are set to the found
886 * branch;
887 * o if we are @adding and @nm was not found, %0 is returned;
888 * o if we are not @adding and @nm was not found, but a dangling branch was
889 * found, then %1 is returned and @zn and @n are set to the dangling branch;
890 * o a negative error code is returned in case of failure.
891 */
892static int fallible_resolve_collision(struct ubifs_info *c,
893 const union ubifs_key *key,
894 struct ubifs_znode **zn, int *n,
895 const struct qstr *nm, int adding)
896{
897 struct ubifs_znode *o_znode = NULL, *znode = *zn;
898 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n;
899
900 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm);
901 if (unlikely(cmp < 0))
902 return cmp;
903 if (cmp == NAME_MATCHES)
904 return 1;
905 if (cmp == NOT_ON_MEDIA) {
906 o_znode = znode;
907 o_n = nn;
908 /*
909 * We are unlucky and hit a dangling branch straight away.
910 * Now we do not really know where to go to find the needed
911 * branch - to the left or to the right. Well, let's try left.
912 */
913 unsure = 1;
914 } else if (!adding)
915 unsure = 1; /* Remove a dangling branch wherever it is */
916
917 if (cmp == NAME_GREATER || unsure) {
918 /* Look left */
919 while (1) {
920 err = tnc_prev(c, zn, n);
921 if (err == -ENOENT) {
922 ubifs_assert(*n == 0);
923 *n = -1;
924 break;
925 }
926 if (err < 0)
927 return err;
928 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) {
929 /* See comments in 'resolve_collision()' */
930 if (*n == (*zn)->child_cnt - 1) {
931 err = tnc_next(c, zn, n);
932 if (err) {
933 /* Should be impossible */
934 ubifs_assert(0);
935 if (err == -ENOENT)
936 err = -EINVAL;
937 return err;
938 }
939 ubifs_assert(*n == 0);
940 *n = -1;
941 }
942 break;
943 }
944 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm);
945 if (err < 0)
946 return err;
947 if (err == NAME_MATCHES)
948 return 1;
949 if (err == NOT_ON_MEDIA) {
950 o_znode = *zn;
951 o_n = *n;
952 continue;
953 }
954 if (!adding)
955 continue;
956 if (err == NAME_LESS)
957 break;
958 else
959 unsure = 0;
960 }
961 }
962
963 if (cmp == NAME_LESS || unsure) {
964 /* Look right */
965 *zn = znode;
966 *n = nn;
967 while (1) {
968 err = tnc_next(c, &znode, &nn);
969 if (err == -ENOENT)
970 break;
971 if (err < 0)
972 return err;
973 if (keys_cmp(c, &znode->zbranch[nn].key, key))
974 break;
975 err = fallible_matches_name(c, &znode->zbranch[nn], nm);
976 if (err < 0)
977 return err;
978 if (err == NAME_GREATER)
979 break;
980 *zn = znode;
981 *n = nn;
982 if (err == NAME_MATCHES)
983 return 1;
984 if (err == NOT_ON_MEDIA) {
985 o_znode = znode;
986 o_n = nn;
987 }
988 }
989 }
990
991 /* Never match a dangling branch when adding */
992 if (adding || !o_znode)
993 return 0;
994
995 dbg_mnt("dangling match LEB %d:%d len %d %s",
996 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs,
997 o_znode->zbranch[o_n].len, DBGKEY(key));
998 *zn = o_znode;
999 *n = o_n;
1000 return 1;
1001}
1002
1003/**
1004 * matches_position - determine if a zbranch matches a given position.
1005 * @zbr: zbranch of dent
1006 * @lnum: LEB number of dent to match
1007 * @offs: offset of dent to match
1008 *
1009 * This function returns %1 if @lnum:@offs matches, and %0 otherwise.
1010 */
1011static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs)
1012{
1013 if (zbr->lnum == lnum && zbr->offs == offs)
1014 return 1;
1015 else
1016 return 0;
1017}
1018
1019/**
1020 * resolve_collision_directly - resolve a collision directly.
1021 * @c: UBIFS file-system description object
1022 * @key: key of directory entry
1023 * @zn: znode is passed and returned here
1024 * @n: zbranch number is passed and returned here
1025 * @lnum: LEB number of dent node to match
1026 * @offs: offset of dent node to match
1027 *
1028 * This function is used for "hashed" keys to make sure the found directory or
1029 * extended attribute entry node is what was looked for. It is used when the
1030 * flash address of the right node is known (@lnum:@offs) which makes it much
1031 * easier to resolve collisions (no need to read entries and match full
1032 * names). This function returns %1 and sets @zn and @n if the collision is
1033 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the
1034 * previous directory entry. Otherwise a negative error code is returned.
1035 */
1036static int resolve_collision_directly(struct ubifs_info *c,
1037 const union ubifs_key *key,
1038 struct ubifs_znode **zn, int *n,
1039 int lnum, int offs)
1040{
1041 struct ubifs_znode *znode;
1042 int nn, err;
1043
1044 znode = *zn;
1045 nn = *n;
1046 if (matches_position(&znode->zbranch[nn], lnum, offs))
1047 return 1;
1048
1049 /* Look left */
1050 while (1) {
1051 err = tnc_prev(c, &znode, &nn);
1052 if (err == -ENOENT)
1053 break;
1054 if (err < 0)
1055 return err;
1056 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1057 break;
1058 if (matches_position(&znode->zbranch[nn], lnum, offs)) {
1059 *zn = znode;
1060 *n = nn;
1061 return 1;
1062 }
1063 }
1064
1065 /* Look right */
1066 znode = *zn;
1067 nn = *n;
1068 while (1) {
1069 err = tnc_next(c, &znode, &nn);
1070 if (err == -ENOENT)
1071 return 0;
1072 if (err < 0)
1073 return err;
1074 if (keys_cmp(c, &znode->zbranch[nn].key, key))
1075 return 0;
1076 *zn = znode;
1077 *n = nn;
1078 if (matches_position(&znode->zbranch[nn], lnum, offs))
1079 return 1;
1080 }
1081}
1082
1083/**
1084 * dirty_cow_bottom_up - dirty a znode and its ancestors.
1085 * @c: UBIFS file-system description object
1086 * @znode: znode to dirty
1087 *
1088 * If we do not have a unique key that resides in a znode, then we cannot
1089 * dirty that znode from the top down (i.e. by using lookup_level0_dirty)
1090 * This function records the path back to the last dirty ancestor, and then
1091 * dirties the znodes on that path.
1092 */
1093static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c,
1094 struct ubifs_znode *znode)
1095{
1096 struct ubifs_znode *zp;
1097 int *path = c->bottom_up_buf, p = 0;
1098
1099 ubifs_assert(c->zroot.znode);
1100 ubifs_assert(znode);
1101 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) {
1102 kfree(c->bottom_up_buf);
1103 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int),
1104 GFP_NOFS);
1105 if (!c->bottom_up_buf)
1106 return ERR_PTR(-ENOMEM);
1107 path = c->bottom_up_buf;
1108 }
1109 if (c->zroot.znode->level) {
1110 /* Go up until parent is dirty */
1111 while (1) {
1112 int n;
1113
1114 zp = znode->parent;
1115 if (!zp)
1116 break;
1117 n = znode->iip;
1118 ubifs_assert(p < c->zroot.znode->level);
1119 path[p++] = n;
1120 if (!zp->cnext && ubifs_zn_dirty(znode))
1121 break;
1122 znode = zp;
1123 }
1124 }
1125
1126 /* Come back down, dirtying as we go */
1127 while (1) {
1128 struct ubifs_zbranch *zbr;
1129
1130 zp = znode->parent;
1131 if (zp) {
1132 ubifs_assert(path[p - 1] >= 0);
1133 ubifs_assert(path[p - 1] < zp->child_cnt);
1134 zbr = &zp->zbranch[path[--p]];
1135 znode = dirty_cow_znode(c, zbr);
1136 } else {
1137 ubifs_assert(znode == c->zroot.znode);
1138 znode = dirty_cow_znode(c, &c->zroot);
1139 }
Hirofumi Nakagawa8d47aef2008-08-21 17:16:40 +03001140 if (IS_ERR(znode) || !p)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001141 break;
1142 ubifs_assert(path[p - 1] >= 0);
1143 ubifs_assert(path[p - 1] < znode->child_cnt);
1144 znode = znode->zbranch[path[p - 1]].znode;
1145 }
1146
1147 return znode;
1148}
1149
1150/**
1151 * ubifs_lookup_level0 - search for zero-level znode.
1152 * @c: UBIFS file-system description object
1153 * @key: key to lookup
1154 * @zn: znode is returned here
1155 * @n: znode branch slot number is returned here
1156 *
1157 * This function looks up the TNC tree and search for zero-level znode which
1158 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1159 * cases:
1160 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1161 * is returned and slot number of the matched branch is stored in @n;
1162 * o not exact match, which means that zero-level znode does not contain
Artem Bityutskiye3c3efc2009-08-27 16:34:19 +03001163 * @key, then %0 is returned and slot number of the closest branch is stored
1164 * in @n;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001165 * o @key is so small that it is even less than the lowest key of the
1166 * leftmost zero-level node, then %0 is returned and %0 is stored in @n.
1167 *
1168 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1169 * function reads corresponding indexing nodes and inserts them to TNC. In
1170 * case of failure, a negative error code is returned.
1171 */
1172int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key,
1173 struct ubifs_znode **zn, int *n)
1174{
1175 int err, exact;
1176 struct ubifs_znode *znode;
1177 unsigned long time = get_seconds();
1178
1179 dbg_tnc("search key %s", DBGKEY(key));
1180
1181 znode = c->zroot.znode;
1182 if (unlikely(!znode)) {
1183 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1184 if (IS_ERR(znode))
1185 return PTR_ERR(znode);
1186 }
1187
1188 znode->time = time;
1189
1190 while (1) {
1191 struct ubifs_zbranch *zbr;
1192
1193 exact = ubifs_search_zbranch(c, znode, key, n);
1194
1195 if (znode->level == 0)
1196 break;
1197
1198 if (*n < 0)
1199 *n = 0;
1200 zbr = &znode->zbranch[*n];
1201
1202 if (zbr->znode) {
1203 znode->time = time;
1204 znode = zbr->znode;
1205 continue;
1206 }
1207
1208 /* znode is not in TNC cache, load it from the media */
1209 znode = ubifs_load_znode(c, zbr, znode, *n);
1210 if (IS_ERR(znode))
1211 return PTR_ERR(znode);
1212 }
1213
1214 *zn = znode;
1215 if (exact || !is_hash_key(c, key) || *n != -1) {
1216 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1217 return exact;
1218 }
1219
1220 /*
1221 * Here is a tricky place. We have not found the key and this is a
1222 * "hashed" key, which may collide. The rest of the code deals with
1223 * situations like this:
1224 *
1225 * | 3 | 5 |
1226 * / \
1227 * | 3 | 5 | | 6 | 7 | (x)
1228 *
1229 * Or more a complex example:
1230 *
1231 * | 1 | 5 |
1232 * / \
1233 * | 1 | 3 | | 5 | 8 |
1234 * \ /
1235 * | 5 | 5 | | 6 | 7 | (x)
1236 *
1237 * In the examples, if we are looking for key "5", we may reach nodes
1238 * marked with "(x)". In this case what we have do is to look at the
1239 * left and see if there is "5" key there. If there is, we have to
1240 * return it.
