[JFFS2] Remove broken insert_point optimisation in jffs2_add_tn_to_tree()
[linux-2.6.git] / fs / jffs2 / readinode.c
1 /*
2  * JFFS2 -- Journalling Flash File System, Version 2.
3  *
4  * Copyright © 2001-2007 Red Hat, Inc.
5  *
6  * Created by David Woodhouse <dwmw2@infradead.org>
7  *
8  * For licensing information, see the file 'LICENCE' in this directory.
9  *
10  */
11
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/fs.h>
16 #include <linux/crc32.h>
17 #include <linux/pagemap.h>
18 #include <linux/mtd/mtd.h>
19 #include <linux/compiler.h>
20 #include "nodelist.h"
21
22 /*
23  * Check the data CRC of the node.
24  *
25  * Returns: 0 if the data CRC is correct;
26  *          1 - if incorrect;
27  *          error code if an error occured.
28  */
29 static int check_node_data(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
30 {
31         struct jffs2_raw_node_ref *ref = tn->fn->raw;
32         int err = 0, pointed = 0;
33         struct jffs2_eraseblock *jeb;
34         unsigned char *buffer;
35         uint32_t crc, ofs, len;
36         size_t retlen;
37
38         BUG_ON(tn->csize == 0);
39
40         if (!jffs2_is_writebuffered(c))
41                 goto adj_acc;
42
43         /* Calculate how many bytes were already checked */
44         ofs = ref_offset(ref) + sizeof(struct jffs2_raw_inode);
45         len = ofs % c->wbuf_pagesize;
46         if (likely(len))
47                 len = c->wbuf_pagesize - len;
48
49         if (len >= tn->csize) {
50                 dbg_readinode("no need to check node at %#08x, data length %u, data starts at %#08x - it has already been checked.\n",
51                         ref_offset(ref), tn->csize, ofs);
52                 goto adj_acc;
53         }
54
55         ofs += len;
56         len = tn->csize - len;
57
58         dbg_readinode("check node at %#08x, data length %u, partial CRC %#08x, correct CRC %#08x, data starts at %#08x, start checking from %#08x - %u bytes.\n",
59                 ref_offset(ref), tn->csize, tn->partial_crc, tn->data_crc, ofs - len, ofs, len);
60
61 #ifndef __ECOS
62         /* TODO: instead, incapsulate point() stuff to jffs2_flash_read(),
63          * adding and jffs2_flash_read_end() interface. */
64         if (c->mtd->point) {
65                 err = c->mtd->point(c->mtd, ofs, len, &retlen, &buffer);
66                 if (!err && retlen < tn->csize) {
67                         JFFS2_WARNING("MTD point returned len too short: %zu instead of %u.\n", retlen, tn->csize);
68                         c->mtd->unpoint(c->mtd, buffer, ofs, len);
69                 } else if (err)
70                         JFFS2_WARNING("MTD point failed: error code %d.\n", err);
71                 else
72                         pointed = 1; /* succefully pointed to device */
73         }
74 #endif
75
76         if (!pointed) {
77                 buffer = kmalloc(len, GFP_KERNEL);
78                 if (unlikely(!buffer))
79                         return -ENOMEM;
80
81                 /* TODO: this is very frequent pattern, make it a separate
82                  * routine */
83                 err = jffs2_flash_read(c, ofs, len, &retlen, buffer);
84                 if (err) {
85                         JFFS2_ERROR("can not read %d bytes from 0x%08x, error code: %d.\n", len, ofs, err);
86                         goto free_out;
87                 }
88
89                 if (retlen != len) {
90                         JFFS2_ERROR("short read at %#08x: %zd instead of %d.\n", ofs, retlen, len);
91                         err = -EIO;
92                         goto free_out;
93                 }
94         }
95
96         /* Continue calculating CRC */
97         crc = crc32(tn->partial_crc, buffer, len);
98         if(!pointed)
99                 kfree(buffer);
100 #ifndef __ECOS
101         else
102                 c->mtd->unpoint(c->mtd, buffer, ofs, len);
103 #endif
104
105         if (crc != tn->data_crc) {
106                 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
107                         ofs, tn->data_crc, crc);
108                 return 1;
109         }
110
111 adj_acc:
112         jeb = &c->blocks[ref->flash_offset / c->sector_size];
113         len = ref_totlen(c, jeb, ref);
114         /* If it should be REF_NORMAL, it'll get marked as such when
115            we build the fragtree, shortly. No need to worry about GC
116            moving it while it's marked REF_PRISTINE -- GC won't happen
117            till we've finished checking every inode anyway. */
118         ref->flash_offset |= REF_PRISTINE;
119         /*
120          * Mark the node as having been checked and fix the
121          * accounting accordingly.