1241 *
1242 * Note, this whole situation is possible because we allow to have
1243 * elements which are equivalent to the next key in the parent in the
1244 * children of current znode. For example, this happens if we split a
1245 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something
1246 * like this:
1247 * | 3 | 5 |
1248 * / \
1249 * | 3 | 5 | | 5 | 6 | 7 |
1250 * ^
1251 * And this becomes what is at the first "picture" after key "5" marked
1252 * with "^" is removed. What could be done is we could prohibit
1253 * splitting in the middle of the colliding sequence. Also, when
1254 * removing the leftmost key, we would have to correct the key of the
1255 * parent node, which would introduce additional complications. Namely,
Artem Bityutskiy7d4e9cc2009-03-20 19:11:12 +02001256 * if we changed the leftmost key of the parent znode, the garbage
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001257 * collector would be unable to find it (GC is doing this when GC'ing
1258 * indexing LEBs). Although we already have an additional RB-tree where
1259 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until
1260 * after the commit. But anyway, this does not look easy to implement
1261 * so we did not try this.
1262 */
1263 err = tnc_prev(c, &znode, n);
1264 if (err == -ENOENT) {
1265 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1266 *n = -1;
1267 return 0;
1268 }
1269 if (unlikely(err < 0))
1270 return err;
1271 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1272 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1273 *n = -1;
1274 return 0;
1275 }
1276
1277 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1278 *zn = znode;
1279 return 1;
1280}
1281
1282/**
1283 * lookup_level0_dirty - search for zero-level znode dirtying.
1284 * @c: UBIFS file-system description object
1285 * @key: key to lookup
1286 * @zn: znode is returned here
1287 * @n: znode branch slot number is returned here
1288 *
1289 * This function looks up the TNC tree and search for zero-level znode which
1290 * refers key @key. The found zero-level znode is returned in @zn. There are 3
1291 * cases:
1292 * o exact match, i.e. the found zero-level znode contains key @key, then %1
1293 * is returned and slot number of the matched branch is stored in @n;
1294 * o not exact match, which means that zero-level znode does not contain @key
1295 * then %0 is returned and slot number of the closed branch is stored in
1296 * @n;
1297 * o @key is so small that it is even less than the lowest key of the
1298 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n.
1299 *
1300 * Additionally all znodes in the path from the root to the located zero-level
1301 * znode are marked as dirty.
1302 *
1303 * Note, when the TNC tree is traversed, some znodes may be absent, then this
1304 * function reads corresponding indexing nodes and inserts them to TNC. In
1305 * case of failure, a negative error code is returned.
1306 */
1307static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key,
1308 struct ubifs_znode **zn, int *n)
1309{
1310 int err, exact;
1311 struct ubifs_znode *znode;
1312 unsigned long time = get_seconds();
1313
1314 dbg_tnc("search and dirty key %s", DBGKEY(key));
1315
1316 znode = c->zroot.znode;
1317 if (unlikely(!znode)) {
1318 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
1319 if (IS_ERR(znode))
1320 return PTR_ERR(znode);
1321 }
1322
1323 znode = dirty_cow_znode(c, &c->zroot);
1324 if (IS_ERR(znode))
1325 return PTR_ERR(znode);
1326
1327 znode->time = time;
1328
1329 while (1) {
1330 struct ubifs_zbranch *zbr;
1331
1332 exact = ubifs_search_zbranch(c, znode, key, n);
1333
1334 if (znode->level == 0)
1335 break;
1336
1337 if (*n < 0)
1338 *n = 0;
1339 zbr = &znode->zbranch[*n];
1340
1341 if (zbr->znode) {
1342 znode->time = time;
1343 znode = dirty_cow_znode(c, zbr);
1344 if (IS_ERR(znode))
1345 return PTR_ERR(znode);
1346 continue;
1347 }
1348
1349 /* znode is not in TNC cache, load it from the media */
1350 znode = ubifs_load_znode(c, zbr, znode, *n);
1351 if (IS_ERR(znode))
1352 return PTR_ERR(znode);
1353 znode = dirty_cow_znode(c, zbr);
1354 if (IS_ERR(znode))
1355 return PTR_ERR(znode);
1356 }
1357
1358 *zn = znode;
1359 if (exact || !is_hash_key(c, key) || *n != -1) {
1360 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n);
1361 return exact;
1362 }
1363
1364 /*
1365 * See huge comment at 'lookup_level0_dirty()' what is the rest of the
1366 * code.
1367 */
1368 err = tnc_prev(c, &znode, n);
1369 if (err == -ENOENT) {
1370 *n = -1;
1371 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1372 return 0;
1373 }
1374 if (unlikely(err < 0))
1375 return err;
1376 if (keys_cmp(c, key, &znode->zbranch[*n].key)) {
1377 *n = -1;
1378 dbg_tnc("found 0, lvl %d, n -1", znode->level);
1379 return 0;
1380 }
1381
1382 if (znode->cnext || !ubifs_zn_dirty(znode)) {
1383 znode = dirty_cow_bottom_up(c, znode);
1384 if (IS_ERR(znode))
1385 return PTR_ERR(znode);
1386 }
1387
1388 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n);
1389 *zn = znode;
1390 return 1;
1391}
1392
1393/**
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001394 * maybe_leb_gced - determine if a LEB may have been garbage collected.
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001395 * @c: UBIFS file-system description object
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001396 * @lnum: LEB number
1397 * @gc_seq1: garbage collection sequence number
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001398 *
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001399 * This function determines if @lnum may have been garbage collected since
1400 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise
1401 * %0 is returned.
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001402 */
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001403static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1)
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001404{
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001405 int gc_seq2, gced_lnum;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001406
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001407 gced_lnum = c->gced_lnum;
1408 smp_rmb();
1409 gc_seq2 = c->gc_seq;
1410 /* Same seq means no GC */
1411 if (gc_seq1 == gc_seq2)
1412 return 0;
1413 /* Different by more than 1 means we don't know */
1414 if (gc_seq1 + 1 != gc_seq2)
1415 return 1;
1416 /*
1417 * We have seen the sequence number has increased by 1. Now we need to
1418 * be sure we read the right LEB number, so read it again.
1419 */
1420 smp_rmb();
1421 if (gced_lnum != c->gced_lnum)
1422 return 1;
1423 /* Finally we can check lnum */
1424 if (gced_lnum == lnum)
1425 return 1;
1426 return 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001427}
1428
1429/**
1430 * ubifs_tnc_locate - look up a file-system node and return it and its location.
1431 * @c: UBIFS file-system description object
1432 * @key: node key to lookup
1433 * @node: the node is returned here
1434 * @lnum: LEB number is returned here
1435 * @offs: offset is returned here
1436 *
Artem Bityutskiye3c3efc2009-08-27 16:34:19 +03001437 * This function looks up and reads node with key @key. The caller has to make
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001438 * sure the @node buffer is large enough to fit the node. Returns zero in case
1439 * of success, %-ENOENT if the node was not found, and a negative error code in
1440 * case of failure. The node location can be returned in @lnum and @offs.
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001441 */
1442int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key,
1443 void *node, int *lnum, int *offs)
1444{
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001445 int found, n, err, safely = 0, gc_seq1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001446 struct ubifs_znode *znode;
1447 struct ubifs_zbranch zbr, *zt;
1448
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001449again:
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001450 mutex_lock(&c->tnc_mutex);
1451 found = ubifs_lookup_level0(c, key, &znode, &n);
1452 if (!found) {
1453 err = -ENOENT;
1454 goto out;
1455 } else if (found < 0) {
1456 err = found;
1457 goto out;
1458 }
1459 zt = &znode->zbranch[n];
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001460 if (lnum) {
1461 *lnum = zt->lnum;
1462 *offs = zt->offs;
1463 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001464 if (is_hash_key(c, key)) {
1465 /*
1466 * In this case the leaf node cache gets used, so we pass the
1467 * address of the zbranch and keep the mutex locked
1468 */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001469 err = tnc_read_node_nm(c, zt, node);
1470 goto out;
1471 }
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001472 if (safely) {
1473 err = ubifs_tnc_read_node(c, zt, node);
1474 goto out;
1475 }
1476 /* Drop the TNC mutex prematurely and race with garbage collection */
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001477 zbr = znode->zbranch[n];
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001478 gc_seq1 = c->gc_seq;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001479 mutex_unlock(&c->tnc_mutex);
1480
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001481 if (ubifs_get_wbuf(c, zbr.lnum)) {
1482 /* We do not GC journal heads */
1483 err = ubifs_tnc_read_node(c, &zbr, node);
1484 return err;
1485 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001486
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001487 err = fallible_read_node(c, key, &zbr, node);
Adrian Hunter6dcfac42008-09-12 12:27:47 +03001488 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) {
Adrian Hunter601c0bc2008-08-22 14:23:35 +03001489 /*
1490 * The node may have been GC'ed out from under us so try again
1491 * while keeping the TNC mutex locked.
1492 */
1493 safely = 1;
1494 goto again;
1495 }
1496 return 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001497
1498out:
1499 mutex_unlock(&c->tnc_mutex);
1500 return err;
1501}
1502
1503/**
Adrian Hunter4793e7c2008-09-02 16:29:46 +03001504 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read.
1505 * @c: UBIFS file-system description object
1506 * @bu: bulk-read parameters and results
1507 *
1508 * Lookup consecutive data node keys for the same inode that reside
Artem Bityutskiy6c0c42c2008-11-18 20:20:05 +02001509 * consecutively in the same LEB. This function returns zero in case of success
1510 * and a negative error code in case of failure.
1511 *
1512 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function
1513 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares
Artem Bityutskiy6f7ab6d2009-01-27 16:12:31 +02001514 * maximum possible amount of nodes for bulk-read.
Adrian Hunter4793e7c2008-09-02 16:29:46 +03001515 */
1516int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu)
1517{
1518 int n, err = 0, lnum = -1, uninitialized_var(offs);
1519 int uninitialized_var(len);
1520 unsigned int block = key_block(c, &bu->key);
1521 struct ubifs_znode *znode;
1522
1523 bu->cnt = 0;
1524 bu->blk_cnt = 0;
1525 bu->eof = 0;
1526
1527 mutex_lock(&c->tnc_mutex);
1528 /* Find first key */
1529 err = ubifs_lookup_level0(c, &bu->key, &znode, &n);
1530 if (err < 0)
1531 goto out;
1532 if (err) {
1533 /* Key found */
1534 len = znode->zbranch[n].len;
1535 /* The buffer must be big enough for at least 1 node */
1536 if (len > bu->buf_len) {
1537 err = -EINVAL;
1538 goto out;
1539 }
1540 /* Add this key */
1541 bu->zbranch[bu->cnt++] = znode->zbranch[n];
1542 bu->blk_cnt += 1;
1543 lnum = znode->zbranch[n].lnum;
1544 offs = ALIGN(znode->zbranch[n].offs + len, 8);
1545 }
1546 while (1) {
1547 struct ubifs_zbranch *zbr;
1548 union ubifs_key *key;
1549 unsigned int next_block;
1550
1551 /* Find next key */
1552 err = tnc_next(c, &znode, &n);
1553 if (err)
1554 goto out;
1555 zbr = &znode->zbranch[n];
1556 key = &zbr->key;
1557 /* See if there is another data key for this file */
1558 if (key_inum(c, key) != key_inum(c, &bu->key) ||
1559 key_type(c, key) != UBIFS_DATA_KEY) {
1560 err = -ENOENT;
1561 goto out;
1562 }
1563 if (lnum < 0) {
1564 /* First key found */
1565 lnum = zbr->lnum;
1566 offs = ALIGN(zbr->offs + zbr->len, 8);
1567 len = zbr->len;
1568 if (len > bu->buf_len) {
1569 err = -EINVAL;
1570 goto out;
1571 }
1572 } else {
1573 /*
1574 * The data nodes must be in consecutive positions in
1575 * the same LEB.
1576 */
1577 if (zbr->lnum != lnum || zbr->offs != offs)
1578 goto out;
1579 offs += ALIGN(zbr->len, 8);
1580 len = ALIGN(len, 8) + zbr->len;
1581 /* Must not exceed buffer length */
1582 if (len > bu->buf_len)
1583 goto out;
1584 }
1585 /* Allow for holes */
1586 next_block = key_block(c, key);
1587 bu->blk_cnt += (next_block - block - 1);
1588 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1589 goto out;
1590 block = next_block;
1591 /* Add this key */
1592 bu->zbranch[bu->cnt++] = *zbr;
1593 bu->blk_cnt += 1;
1594 /* See if we have room for more */
1595 if (bu->cnt >= UBIFS_MAX_BULK_READ)
1596 goto out;
1597 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ)
1598 goto out;
1599 }
1600out:
1601 if (err == -ENOENT) {
1602 bu->eof = 1;
1603 err = 0;
1604 }
1605 bu->gc_seq = c->gc_seq;
1606 mutex_unlock(&c->tnc_mutex);
1607 if (err)
1608 return err;
1609 /*
1610 * An enormous hole could cause bulk-read to encompass too many
1611 * page cache pages, so limit the number here.
1612 */
Adrian Hunter63c300b2008-09-17 12:11:13 +03001613 if (bu->blk_cnt > UBIFS_MAX_BULK_READ)
Adrian Hunter4793e7c2008-09-02 16:29:46 +03001614 bu->blk_cnt = UBIFS_MAX_BULK_READ;
1615 /*
1616 * Ensure that bulk-read covers a whole number of page cache
1617 * pages.
1618 */
1619 if (UBIFS_BLOCKS_PER_PAGE == 1 ||
1620 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1)))
1621 return 0;
1622 if (bu->eof) {
1623 /* At the end of file we can round up */
1624 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1;
1625 return 0;
1626 }
1627 /* Exclude data nodes that do not make up a whole page cache page */
1628 block = key_block(c, &bu->key) + bu->blk_cnt;
1629 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1);
1630 while (bu->cnt) {
1631 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block)
1632 break;
1633 bu->cnt -= 1;
1634 }
1635 return 0;
1636}
1637
1638/**
1639 * read_wbuf - bulk-read from a LEB with a wbuf.
1640 * @wbuf: wbuf that may overlap the read
1641 * @buf: buffer into which to read
1642 * @len: read length
1643 * @lnum: LEB number from which to read
1644 * @offs: offset from which to read
1645 *
1646 * This functions returns %0 on success or a negative error code on failure.
1647 */
1648static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum,
1649 int offs)
1650{
1651 const struct ubifs_info *c = wbuf->c;
1652 int rlen, overlap;
1653
1654 dbg_io("LEB %d:%d, length %d", lnum, offs, len);
1655 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0);
1656 ubifs_assert(!(offs & 7) && offs < c->leb_size);
1657 ubifs_assert(offs + len <= c->leb_size);
1658
1659 spin_lock(&wbuf->lock);
1660 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs);
1661 if (!overlap) {
1662 /* We may safely unlock the write-buffer and read the data */
1663 spin_unlock(&wbuf->lock);
1664 return ubi_read(c->ubi, lnum, buf, offs, len);
1665 }
1666
1667 /* Don't read under wbuf */
1668 rlen = wbuf->offs - offs;
1669 if (rlen < 0)
1670 rlen = 0;
1671
1672 /* Copy the rest from the write-buffer */
1673 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen);
1674 spin_unlock(&wbuf->lock);
1675
1676 if (rlen > 0)
1677 /* Read everything that goes before write-buffer */
1678 return ubi_read(c->ubi, lnum, buf, offs, rlen);
1679
1680 return 0;
1681}
1682
1683/**
1684 * validate_data_node - validate data nodes for bulk-read.
1685 * @c: UBIFS file-system description object
1686 * @buf: buffer containing data node to validate
1687 * @zbr: zbranch of data node to validate
1688 *
1689 * This functions returns %0 on success or a negative error code on failure.
1690 */
1691static int validate_data_node(struct ubifs_info *c, void *buf,
1692 struct ubifs_zbranch *zbr)
1693{
1694 union ubifs_key key1;
1695 struct ubifs_ch *ch = buf;
1696 int err, len;
1697
1698 if (ch->node_type != UBIFS_DATA_NODE) {
1699 ubifs_err("bad node type (%d but expected %d)",
1700 ch->node_type, UBIFS_DATA_NODE);
1701 goto out_err;
1702 }
1703
Adrian Hunter2953e732008-09-04 16:26:00 +03001704 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0);
Adrian Hunter4793e7c2008-09-02 16:29:46 +03001705 if (err) {
1706 ubifs_err("expected node type %d", UBIFS_DATA_NODE);
1707 goto out;
1708 }
1709
1710 len = le32_to_cpu(ch->len);
1711 if (len != zbr->len) {
1712 ubifs_err("bad node length %d, expected %d", len, zbr->len);
1713 goto out_err;
1714 }
1715
1716 /* Make sure the key of the read node is correct */
1717 key_read(c, buf + UBIFS_KEY_OFFSET, &key1);
1718 if (!keys_eq(c, &zbr->key, &key1)) {
1719 ubifs_err("bad key in node at LEB %d:%d",
1720 zbr->lnum, zbr->offs);
1721 dbg_tnc("looked for key %s found node's key %s",
1722 DBGKEY(&zbr->key), DBGKEY1(&key1));
1723 goto out_err;
1724 }
1725
1726 return 0;
1727
1728out_err:
1729 err = -EINVAL;
1730out:
1731 ubifs_err("bad node at LEB %d:%d", zbr->lnum, zbr->offs);
1732 dbg_dump_node(c, buf);
1733 dbg_dump_stack();
1734 return err;
1735}
1736
1737/**
1738 * ubifs_tnc_bulk_read - read a number of data nodes in one go.
1739 * @c: UBIFS file-system description object
1740 * @bu: bulk-read parameters and results
1741 *
1742 * This functions reads and validates the data nodes that were identified by the
1743 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success,
1744 * -EAGAIN to indicate a race with GC, or another negative error code on
1745 * failure.
1746 */
1747int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu)
1748{
1749 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i;
1750 struct ubifs_wbuf *wbuf;
1751 void *buf;
1752
1753 len = bu->zbranch[bu->cnt - 1].offs;
1754 len += bu->zbranch[bu->cnt - 1].len - offs;
1755 if (len > bu->buf_len) {
1756 ubifs_err("buffer too small %d vs %d", bu->buf_len, len);
1757 return -EINVAL;
1758 }
1759
1760 /* Do the read */
1761 wbuf = ubifs_get_wbuf(c, lnum);
1762 if (wbuf)
1763 err = read_wbuf(wbuf, bu->buf, len, lnum, offs);
1764 else
1765 err = ubi_read(c->ubi, lnum, bu->buf, offs, len);
1766
1767 /* Check for a race with GC */
1768 if (maybe_leb_gced(c, lnum, bu->gc_seq))
1769 return -EAGAIN;
1770
1771 if (err && err != -EBADMSG) {
1772 ubifs_err("failed to read from LEB %d:%d, error %d",
1773 lnum, offs, err);
1774 dbg_dump_stack();
1775 dbg_tnc("key %s", DBGKEY(&bu->key));
1776 return err;
1777 }
1778
1779 /* Validate the nodes read */
1780 buf = bu->buf;
1781 for (i = 0; i < bu->cnt; i++) {
1782 err = validate_data_node(c, buf, &bu->zbranch[i]);
1783 if (err)
1784 return err;
1785 buf = buf + ALIGN(bu->zbranch[i].len, 8);
1786 }
1787
1788 return 0;
1789}
1790
1791/**
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001792 * do_lookup_nm- look up a "hashed" node.
1793 * @c: UBIFS file-system description object
1794 * @key: node key to lookup
1795 * @node: the node is returned here
1796 * @nm: node name
1797 *
1798 * This function look up and reads a node which contains name hash in the key.
1799 * Since the hash may have collisions, there may be many nodes with the same
1800 * key, so we have to sequentially look to all of them until the needed one is
1801 * found. This function returns zero in case of success, %-ENOENT if the node
1802 * was not found, and a negative error code in case of failure.