122          */
123         spin_lock(&c->erase_completion_lock);
124         jeb->used_size += len;
125         jeb->unchecked_size -= len;
126         c->used_size += len;
127         c->unchecked_size -= len;
128         jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
129         spin_unlock(&c->erase_completion_lock);
130
131         return 0;
132
133 free_out:
134         if(!pointed)
135                 kfree(buffer);
136 #ifndef __ECOS
137         else
138                 c->mtd->unpoint(c->mtd, buffer, ofs, len);
139 #endif
140         return err;
141 }
142
143 /*
144  * Helper function for jffs2_add_older_frag_to_fragtree().
145  *
146  * Checks the node if we are in the checking stage.
147  */
148 static int check_tn_node(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
149 {
150         int ret;
151
152         BUG_ON(ref_obsolete(tn->fn->raw));
153
154         /* We only check the data CRC of unchecked nodes */
155         if (ref_flags(tn->fn->raw) != REF_UNCHECKED)
156                 return 0;
157
158         dbg_readinode("check node %#04x-%#04x, phys offs %#08x\n",
159                       tn->fn->ofs, tn->fn->ofs + tn->fn->size, ref_offset(tn->fn->raw));
160
161         ret = check_node_data(c, tn);
162         if (unlikely(ret < 0)) {
163                 JFFS2_ERROR("check_node_data() returned error: %d.\n",
164                         ret);
165         } else if (unlikely(ret > 0)) {
166                 dbg_readinode("CRC error, mark it obsolete.\n");
167                 jffs2_mark_node_obsolete(c, tn->fn->raw);
168         }
169
170         return ret;
171 }
172
173 static struct jffs2_tmp_dnode_info *jffs2_lookup_tn(struct rb_root *tn_root, uint32_t offset)
174 {
175         struct rb_node *next;
176         struct jffs2_tmp_dnode_info *tn = NULL;
177
178         dbg_readinode("root %p, offset %d\n", tn_root, offset);
179
180         next = tn_root->rb_node;
181
182         while (next) {
183                 tn = rb_entry(next, struct jffs2_tmp_dnode_info, rb);
184
185                 if (tn->fn->ofs < offset)
186                         next = tn->rb.rb_right;
187                 else if (tn->fn->ofs >= offset)
188                         next = tn->rb.rb_left;
189                 else
190                         break;
191         }
192
193         return tn;
194 }
195
196
197 static void jffs2_kill_tn(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
198 {
199         jffs2_mark_node_obsolete(c, tn->fn->raw);
200         jffs2_free_full_dnode(tn->fn);
201         jffs2_free_tmp_dnode_info(tn);
202 }
203 /*
204  * This function is used when we read an inode. Data nodes arrive in
205  * arbitrary order -- they may be older or newer than the nodes which
206  * are already in the tree. Where overlaps occur, the older node can
207  * be discarded as long as the newer passes the CRC check. We don't
208  * bother to keep track of holes in this rbtree, and neither do we deal
209  * with frags -- we can have multiple entries starting at the same
210  * offset, and the one with the smallest length will come first in the
211  * ordering.
212  *
213  * Returns 0 if the node was inserted
214  *         1 if the node is obsolete (because we can't mark it so yet)
215  *         < 0 an if error occurred
216  */
217 static int jffs2_add_tn_to_tree(struct jffs2_sb_info *c,
218                                 struct jffs2_readinode_info *rii,
219                                 struct jffs2_tmp_dnode_info *tn)
220 {
221         uint32_t fn_end = tn->fn->ofs + tn->fn->size;
222         struct jffs2_tmp_dnode_info *this;
223
224         dbg_readinode("insert fragment %#04x-%#04x, ver %u\n", tn->fn->ofs, fn_end, tn->version);
225
226         /* If a node has zero dsize, we only have to keep if it if it might be the
227            node with highest version -- i.e. the one which will end up as f->metadata.