1803 */
1804static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1805 void *node, const struct qstr *nm)
1806{
1807 int found, n, err;
1808 struct ubifs_znode *znode;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001809
1810 dbg_tnc("name '%.*s' key %s", nm->len, nm->name, DBGKEY(key));
1811 mutex_lock(&c->tnc_mutex);
1812 found = ubifs_lookup_level0(c, key, &znode, &n);
1813 if (!found) {
1814 err = -ENOENT;
1815 goto out_unlock;
1816 } else if (found < 0) {
1817 err = found;
1818 goto out_unlock;
1819 }
1820
1821 ubifs_assert(n >= 0);
1822
1823 err = resolve_collision(c, key, &znode, &n, nm);
1824 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
1825 if (unlikely(err < 0))
1826 goto out_unlock;
1827 if (err == 0) {
1828 err = -ENOENT;
1829 goto out_unlock;
1830 }
1831
Adrian Hunter761e29f2008-08-20 16:32:40 +03001832 err = tnc_read_node_nm(c, &znode->zbranch[n], node);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001833
1834out_unlock:
1835 mutex_unlock(&c->tnc_mutex);
1836 return err;
1837}
1838
1839/**
1840 * ubifs_tnc_lookup_nm - look up a "hashed" node.
1841 * @c: UBIFS file-system description object
1842 * @key: node key to lookup
1843 * @node: the node is returned here
1844 * @nm: node name
1845 *
1846 * This function look up and reads a node which contains name hash in the key.
1847 * Since the hash may have collisions, there may be many nodes with the same
1848 * key, so we have to sequentially look to all of them until the needed one is
1849 * found. This function returns zero in case of success, %-ENOENT if the node
1850 * was not found, and a negative error code in case of failure.
1851 */
1852int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key,
1853 void *node, const struct qstr *nm)
1854{
1855 int err, len;
1856 const struct ubifs_dent_node *dent = node;
1857
1858 /*
1859 * We assume that in most of the cases there are no name collisions and
1860 * 'ubifs_tnc_lookup()' returns us the right direntry.
1861 */
1862 err = ubifs_tnc_lookup(c, key, node);
1863 if (err)
1864 return err;
1865
1866 len = le16_to_cpu(dent->nlen);
1867 if (nm->len == len && !memcmp(dent->name, nm->name, len))
1868 return 0;
1869
1870 /*
1871 * Unluckily, there are hash collisions and we have to iterate over
1872 * them look at each direntry with colliding name hash sequentially.
1873 */
1874 return do_lookup_nm(c, key, node, nm);
1875}
1876
1877/**
1878 * correct_parent_keys - correct parent znodes' keys.
1879 * @c: UBIFS file-system description object
1880 * @znode: znode to correct parent znodes for
1881 *
1882 * This is a helper function for 'tnc_insert()'. When the key of the leftmost
1883 * zbranch changes, keys of parent znodes have to be corrected. This helper
1884 * function is called in such situations and corrects the keys if needed.
1885 */
1886static void correct_parent_keys(const struct ubifs_info *c,
1887 struct ubifs_znode *znode)
1888{
1889 union ubifs_key *key, *key1;
1890
1891 ubifs_assert(znode->parent);
1892 ubifs_assert(znode->iip == 0);
1893
1894 key = &znode->zbranch[0].key;
1895 key1 = &znode->parent->zbranch[0].key;
1896
1897 while (keys_cmp(c, key, key1) < 0) {
1898 key_copy(c, key, key1);
1899 znode = znode->parent;
1900 znode->alt = 1;
1901 if (!znode->parent || znode->iip)
1902 break;
1903 key1 = &znode->parent->zbranch[0].key;
1904 }
1905}
1906
1907/**
1908 * insert_zbranch - insert a zbranch into a znode.
1909 * @znode: znode into which to insert
1910 * @zbr: zbranch to insert
1911 * @n: slot number to insert to
1912 *
1913 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in
1914 * znode's array of zbranches and keeps zbranches consolidated, so when a new
1915 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th
1916 * slot, zbranches starting from @n have to be moved right.
1917 */
1918static void insert_zbranch(struct ubifs_znode *znode,
1919 const struct ubifs_zbranch *zbr, int n)
1920{
1921 int i;
1922
1923 ubifs_assert(ubifs_zn_dirty(znode));
1924
1925 if (znode->level) {
1926 for (i = znode->child_cnt; i > n; i--) {
1927 znode->zbranch[i] = znode->zbranch[i - 1];
1928 if (znode->zbranch[i].znode)
1929 znode->zbranch[i].znode->iip = i;
1930 }
1931 if (zbr->znode)
1932 zbr->znode->iip = n;
1933 } else
1934 for (i = znode->child_cnt; i > n; i--)
1935 znode->zbranch[i] = znode->zbranch[i - 1];
1936
1937 znode->zbranch[n] = *zbr;
1938 znode->child_cnt += 1;
1939
1940 /*
1941 * After inserting at slot zero, the lower bound of the key range of
1942 * this znode may have changed. If this znode is subsequently split
1943 * then the upper bound of the key range may change, and furthermore
1944 * it could change to be lower than the original lower bound. If that
1945 * happens, then it will no longer be possible to find this znode in the
1946 * TNC using the key from the index node on flash. That is bad because
1947 * if it is not found, we will assume it is obsolete and may overwrite
1948 * it. Then if there is an unclean unmount, we will start using the
1949 * old index which will be broken.
1950 *
1951 * So we first mark znodes that have insertions at slot zero, and then
1952 * if they are split we add their lnum/offs to the old_idx tree.
1953 */
1954 if (n == 0)
1955 znode->alt = 1;
1956}
1957
1958/**
1959 * tnc_insert - insert a node into TNC.
1960 * @c: UBIFS file-system description object
1961 * @znode: znode to insert into
1962 * @zbr: branch to insert
1963 * @n: slot number to insert new zbranch to
1964 *
1965 * This function inserts a new node described by @zbr into znode @znode. If
1966 * znode does not have a free slot for new zbranch, it is split. Parent znodes
1967 * are splat as well if needed. Returns zero in case of success or a negative
1968 * error code in case of failure.
1969 */
1970static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode,
1971 struct ubifs_zbranch *zbr, int n)
1972{
1973 struct ubifs_znode *zn, *zi, *zp;
1974 int i, keep, move, appending = 0;
Adrian Hunter2242c682008-09-05 11:56:05 +03001975 union ubifs_key *key = &zbr->key, *key1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001976
1977 ubifs_assert(n >= 0 && n <= c->fanout);
1978
1979 /* Implement naive insert for now */
1980again:
1981 zp = znode->parent;
1982 if (znode->child_cnt < c->fanout) {
1983 ubifs_assert(n != c->fanout);
1984 dbg_tnc("inserted at %d level %d, key %s", n, znode->level,
1985 DBGKEY(key));
1986
1987 insert_zbranch(znode, zbr, n);
1988
1989 /* Ensure parent's key is correct */
1990 if (n == 0 && zp && znode->iip == 0)
1991 correct_parent_keys(c, znode);
1992
1993 return 0;
1994 }
1995
1996 /*
1997 * Unfortunately, @znode does not have more empty slots and we have to
1998 * split it.
1999 */
2000 dbg_tnc("splitting level %d, key %s", znode->level, DBGKEY(key));
2001
2002 if (znode->alt)
2003 /*
2004 * We can no longer be sure of finding this znode by key, so we
2005 * record it in the old_idx tree.
2006 */
2007 ins_clr_old_idx_znode(c, znode);
2008
2009 zn = kzalloc(c->max_znode_sz, GFP_NOFS);
2010 if (!zn)
2011 return -ENOMEM;
2012 zn->parent = zp;
2013 zn->level = znode->level;
2014
2015 /* Decide where to split */
Adrian Hunter2242c682008-09-05 11:56:05 +03002016 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) {
2017 /* Try not to split consecutive data keys */
2018 if (n == c->fanout) {
2019 key1 = &znode->zbranch[n - 1].key;
2020 if (key_inum(c, key1) == key_inum(c, key) &&
2021 key_type(c, key1) == UBIFS_DATA_KEY)
2022 appending = 1;
2023 } else
2024 goto check_split;
2025 } else if (appending && n != c->fanout) {
2026 /* Try not to split consecutive data keys */
2027 appending = 0;
2028check_split:
2029 if (n >= (c->fanout + 1) / 2) {
2030 key1 = &znode->zbranch[0].key;
2031 if (key_inum(c, key1) == key_inum(c, key) &&
2032 key_type(c, key1) == UBIFS_DATA_KEY) {
2033 key1 = &znode->zbranch[n].key;
2034 if (key_inum(c, key1) != key_inum(c, key) ||
2035 key_type(c, key1) != UBIFS_DATA_KEY) {
2036 keep = n;
2037 move = c->fanout - keep;
2038 zi = znode;
2039 goto do_split;
2040 }
2041 }
2042 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002043 }
2044
2045 if (appending) {
2046 keep = c->fanout;
2047 move = 0;
2048 } else {
2049 keep = (c->fanout + 1) / 2;
2050 move = c->fanout - keep;
2051 }
2052
2053 /*
2054 * Although we don't at present, we could look at the neighbors and see
2055 * if we can move some zbranches there.
2056 */
2057
2058 if (n < keep) {
2059 /* Insert into existing znode */
2060 zi = znode;
2061 move += 1;
2062 keep -= 1;
2063 } else {
2064 /* Insert into new znode */
2065 zi = zn;
2066 n -= keep;
2067 /* Re-parent */
2068 if (zn->level != 0)
2069 zbr->znode->parent = zn;
2070 }
2071
Adrian Hunter2242c682008-09-05 11:56:05 +03002072do_split:
2073
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002074 __set_bit(DIRTY_ZNODE, &zn->flags);
2075 atomic_long_inc(&c->dirty_zn_cnt);
2076
2077 zn->child_cnt = move;
2078 znode->child_cnt = keep;
2079
2080 dbg_tnc("moving %d, keeping %d", move, keep);
2081
2082 /* Move zbranch */
2083 for (i = 0; i < move; i++) {
2084 zn->zbranch[i] = znode->zbranch[keep + i];
2085 /* Re-parent */
2086 if (zn->level != 0)
2087 if (zn->zbranch[i].znode) {
2088 zn->zbranch[i].znode->parent = zn;
2089 zn->zbranch[i].znode->iip = i;
2090 }
2091 }
2092
2093 /* Insert new key and branch */
2094 dbg_tnc("inserting at %d level %d, key %s", n, zn->level, DBGKEY(key));
2095
2096 insert_zbranch(zi, zbr, n);
2097
2098 /* Insert new znode (produced by spitting) into the parent */
2099 if (zp) {
Adrian Hunter2242c682008-09-05 11:56:05 +03002100 if (n == 0 && zi == znode && znode->iip == 0)
2101 correct_parent_keys(c, znode);
2102
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002103 /* Locate insertion point */
2104 n = znode->iip + 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002105
2106 /* Tail recursion */
2107 zbr->key = zn->zbranch[0].key;
2108 zbr->znode = zn;
2109 zbr->lnum = 0;
2110 zbr->offs = 0;
2111 zbr->len = 0;
2112 znode = zp;
2113
2114 goto again;
2115 }
2116
2117 /* We have to split root znode */
2118 dbg_tnc("creating new zroot at level %d", znode->level + 1);
2119
2120 zi = kzalloc(c->max_znode_sz, GFP_NOFS);
2121 if (!zi)
2122 return -ENOMEM;
2123
2124 zi->child_cnt = 2;
2125 zi->level = znode->level + 1;
2126
2127 __set_bit(DIRTY_ZNODE, &zi->flags);
2128 atomic_long_inc(&c->dirty_zn_cnt);
2129
2130 zi->zbranch[0].key = znode->zbranch[0].key;
2131 zi->zbranch[0].znode = znode;
2132 zi->zbranch[0].lnum = c->zroot.lnum;
2133 zi->zbranch[0].offs = c->zroot.offs;
2134 zi->zbranch[0].len = c->zroot.len;
2135 zi->zbranch[1].key = zn->zbranch[0].key;
2136 zi->zbranch[1].znode = zn;
2137
2138 c->zroot.lnum = 0;
2139 c->zroot.offs = 0;
2140 c->zroot.len = 0;
2141 c->zroot.znode = zi;
2142
2143 zn->parent = zi;
2144 zn->iip = 1;
2145 znode->parent = zi;
2146 znode->iip = 0;
2147
2148 return 0;
2149}
2150
2151/**
2152 * ubifs_tnc_add - add a node to TNC.
2153 * @c: UBIFS file-system description object
2154 * @key: key to add
2155 * @lnum: LEB number of node
2156 * @offs: node offset
2157 * @len: node length
2158 *
2159 * This function adds a node with key @key to TNC. The node may be new or it may
2160 * obsolete some existing one. Returns %0 on success or negative error code on
2161 * failure.
2162 */
2163int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum,
2164 int offs, int len)
2165{
2166 int found, n, err = 0;
2167 struct ubifs_znode *znode;
2168
2169 mutex_lock(&c->tnc_mutex);
2170 dbg_tnc("%d:%d, len %d, key %s", lnum, offs, len, DBGKEY(key));
2171 found = lookup_level0_dirty(c, key, &znode, &n);
2172 if (!found) {
2173 struct ubifs_zbranch zbr;
2174
2175 zbr.znode = NULL;
2176 zbr.lnum = lnum;
2177 zbr.offs = offs;
2178 zbr.len = len;
2179 key_copy(c, key, &zbr.key);
2180 err = tnc_insert(c, znode, &zbr, n + 1);
2181 } else if (found == 1) {
2182 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2183
2184 lnc_free(zbr);
2185 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2186 zbr->lnum = lnum;
2187 zbr->offs = offs;
2188 zbr->len = len;
2189 } else
2190 err = found;
2191 if (!err)
2192 err = dbg_check_tnc(c, 0);
2193 mutex_unlock(&c->tnc_mutex);
2194
2195 return err;
2196}
2197
2198/**
2199 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found.
2200 * @c: UBIFS file-system description object
2201 * @key: key to add
2202 * @old_lnum: LEB number of old node
2203 * @old_offs: old node offset
2204 * @lnum: LEB number of node
2205 * @offs: node offset
2206 * @len: node length
2207 *
2208 * This function replaces a node with key @key in the TNC only if the old node
2209 * is found. This function is called by garbage collection when node are moved.
2210 * Returns %0 on success or negative error code on failure.
2211 */
2212int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key,
2213 int old_lnum, int old_offs, int lnum, int offs, int len)
2214{
2215 int found, n, err = 0;
2216 struct ubifs_znode *znode;
2217
2218 mutex_lock(&c->tnc_mutex);
2219 dbg_tnc("old LEB %d:%d, new LEB %d:%d, len %d, key %s", old_lnum,
2220 old_offs, lnum, offs, len, DBGKEY(key));
2221 found = lookup_level0_dirty(c, key, &znode, &n);
2222 if (found < 0) {
2223 err = found;
2224 goto out_unlock;
2225 }
2226
2227 if (found == 1) {
2228 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2229
2230 found = 0;
2231 if (zbr->lnum == old_lnum && zbr->offs == old_offs) {
2232 lnc_free(zbr);
2233 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2234 if (err)
2235 goto out_unlock;
2236 zbr->lnum = lnum;
2237 zbr->offs = offs;
2238 zbr->len = len;
2239 found = 1;
2240 } else if (is_hash_key(c, key)) {
2241 found = resolve_collision_directly(c, key, &znode, &n,
2242 old_lnum, old_offs);
2243 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d",
2244 found, znode, n, old_lnum, old_offs);
2245 if (found < 0) {
2246 err = found;
2247 goto out_unlock;
2248 }
2249
2250 if (found) {
2251 /* Ensure the znode is dirtied */
2252 if (znode->cnext || !ubifs_zn_dirty(znode)) {
Artem Bityutskiyf92b9822008-12-28 11:34:26 +02002253 znode = dirty_cow_bottom_up(c, znode);
2254 if (IS_ERR(znode)) {
2255 err = PTR_ERR(znode);
2256 goto out_unlock;
2257 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002258 }
2259 zbr = &znode->zbranch[n];
2260 lnc_free(zbr);
2261 err = ubifs_add_dirt(c, zbr->lnum,
2262 zbr->len);
2263 if (err)
2264 goto out_unlock;
2265 zbr->lnum = lnum;
2266 zbr->offs = offs;
2267 zbr->len = len;
2268 }
2269 }
2270 }
2271
2272 if (!found)
2273 err = ubifs_add_dirt(c, lnum, len);
2274
2275 if (!err)
2276 err = dbg_check_tnc(c, 0);
2277
2278out_unlock:
2279 mutex_unlock(&c->tnc_mutex);
2280 return err;
2281}
2282
2283/**
2284 * ubifs_tnc_add_nm - add a "hashed" node to TNC.
2285 * @c: UBIFS file-system description object
2286 * @key: key to add
2287 * @lnum: LEB number of node
2288 * @offs: node offset
2289 * @len: node length
2290 * @nm: node name
2291 *
2292 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which
2293 * may have collisions, like directory entry keys.
2294 */
2295int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key,
2296 int lnum, int offs, int len, const struct qstr *nm)
2297{
2298 int found, n, err = 0;
2299 struct ubifs_znode *znode;
2300
2301 mutex_lock(&c->tnc_mutex);
2302 dbg_tnc("LEB %d:%d, name '%.*s', key %s", lnum, offs, nm->len, nm->name,
2303 DBGKEY(key));
2304 found = lookup_level0_dirty(c, key, &znode, &n);
2305 if (found < 0) {
2306 err = found;
2307 goto out_unlock;
2308 }
2309
2310 if (found == 1) {
2311 if (c->replaying)
2312 found = fallible_resolve_collision(c, key, &znode, &n,
2313 nm, 1);
2314 else
2315 found = resolve_collision(c, key, &znode, &n, nm);
2316 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n);
2317 if (found < 0) {
2318 err = found;
2319 goto out_unlock;
2320 }
2321
2322 /* Ensure the znode is dirtied */
2323 if (znode->cnext || !ubifs_zn_dirty(znode)) {
Artem Bityutskiyf92b9822008-12-28 11:34:26 +02002324 znode = dirty_cow_bottom_up(c, znode);
2325 if (IS_ERR(znode)) {
2326 err = PTR_ERR(znode);
2327 goto out_unlock;
2328 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002329 }
2330
2331 if (found == 1) {
2332 struct ubifs_zbranch *zbr = &znode->zbranch[n];
2333
2334 lnc_free(zbr);
2335 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2336 zbr->lnum = lnum;
2337 zbr->offs = offs;
2338 zbr->len = len;
2339 goto out_unlock;
2340 }
2341 }
2342
2343 if (!found) {
2344 struct ubifs_zbranch zbr;
2345
2346 zbr.znode = NULL;
2347 zbr.lnum = lnum;
2348 zbr.offs = offs;
2349 zbr.len = len;
2350 key_copy(c, key, &zbr.key);
2351 err = tnc_insert(c, znode, &zbr, n + 1);
2352 if (err)
2353 goto out_unlock;
2354 if (c->replaying) {
2355 /*
2356 * We did not find it in the index so there may be a
2357 * dangling branch still in the index. So we remove it
2358 * by passing 'ubifs_tnc_remove_nm()' the same key but
2359 * an unmatchable name.
2360 */
2361 struct qstr noname = { .len = 0, .name = "" };
2362
2363 err = dbg_check_tnc(c, 0);
2364 mutex_unlock(&c->tnc_mutex);
2365 if (err)
2366 return err;
2367 return ubifs_tnc_remove_nm(c, key, &noname);
2368 }
2369 }
2370
2371out_unlock:
2372 if (!err)
2373 err = dbg_check_tnc(c, 0);
2374 mutex_unlock(&c->tnc_mutex);
2375 return err;
2376}
2377
2378/**
2379 * tnc_delete - delete a znode form TNC.
2380 * @c: UBIFS file-system description object
2381 * @znode: znode to delete from
2382 * @n: zbranch slot number to delete
2383 *
2384 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in
2385 * case of success and a negative error code in case of failure.
2386 */
2387static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n)
2388{
2389 struct ubifs_zbranch *zbr;
2390 struct ubifs_znode *zp;
2391 int i, err;
2392
2393 /* Delete without merge for now */
2394 ubifs_assert(znode->level == 0);
2395 ubifs_assert(n >= 0 && n < c->fanout);
2396 dbg_tnc("deleting %s", DBGKEY(&znode->zbranch[n].key));
2397
2398 zbr = &znode->zbranch[n];
2399 lnc_free(zbr);
2400
2401 err = ubifs_add_dirt(c, zbr->lnum, zbr->len);
2402 if (err) {
2403 dbg_dump_znode(c, znode);
2404 return err;
2405 }
2406
2407 /* We do not "gap" zbranch slots */
2408 for (i = n; i < znode->child_cnt - 1; i++)
2409 znode->zbranch[i] = znode->zbranch[i + 1];
2410 znode->child_cnt -= 1;
2411
2412 if (znode->child_cnt > 0)
2413 return 0;
2414
2415 /*
2416 * This was the last zbranch, we have to delete this znode from the
2417 * parent.