228            Note that such nodes won't be REF_UNCHECKED since there are no data to
229            check anyway. */
230         if (!tn->fn->size) {
231                 if (rii->mdata_tn) {
232                         /* We had a candidate mdata node already */
233                         dbg_readinode("kill old mdata with ver %d\n", rii->mdata_tn->version);
234                         jffs2_kill_tn(c, rii->mdata_tn);
235                 }
236                 rii->mdata_tn = tn;
237                 dbg_readinode("keep new mdata with ver %d\n", tn->version);
238                 return 0;
239         }
240
241         /* Find the earliest node which _may_ be relevant to this one */
242         this = jffs2_lookup_tn(&rii->tn_root, tn->fn->ofs);
243         if (!this) {
244                 /* First addition to empty tree. $DEITY how I love the easy cases */
245                 rb_link_node(&tn->rb, NULL, &rii->tn_root.rb_node);
246                 rb_insert_color(&tn->rb, &rii->tn_root);
247                 dbg_readinode("keep new frag\n");
248                 return 0;
249         }
250
251         /* If the node is coincident with another at a lower address,
252            back up until the other node is found. It may be relevant */
253         while (tn->overlapped)
254                 tn = tn_prev(tn);
255
256         dbg_readinode("'this' found %#04x-%#04x (%s)\n", this->fn->ofs, this->fn->ofs + this->fn->size, this->fn ? "data" : "hole");
257
258         while (this) {
259                 if (this->fn->ofs > fn_end)
260                         break;
261                 dbg_readinode("Ponder this ver %d, 0x%x-0x%x\n",
262                               this->version, this->fn->ofs, this->fn->size);
263
264                 if (this->version == tn->version) {
265                         /* Version number collision means REF_PRISTINE GC. Accept either of them
266                            as long as the CRC is correct. Check the one we have already...  */
267                         if (!check_tn_node(c, this)) {
268                                 /* The one we already had was OK. Keep it and throw away the new one */
269                                 dbg_readinode("Like old node. Throw away new\n");
270                                 jffs2_kill_tn(c, tn);
271                                 return 0;
272                         } else {
273                                 /* Who cares if the new one is good; keep it for now anyway. */
274                                 rb_replace_node(&this->rb, &tn->rb, &rii->tn_root);
275                                 /* Same overlapping from in front and behind */
276                                 jffs2_kill_tn(c, this);
277                                 dbg_readinode("Like new node. Throw away old\n");
278                                 goto calc_overlaps;
279                         }
280                 }
281                 if (this->version < tn->version &&
282                     this->fn->ofs >= tn->fn->ofs &&
283                     this->fn->ofs + this->fn->size <= fn_end) {
284                         /* New node entirely overlaps 'this' */
285                         if (check_tn_node(c, tn)) {
286                                 dbg_readinode("new node bad CRC\n");
287                                 jffs2_kill_tn(c, tn);
288                                 return 0;
289                         }
290                         /* ... and is good. Kill 'this'... */
291                         rb_replace_node(&this->rb, &tn->rb, &rii->tn_root);
292                         jffs2_kill_tn(c, this);
293                         /* ... and any subsequent nodes which are also overlapped */
294                         this = tn_next(tn);
295                         while (this && this->fn->ofs + this->fn->size < fn_end) {
296                                 struct jffs2_tmp_dnode_info *next = tn_next(this);
297                                 if (this->version < tn->version) {
298                                         tn_erase(this, &rii->tn_root);
299                                         dbg_readinode("Kill overlapped ver %d, 0x%x-0x%x\n",
300                                                       this->version, this->fn->ofs,
301                                                       this->fn->ofs+this->fn->size);
302                                         jffs2_kill_tn(c, this);
303                                 }
304                                 this = next;
305                         }
306                         dbg_readinode("Done inserting new\n");
307                         goto calc_overlaps;
308                 }
309                 if (this->version > tn->version &&
310                     this->fn->ofs <= tn->fn->ofs &&
311                     this->fn->ofs+this->fn->size >= fn_end) {
312                         /* New node entirely overlapped by 'this' */