2418 */
2419
2420 do {
2421 ubifs_assert(!test_bit(OBSOLETE_ZNODE, &znode->flags));
2422 ubifs_assert(ubifs_zn_dirty(znode));
2423
2424 zp = znode->parent;
2425 n = znode->iip;
2426
2427 atomic_long_dec(&c->dirty_zn_cnt);
2428
2429 err = insert_old_idx_znode(c, znode);
2430 if (err)
2431 return err;
2432
2433 if (znode->cnext) {
2434 __set_bit(OBSOLETE_ZNODE, &znode->flags);
2435 atomic_long_inc(&c->clean_zn_cnt);
2436 atomic_long_inc(&ubifs_clean_zn_cnt);
2437 } else
2438 kfree(znode);
2439 znode = zp;
2440 } while (znode->child_cnt == 1); /* while removing last child */
2441
2442 /* Remove from znode, entry n - 1 */
2443 znode->child_cnt -= 1;
2444 ubifs_assert(znode->level != 0);
2445 for (i = n; i < znode->child_cnt; i++) {
2446 znode->zbranch[i] = znode->zbranch[i + 1];
2447 if (znode->zbranch[i].znode)
2448 znode->zbranch[i].znode->iip = i;
2449 }
2450
2451 /*
2452 * If this is the root and it has only 1 child then
2453 * collapse the tree.
2454 */
2455 if (!znode->parent) {
2456 while (znode->child_cnt == 1 && znode->level != 0) {
2457 zp = znode;
2458 zbr = &znode->zbranch[0];
2459 znode = get_znode(c, znode, 0);
2460 if (IS_ERR(znode))
2461 return PTR_ERR(znode);
2462 znode = dirty_cow_znode(c, zbr);
2463 if (IS_ERR(znode))
2464 return PTR_ERR(znode);
2465 znode->parent = NULL;
2466 znode->iip = 0;
2467 if (c->zroot.len) {
2468 err = insert_old_idx(c, c->zroot.lnum,
2469 c->zroot.offs);
2470 if (err)
2471 return err;
2472 }
2473 c->zroot.lnum = zbr->lnum;
2474 c->zroot.offs = zbr->offs;
2475 c->zroot.len = zbr->len;
2476 c->zroot.znode = znode;
2477 ubifs_assert(!test_bit(OBSOLETE_ZNODE,
2478 &zp->flags));
2479 ubifs_assert(test_bit(DIRTY_ZNODE, &zp->flags));
2480 atomic_long_dec(&c->dirty_zn_cnt);
2481
2482 if (zp->cnext) {
2483 __set_bit(OBSOLETE_ZNODE, &zp->flags);
2484 atomic_long_inc(&c->clean_zn_cnt);
2485 atomic_long_inc(&ubifs_clean_zn_cnt);
2486 } else
2487 kfree(zp);
2488 }
2489 }
2490
2491 return 0;
2492}
2493
2494/**
2495 * ubifs_tnc_remove - remove an index entry of a node.
2496 * @c: UBIFS file-system description object
2497 * @key: key of node
2498 *
2499 * Returns %0 on success or negative error code on failure.
2500 */
2501int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key)
2502{
2503 int found, n, err = 0;
2504 struct ubifs_znode *znode;
2505
2506 mutex_lock(&c->tnc_mutex);
2507 dbg_tnc("key %s", DBGKEY(key));
2508 found = lookup_level0_dirty(c, key, &znode, &n);
2509 if (found < 0) {
2510 err = found;
2511 goto out_unlock;
2512 }
2513 if (found == 1)
2514 err = tnc_delete(c, znode, n);
2515 if (!err)
2516 err = dbg_check_tnc(c, 0);
2517
2518out_unlock:
2519 mutex_unlock(&c->tnc_mutex);
2520 return err;
2521}
2522
2523/**
2524 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node.
2525 * @c: UBIFS file-system description object
2526 * @key: key of node
2527 * @nm: directory entry name
2528 *
2529 * Returns %0 on success or negative error code on failure.
2530 */
2531int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key,
2532 const struct qstr *nm)
2533{
2534 int n, err;
2535 struct ubifs_znode *znode;
2536
2537 mutex_lock(&c->tnc_mutex);
2538 dbg_tnc("%.*s, key %s", nm->len, nm->name, DBGKEY(key));
2539 err = lookup_level0_dirty(c, key, &znode, &n);
2540 if (err < 0)
2541 goto out_unlock;
2542
2543 if (err) {
2544 if (c->replaying)
2545 err = fallible_resolve_collision(c, key, &znode, &n,
2546 nm, 0);
2547 else
2548 err = resolve_collision(c, key, &znode, &n, nm);
2549 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n);
2550 if (err < 0)
2551 goto out_unlock;
2552 if (err) {
2553 /* Ensure the znode is dirtied */
2554 if (znode->cnext || !ubifs_zn_dirty(znode)) {
2555 znode = dirty_cow_bottom_up(c, znode);
2556 if (IS_ERR(znode)) {
2557 err = PTR_ERR(znode);
2558 goto out_unlock;
2559 }
2560 }
2561 err = tnc_delete(c, znode, n);
2562 }
2563 }
2564
2565out_unlock:
2566 if (!err)
2567 err = dbg_check_tnc(c, 0);
2568 mutex_unlock(&c->tnc_mutex);
2569 return err;
2570}
2571
2572/**
2573 * key_in_range - determine if a key falls within a range of keys.
2574 * @c: UBIFS file-system description object
2575 * @key: key to check
2576 * @from_key: lowest key in range
2577 * @to_key: highest key in range
2578 *
2579 * This function returns %1 if the key is in range and %0 otherwise.
2580 */
2581static int key_in_range(struct ubifs_info *c, union ubifs_key *key,
2582 union ubifs_key *from_key, union ubifs_key *to_key)
2583{
2584 if (keys_cmp(c, key, from_key) < 0)
2585 return 0;
2586 if (keys_cmp(c, key, to_key) > 0)
2587 return 0;
2588 return 1;
2589}
2590
2591/**
2592 * ubifs_tnc_remove_range - remove index entries in range.
2593 * @c: UBIFS file-system description object
2594 * @from_key: lowest key to remove
2595 * @to_key: highest key to remove
2596 *
2597 * This function removes index entries starting at @from_key and ending at
2598 * @to_key. This function returns zero in case of success and a negative error
2599 * code in case of failure.
2600 */
2601int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key,
2602 union ubifs_key *to_key)
2603{
2604 int i, n, k, err = 0;
2605 struct ubifs_znode *znode;
2606 union ubifs_key *key;
2607
2608 mutex_lock(&c->tnc_mutex);
2609 while (1) {
2610 /* Find first level 0 znode that contains keys to remove */
2611 err = ubifs_lookup_level0(c, from_key, &znode, &n);
2612 if (err < 0)
2613 goto out_unlock;
2614
2615 if (err)
2616 key = from_key;
2617 else {
2618 err = tnc_next(c, &znode, &n);
2619 if (err == -ENOENT) {
2620 err = 0;
2621 goto out_unlock;
2622 }
2623 if (err < 0)
2624 goto out_unlock;
2625 key = &znode->zbranch[n].key;
2626 if (!key_in_range(c, key, from_key, to_key)) {
2627 err = 0;
2628 goto out_unlock;
2629 }
2630 }
2631
2632 /* Ensure the znode is dirtied */
2633 if (znode->cnext || !ubifs_zn_dirty(znode)) {
Artem Bityutskiyf92b9822008-12-28 11:34:26 +02002634 znode = dirty_cow_bottom_up(c, znode);
2635 if (IS_ERR(znode)) {
2636 err = PTR_ERR(znode);
2637 goto out_unlock;
2638 }
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002639 }
2640
2641 /* Remove all keys in range except the first */
2642 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) {
2643 key = &znode->zbranch[i].key;
2644 if (!key_in_range(c, key, from_key, to_key))
2645 break;
2646 lnc_free(&znode->zbranch[i]);
2647 err = ubifs_add_dirt(c, znode->zbranch[i].lnum,
2648 znode->zbranch[i].len);
2649 if (err) {
2650 dbg_dump_znode(c, znode);
2651 goto out_unlock;
2652 }
2653 dbg_tnc("removing %s", DBGKEY(key));
2654 }
2655 if (k) {
2656 for (i = n + 1 + k; i < znode->child_cnt; i++)
2657 znode->zbranch[i - k] = znode->zbranch[i];
2658 znode->child_cnt -= k;
2659 }
2660
2661 /* Now delete the first */
2662 err = tnc_delete(c, znode, n);
2663 if (err)
2664 goto out_unlock;
2665 }
2666
2667out_unlock:
2668 if (!err)
2669 err = dbg_check_tnc(c, 0);
2670 mutex_unlock(&c->tnc_mutex);
2671 return err;
2672}
2673
2674/**
2675 * ubifs_tnc_remove_ino - remove an inode from TNC.
2676 * @c: UBIFS file-system description object
2677 * @inum: inode number to remove
2678 *
2679 * This function remove inode @inum and all the extended attributes associated
2680 * with the anode from TNC and returns zero in case of success or a negative
2681 * error code in case of failure.
2682 */
2683int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum)
2684{
2685 union ubifs_key key1, key2;
2686 struct ubifs_dent_node *xent, *pxent = NULL;
2687 struct qstr nm = { .name = NULL };
2688
Artem Bityutskiye84461a2008-10-29 12:08:43 +02002689 dbg_tnc("ino %lu", (unsigned long)inum);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002690
2691 /*
2692 * Walk all extended attribute entries and remove them together with
2693 * corresponding extended attribute inodes.
2694 */
2695 lowest_xent_key(c, &key1, inum);
2696 while (1) {
2697 ino_t xattr_inum;
2698 int err;
2699
2700 xent = ubifs_tnc_next_ent(c, &key1, &nm);
2701 if (IS_ERR(xent)) {
2702 err = PTR_ERR(xent);
2703 if (err == -ENOENT)
2704 break;
2705 return err;
2706 }
2707
2708 xattr_inum = le64_to_cpu(xent->inum);
Artem Bityutskiye84461a2008-10-29 12:08:43 +02002709 dbg_tnc("xent '%s', ino %lu", xent->name,
2710 (unsigned long)xattr_inum);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03002711
2712 nm.name = xent->name;
2713 nm.len = le16_to_cpu(xent->nlen);
2714 err = ubifs_tnc_remove_nm(c, &key1, &nm);
2715 if (err) {
2716 kfree(xent);
2717 return err;
2718 }
2719
2720 lowest_ino_key(c, &key1, xattr_inum);
2721 highest_ino_key(c, &key2, xattr_inum);
2722 err = ubifs_tnc_remove_range(c, &key1, &key2);
2723 if (err) {
2724 kfree(xent);
2725 return err;
2726 }
2727
2728 kfree(pxent);
2729 pxent = xent;
2730 key_read(c, &xent->key, &key1);
2731 }
2732
2733 kfree(pxent);
2734 lowest_ino_key(c, &key1, inum);
2735 highest_ino_key(c, &key2, inum);
2736
2737 return ubifs_tnc_remove_range(c, &key1, &key2);
2738}
2739
2740/**
2741 * ubifs_tnc_next_ent - walk directory or extended attribute entries.
2742 * @c: UBIFS file-system description object
2743 * @key: key of last entry
2744 * @nm: name of last entry found or %NULL
2745 *
2746 * This function finds and reads the next directory or extended attribute entry
2747 * after the given key (@key) if there is one. @nm is used to resolve
2748 * collisions.
2749 *
2750 * If the name of the current entry is not known and only the key is known,
2751 * @nm->name has to be %NULL. In this case the semantics of this function is a
2752 * little bit different and it returns the entry corresponding to this key, not
2753 * the next one. If the key was not found, the closest "right" entry is
2754 * returned.
2755 *
2756 * If the fist entry has to be found, @key has to contain the lowest possible
2757 * key value for this inode and @name has to be %NULL.
2758 *
2759 * This function returns the found directory or extended attribute entry node
2760 * in case of success, %-ENOENT is returned if no entry was found, and a
2761 * negative error code is returned in case of failure.
2762 */
2763struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c,
2764 union ubifs_key *key,
2765 const struct qstr *nm)
2766{
2767 int n, err, type = key_type(c, key);
2768 struct ubifs_znode *znode;
2769 struct ubifs_dent_node *dent;
2770 struct ubifs_zbranch *zbr;
2771 union ubifs_key *dkey;
2772
2773 dbg_tnc("%s %s", nm->name ? (char *)nm->name : "(lowest)", DBGKEY(key));
2774 ubifs_assert(is_hash_key(c, key));
2775
2776 mutex_lock(&c->tnc_mutex);
2777 err = ubifs_lookup_level0(c, key, &znode, &n);
2778 if (unlikely(err < 0))
2779 goto out_unlock;
2780
2781 if (nm->name) {
2782 if (err) {
2783 /* Handle collisions */
2784 err = resolve_collision(c, key, &znode, &n, nm);
2785 dbg_tnc("rc returned %d, znode %p, n %d",
2786 err, znode, n);
2787 if (unlikely(err < 0))
2788 goto out_unlock;
2789 }
2790
2791 /* Now find next entry */
2792 err = tnc_next(c, &znode, &n);
2793 if (unlikely(err))
2794 goto out_unlock;
2795 } else {
2796 /*
2797 * The full name of the entry was not given, in which case the
2798 * behavior of this function is a little different and it
2799 * returns current entry, not the next one.
2800 */
2801 if (!err) {
2802 /*
2803 * However, the given key does not exist in the TNC
2804 * tree and @znode/@n variables contain the closest
2805 * "preceding" element. Switch to the next one.
2806 */
2807 err = tnc_next(c, &znode, &n);
2808 if (err)
2809 goto out_unlock;
2810 }
2811 }
2812
2813 zbr = &znode->zbranch[n];
2814 dent = kmalloc(zbr->len, GFP_NOFS);
2815 if (unlikely(!dent)) {
2816 err = -ENOMEM;
2817 goto out_unlock;
2818 }
2819
2820 /*
2821 * The above 'tnc_next()' call could lead us to the next inode, check
2822 * this.
2823 */
2824 dkey = &zbr->key;
2825 if (key_inum(c, dkey) != key_inum(c, key) ||
2826 key_type(c, dkey) != type) {
2827 err = -ENOENT;
2828 goto out_free;
2829 }
2830
2831 err = tnc_read_node_nm(c, zbr, dent);
2832 if (unlikely(err))
2833 goto out_free;
2834
2835 mutex_unlock(&c->tnc_mutex);
2836 return dent;
2837
2838out_free:
2839 kfree(dent);
2840out_unlock:
2841 mutex_unlock(&c->tnc_mutex);
2842 return ERR_PTR(err);
2843}
2844
2845/**
2846 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit.
2847 * @c: UBIFS file-system description object
2848 *
2849 * Destroy left-over obsolete znodes from a failed commit.
2850 */
2851static void tnc_destroy_cnext(struct ubifs_info *c)
2852{
2853 struct ubifs_znode *cnext;
2854
2855 if (!c->cnext)
2856 return;
2857 ubifs_assert(c->cmt_state == COMMIT_BROKEN);
2858 cnext = c->cnext;
2859 do {
2860 struct ubifs_znode *znode = cnext;
2861
2862 cnext = cnext->cnext;
2863 if (test_bit(OBSOLETE_ZNODE, &znode->flags))
2864 kfree(znode);
2865 } while (cnext && cnext != c->cnext);
2866}
2867
2868/**
2869 * ubifs_tnc_close - close TNC subsystem and free all related resources.
2870 * @c: UBIFS file-system description object
2871 */
2872void ubifs_tnc_close(struct ubifs_info *c)
2873{
2874 long clean_freed;
2875
2876 tnc_destroy_cnext(c);
2877 if (c->zroot.znode) {
2878 clean_freed = ubifs_destroy_tnc_subtree(c->zroot.znode);
2879 atomic_long_sub(clean_freed, &ubifs_clean_zn_cnt);
2880 }
2881 kfree(c->gap_lebs);
2882 kfree(c->ilebs);
2883 destroy_old_idx(c);
2884}
2885
2886/**
2887 * left_znode - get the znode to the left.
2888 * @c: UBIFS file-system description object
2889 * @znode: znode
2890 *
2891 * This function returns a pointer to the znode to the left of @znode or NULL if
2892 * there is not one. A negative error code is returned on failure.
2893 */
2894static struct ubifs_znode *left_znode(struct ubifs_info *c,
2895 struct ubifs_znode *znode)
2896{
2897 int level = znode->level;
2898
2899 while (1) {
2900 int n = znode->iip - 1;
2901
2902 /* Go up until we can go left */
2903 znode = znode->parent;
2904 if (!znode)
2905 return NULL;
2906 if (n >= 0) {
2907 /* Now go down the rightmost branch to 'level' */
2908 znode = get_znode(c, znode, n);
2909 if (IS_ERR(znode))
2910 return znode;
2911 while (znode->level != level) {
2912 n = znode->child_cnt - 1;
2913 znode = get_znode(c, znode, n);
2914 if (IS_ERR(znode))
2915 return znode;
2916 }
2917 break;
2918 }
2919 }
2920 return znode;
2921}
2922
2923/**
2924 * right_znode - get the znode to the right.
2925 * @c: UBIFS file-system description object
2926 * @znode: znode
2927 *
2928 * This function returns a pointer to the znode to the right of @znode or NULL
2929 * if there is not one. A negative error code is returned on failure.
2930 */
2931static struct ubifs_znode *right_znode(struct ubifs_info *c,
2932 struct ubifs_znode *znode)
2933{
2934 int level = znode->level;
2935
2936 while (1) {
2937 int n = znode->iip + 1;
2938
2939 /* Go up until we can go right */
2940 znode = znode->parent;
2941 if (!znode)
2942 return NULL;
2943 if (n < znode->child_cnt) {
2944 /* Now go down the leftmost branch to 'level' */
2945 znode = get_znode(c, znode, n);
2946 if (IS_ERR(znode))
2947 return znode;
2948 while (znode->level != level) {
2949 znode = get_znode(c, znode, 0);
2950 if (IS_ERR(znode))
2951 return znode;
2952 }
2953 break;
2954 }
2955 }
2956 return znode;
2957}
2958
2959/**
2960 * lookup_znode - find a particular indexing node from TNC.
2961 * @c: UBIFS file-system description object
2962 * @key: index node key to lookup
2963 * @level: index node level
2964 * @lnum: index node LEB number
2965 * @offs: index node offset
2966 *
2967 * This function searches an indexing node by its first key @key and its
2968 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing
2969 * nodes it traverses to TNC. This function is called fro indexing nodes which
2970 * were found on the media by scanning, for example when garbage-collecting or
2971 * when doing in-the-gaps commit. This means that the indexing node which is
2972 * looked for does not have to have exactly the same leftmost key @key, because
2973 * the leftmost key may have been changed, in which case TNC will contain a
2974 * dirty znode which still refers the same @lnum:@offs. This function is clever
2975 * enough to recognize such indexing nodes.
2976 *
2977 * Note, if a znode was deleted or changed too much, then this function will
2978 * not find it. For situations like this UBIFS has the old index RB-tree
2979 * (indexed by @lnum:@offs).
2980 *
2981 * This function returns a pointer to the znode found or %NULL if it is not
2982 * found. A negative error code is returned on failure.
2983 */
2984static struct ubifs_znode *lookup_znode(struct ubifs_info *c,
2985 union ubifs_key *key, int level,
2986 int lnum, int offs)
2987{
2988 struct ubifs_znode *znode, *zn;
2989 int n, nn;
2990
2991 /*
2992 * The arguments have probably been read off flash, so don't assume
2993 * they are valid.
2994 */
2995 if (level < 0)
2996 return ERR_PTR(-EINVAL);
2997
2998 /* Get the root znode */
2999 znode = c->zroot.znode;
3000 if (!znode) {
3001 znode = ubifs_load_znode(c, &c->zroot, NULL, 0);
3002 if (IS_ERR(znode))
3003 return znode;
3004 }
3005 /* Check if it is the one we are looking for */
3006 if (c->zroot.lnum == lnum && c->zroot.offs == offs)
3007 return znode;
3008 /* Descend to the parent level i.e. (level + 1) */
3009 if (level >= znode->level)
3010 return NULL;
3011 while (1) {
3012 ubifs_search_zbranch(c, znode, key, &n);
3013 if (n < 0) {
3014 /*
3015 * We reached a znode where the leftmost key is greater
3016 * than the key we are searching for. This is the same
3017 * situation as the one described in a huge comment at
3018 * the end of the 'ubifs_lookup_level0()' function. And
3019 * for exactly the same reasons we have to try to look
3020 * left before giving up.