313                         if (!check_tn_node(c, this)) {
314                                 dbg_readinode("Good CRC on old node. Kill new\n");
315                                 jffs2_kill_tn(c, tn);
316                                 return 0;
317                         }
318                         /* ... but 'this' was bad. Replace it... */
319                         tn->overlapped = this->overlapped;
320                         rb_replace_node(&this->rb, &tn->rb, &rii->tn_root);
321                         dbg_readinode("Bad CRC on old overlapping node. Kill it\n");
322                         jffs2_kill_tn(c, this);
323                         return 0;
324                 }
325
326                 this = tn_next(this);
327         }
328
329         /* We neither completely obsoleted nor were completely
330            obsoleted by an earlier node. Insert into the tree */
331         {
332                 struct rb_node *parent;
333                 struct rb_node **link = &rii->tn_root.rb_node;
334                 struct jffs2_tmp_dnode_info *insert_point;
335
336                 while (*link) {
337                         parent = *link;
338                         insert_point = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
339                         if (tn->fn->ofs > insert_point->fn->ofs)
340                                 link = &insert_point->rb.rb_right;
341                         else if (tn->fn->ofs < insert_point->fn->ofs ||
342                                  tn->fn->size < insert_point->fn->size)
343                                 link = &insert_point->rb.rb_left;
344                         else
345                                 link = &insert_point->rb.rb_right;
346                 }
347                 rb_link_node(&tn->rb, &insert_point->rb, link);
348                 rb_insert_color(&tn->rb, &rii->tn_root);
349         }
350
351  calc_overlaps:
352         /* If there's anything behind that overlaps us, note it */
353         this = tn_prev(tn);
354         if (this) {
355                 while (1) {
356                         if (this->fn->ofs + this->fn->size > tn->fn->ofs) {
357                                 dbg_readinode("Node is overlapped by %p (v %d, 0x%x-0x%x)\n",
358                                               this, this->version, this->fn->ofs,
359                                               this->fn->ofs+this->fn->size);
360                                 tn->overlapped = 1;
361                                 break;
362                         }
363                         if (!this->overlapped)
364                                 break;
365                         this = tn_prev(this);
366                 }
367         }
368
369         /* If the new node overlaps anything ahead, note it */
370         this = tn_next(tn);
371         while (this && this->fn->ofs < fn_end) {
372                 this->overlapped = 1;
373                 dbg_readinode("Node ver %d, 0x%x-0x%x is overlapped\n",
374                               this->version, this->fn->ofs,
375                               this->fn->ofs+this->fn->size);
376                 this = tn_next(this);
377         }
378         return 0;
379 }
380
381 /* Trivial function to remove the last node in the tree. Which by definition
382    has no right-hand -- so can be removed just by making its only child (if
383    any) take its place under its parent. */
384 static void eat_last(struct rb_root *root, struct rb_node *node)
385 {
386         struct rb_node *parent = rb_parent(node);
387         struct rb_node **link;
388
389         /* LAST! */
390         BUG_ON(node->rb_right);
391
392         if (!parent)
393                 link = &root->rb_node;
394         else if (node == parent->rb_left)
395                 link = &parent->rb_left;
396         else
397                 link = &parent->rb_right;
398
399         *link = node->rb_left;
400         /* Colour doesn't matter now. Only the parent pointer. */
401         if (node->rb_left)
402                 node->rb_left->rb_parent_color = node->rb_parent_color;
403 }
404
405 /* We put this in reverse order, so we can just use eat_last */
406 static void ver_insert(struct rb_root *ver_root, struct jffs2_tmp_dnode_info *tn)
407 {
408         struct rb_node **link = &ver_root->rb_node;
409         struct rb_node *parent = NULL;
410         struct jffs2_tmp_dnode_info *this_tn;
411
412         while (*link) {
413                 parent = *link;
414                 this_tn = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
415
416                 if (tn->version > this_tn->version)
417                         link = &parent->rb_left;
418                 else
419                         link = &parent->rb_right;
420         }
421         dbg_readinode("Link new node at %p (root is %p)\n", link, ver_root);