3021 */
3022 znode = left_znode(c, znode);
3023 if (!znode)
3024 return NULL;
3025 if (IS_ERR(znode))
3026 return znode;
3027 ubifs_search_zbranch(c, znode, key, &n);
3028 ubifs_assert(n >= 0);
3029 }
3030 if (znode->level == level + 1)
3031 break;
3032 znode = get_znode(c, znode, n);
3033 if (IS_ERR(znode))
3034 return znode;
3035 }
3036 /* Check if the child is the one we are looking for */
3037 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs)
3038 return get_znode(c, znode, n);
3039 /* If the key is unique, there is nowhere else to look */
3040 if (!is_hash_key(c, key))
3041 return NULL;
3042 /*
3043 * The key is not unique and so may be also in the znodes to either
3044 * side.
3045 */
3046 zn = znode;
3047 nn = n;
3048 /* Look left */
3049 while (1) {
3050 /* Move one branch to the left */
3051 if (n)
3052 n -= 1;
3053 else {
3054 znode = left_znode(c, znode);
3055 if (!znode)
3056 break;
3057 if (IS_ERR(znode))
3058 return znode;
3059 n = znode->child_cnt - 1;
3060 }
3061 /* Check it */
3062 if (znode->zbranch[n].lnum == lnum &&
3063 znode->zbranch[n].offs == offs)
3064 return get_znode(c, znode, n);
3065 /* Stop if the key is less than the one we are looking for */
3066 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0)
3067 break;
3068 }
3069 /* Back to the middle */
3070 znode = zn;
3071 n = nn;
3072 /* Look right */
3073 while (1) {
3074 /* Move one branch to the right */
3075 if (++n >= znode->child_cnt) {
3076 znode = right_znode(c, znode);
3077 if (!znode)
3078 break;
3079 if (IS_ERR(znode))
3080 return znode;
3081 n = 0;
3082 }
3083 /* Check it */
3084 if (znode->zbranch[n].lnum == lnum &&
3085 znode->zbranch[n].offs == offs)
3086 return get_znode(c, znode, n);
3087 /* Stop if the key is greater than the one we are looking for */
3088 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0)
3089 break;
3090 }
3091 return NULL;
3092}
3093
3094/**
3095 * is_idx_node_in_tnc - determine if an index node is in the TNC.
3096 * @c: UBIFS file-system description object
3097 * @key: key of index node
3098 * @level: index node level
3099 * @lnum: LEB number of index node
3100 * @offs: offset of index node
3101 *
3102 * This function returns %0 if the index node is not referred to in the TNC, %1
3103 * if the index node is referred to in the TNC and the corresponding znode is
3104 * dirty, %2 if an index node is referred to in the TNC and the corresponding
3105 * znode is clean, and a negative error code in case of failure.
3106 *
3107 * Note, the @key argument has to be the key of the first child. Also note,
3108 * this function relies on the fact that 0:0 is never a valid LEB number and
3109 * offset for a main-area node.
3110 */
3111int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level,
3112 int lnum, int offs)
3113{
3114 struct ubifs_znode *znode;
3115
3116 znode = lookup_znode(c, key, level, lnum, offs);
3117 if (!znode)
3118 return 0;
3119 if (IS_ERR(znode))
3120 return PTR_ERR(znode);
3121
3122 return ubifs_zn_dirty(znode) ? 1 : 2;
3123}
3124
3125/**
3126 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC.
3127 * @c: UBIFS file-system description object
3128 * @key: node key
3129 * @lnum: node LEB number
3130 * @offs: node offset
3131 *
3132 * This function returns %1 if the node is referred to in the TNC, %0 if it is
3133 * not, and a negative error code in case of failure.
3134 *
3135 * Note, this function relies on the fact that 0:0 is never a valid LEB number
3136 * and offset for a main-area node.
3137 */
3138static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key,
3139 int lnum, int offs)
3140{
3141 struct ubifs_zbranch *zbr;
3142 struct ubifs_znode *znode, *zn;
3143 int n, found, err, nn;
3144 const int unique = !is_hash_key(c, key);
3145
3146 found = ubifs_lookup_level0(c, key, &znode, &n);
3147 if (found < 0)
3148 return found; /* Error code */
3149 if (!found)
3150 return 0;
3151 zbr = &znode->zbranch[n];
3152 if (lnum == zbr->lnum && offs == zbr->offs)
3153 return 1; /* Found it */
3154 if (unique)
3155 return 0;
3156 /*
3157 * Because the key is not unique, we have to look left
3158 * and right as well
3159 */
3160 zn = znode;
3161 nn = n;
3162 /* Look left */
3163 while (1) {
3164 err = tnc_prev(c, &znode, &n);
3165 if (err == -ENOENT)
3166 break;
3167 if (err)
3168 return err;
3169 if (keys_cmp(c, key, &znode->zbranch[n].key))
3170 break;
3171 zbr = &znode->zbranch[n];
3172 if (lnum == zbr->lnum && offs == zbr->offs)
3173 return 1; /* Found it */
3174 }
3175 /* Look right */
3176 znode = zn;
3177 n = nn;
3178 while (1) {
3179 err = tnc_next(c, &znode, &n);
3180 if (err) {
3181 if (err == -ENOENT)
3182 return 0;
3183 return err;
3184 }
3185 if (keys_cmp(c, key, &znode->zbranch[n].key))
3186 break;
3187 zbr = &znode->zbranch[n];
3188 if (lnum == zbr->lnum && offs == zbr->offs)
3189 return 1; /* Found it */
3190 }
3191 return 0;
3192}
3193
3194/**
3195 * ubifs_tnc_has_node - determine whether a node is in the TNC.
3196 * @c: UBIFS file-system description object
3197 * @key: node key
3198 * @level: index node level (if it is an index node)
3199 * @lnum: node LEB number
3200 * @offs: node offset
3201 * @is_idx: non-zero if the node is an index node
3202 *
3203 * This function returns %1 if the node is in the TNC, %0 if it is not, and a
3204 * negative error code in case of failure. For index nodes, @key has to be the
3205 * key of the first child. An index node is considered to be in the TNC only if
3206 * the corresponding znode is clean or has not been loaded.
3207 */
3208int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level,
3209 int lnum, int offs, int is_idx)
3210{
3211 int err;
3212
3213 mutex_lock(&c->tnc_mutex);
3214 if (is_idx) {
3215 err = is_idx_node_in_tnc(c, key, level, lnum, offs);
3216 if (err < 0)
3217 goto out_unlock;
3218 if (err == 1)
3219 /* The index node was found but it was dirty */
3220 err = 0;
3221 else if (err == 2)
3222 /* The index node was found and it was clean */
3223 err = 1;
3224 else
3225 BUG_ON(err != 0);
3226 } else
3227 err = is_leaf_node_in_tnc(c, key, lnum, offs);
3228
3229out_unlock:
3230 mutex_unlock(&c->tnc_mutex);
3231 return err;
3232}
3233
3234/**
3235 * ubifs_dirty_idx_node - dirty an index node.
3236 * @c: UBIFS file-system description object
3237 * @key: index node key
3238 * @level: index node level
3239 * @lnum: index node LEB number
3240 * @offs: index node offset
3241 *
3242 * This function loads and dirties an index node so that it can be garbage
3243 * collected. The @key argument has to be the key of the first child. This
3244 * function relies on the fact that 0:0 is never a valid LEB number and offset
3245 * for a main-area node. Returns %0 on success and a negative error code on
3246 * failure.
3247 */
3248int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level,
3249 int lnum, int offs)
3250{
3251 struct ubifs_znode *znode;
3252 int err = 0;
3253
3254 mutex_lock(&c->tnc_mutex);
3255 znode = lookup_znode(c, key, level, lnum, offs);
3256 if (!znode)
3257 goto out_unlock;
3258 if (IS_ERR(znode)) {
3259 err = PTR_ERR(znode);
3260 goto out_unlock;
3261 }
3262 znode = dirty_cow_bottom_up(c, znode);
3263 if (IS_ERR(znode)) {
3264 err = PTR_ERR(znode);
3265 goto out_unlock;
3266 }
3267
3268out_unlock:
3269 mutex_unlock(&c->tnc_mutex);
3270 return err;
3271}
Artem Bityutskiye3c3efc2009-08-27 16:34:19 +03003272
3273#ifdef CONFIG_UBIFS_FS_DEBUG
3274
3275/**
3276 * dbg_check_inode_size - check if inode size is correct.
3277 * @c: UBIFS file-system description object
3278 * @inum: inode number
3279 * @size: inode size
3280 *
3281 * This function makes sure that the inode size (@size) is correct and it does
3282 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL
3283 * if it has a data page beyond @size, and other negative error code in case of
3284 * other errors.
3285 */
3286int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode,
3287 loff_t size)
3288{
3289 int err, n;
3290 union ubifs_key from_key, to_key, *key;
3291 struct ubifs_znode *znode;
3292 unsigned int block;
3293
3294 if (!S_ISREG(inode->i_mode))
3295 return 0;
3296 if (!(ubifs_chk_flags & UBIFS_CHK_GEN))
3297 return 0;
3298
3299 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
3300 data_key_init(c, &from_key, inode->i_ino, block);
3301 highest_data_key(c, &to_key, inode->i_ino);
3302
3303 mutex_lock(&c->tnc_mutex);
3304 err = ubifs_lookup_level0(c, &from_key, &znode, &n);
3305 if (err < 0)
3306 goto out_unlock;
3307
3308 if (err) {
3309 err = -EINVAL;
3310 key = &from_key;
3311 goto out_dump;
3312 }
3313
3314 err = tnc_next(c, &znode, &n);
3315 if (err == -ENOENT) {
3316 err = 0;
3317 goto out_unlock;
3318 }
3319 if (err < 0)
3320 goto out_unlock;
3321
3322 ubifs_assert(err == 0);
3323 key = &znode->zbranch[n].key;
3324 if (!key_in_range(c, key, &from_key, &to_key))
3325 goto out_unlock;
3326
3327out_dump:
3328 block = key_block(c, key);
3329 ubifs_err("inode %lu has size %lld, but there are data at offset %lld "
3330 "(data key %s)", (unsigned long)inode->i_ino, size,
3331 ((loff_t)block) << UBIFS_BLOCK_SHIFT, DBGKEY(key));
3332 dbg_dump_inode(c, inode);
3333 dbg_dump_stack();
3334 err = -EINVAL;
3335
3336out_unlock:
3337 mutex_unlock(&c->tnc_mutex);
3338 return err;
3339}
3340
3341#endif /* CONFIG_UBIFS_FS_DEBUG */