422         rb_link_node(&tn->rb, parent, link);
423         rb_insert_color(&tn->rb, ver_root);
424 }
425
426 /* Build final, normal fragtree from tn tree. It doesn't matter which order
427    we add nodes to the real fragtree, as long as they don't overlap. And
428    having thrown away the majority of overlapped nodes as we went, there
429    really shouldn't be many sets of nodes which do overlap. If we start at
430    the end, we can use the overlap markers -- we can just eat nodes which
431    aren't overlapped, and when we encounter nodes which _do_ overlap we
432    sort them all into a temporary tree in version order before replaying them. */
433 static int jffs2_build_inode_fragtree(struct jffs2_sb_info *c,
434                                       struct jffs2_inode_info *f,
435                                       struct jffs2_readinode_info *rii)
436 {
437         struct jffs2_tmp_dnode_info *pen, *last, *this;
438         struct rb_root ver_root = RB_ROOT;
439         uint32_t high_ver = 0;
440
441         if (rii->mdata_tn) {
442                 dbg_readinode("potential mdata is ver %d at %p\n", rii->mdata_tn->version, rii->mdata_tn);
443                 high_ver = rii->mdata_tn->version;
444                 rii->latest_ref = rii->mdata_tn->fn->raw;
445         }
446 #ifdef JFFS2_DBG_READINODE_MESSAGES
447         this = tn_last(&rii->tn_root);
448         while (this) {
449                 dbg_readinode("tn %p ver %d range 0x%x-0x%x ov %d\n", this, this->version, this->fn->ofs,
450                               this->fn->ofs+this->fn->size, this->overlapped);
451                 this = tn_prev(this);
452         }
453 #endif
454         pen = tn_last(&rii->tn_root);
455         while ((last = pen)) {
456                 pen = tn_prev(last);
457
458                 eat_last(&rii->tn_root, &last->rb);
459                 ver_insert(&ver_root, last);
460
461                 if (unlikely(last->overlapped))
462                         continue;
463
464                 /* Now we have a bunch of nodes in reverse version
465                    order, in the tree at ver_root. Most of the time,
466                    there'll actually be only one node in the 'tree',
467                    in fact. */
468                 this = tn_last(&ver_root);
469
470                 while (this) {
471                         struct jffs2_tmp_dnode_info *vers_next;
472                         int ret;
473                         vers_next = tn_prev(this);
474                         eat_last(&ver_root, &this->rb);
475                         if (check_tn_node(c, this)) {
476                                 dbg_readinode("node ver %d, 0x%x-0x%x failed CRC\n",
477                                              this->version, this->fn->ofs,
478                                              this->fn->ofs+this->fn->size);
479                                 jffs2_kill_tn(c, this);
480                         } else {
481                                 if (this->version > high_ver) {
482                                         /* Note that this is different from the other
483                                            highest_version, because this one is only
484                                            counting _valid_ nodes which could give the
485                                            latest inode metadata */
486                                         high_ver = this->version;
487                                         rii->latest_ref = this->fn->raw;
488                                 }
489                                 dbg_readinode("Add %p (v %d, 0x%x-0x%x, ov %d) to fragtree\n",
490                                              this, this->version, this->fn->ofs,
491                                              this->fn->ofs+this->fn->size, this->overlapped);
492
493                                 ret = jffs2_add_full_dnode_to_inode(c, f, this->fn);
494                                 if (ret) {
495                                         /* Free the nodes in vers_root; let the caller
496                                            deal with the rest */
497                                         JFFS2_ERROR("Add node to tree failed %d\n", ret);
498                                         while (1) {
499                                                 vers_next = tn_prev(this);
500                                                 if (check_tn_node(c, this))
501                                                         jffs2_mark_node_obsolete(c, this->fn->raw);
502                                                 jffs2_free_full_dnode(this->fn);
503                                                 jffs2_free_tmp_dnode_info(this);
504                                                 this = vers_next;
505                                                 if (!this)
506                                                         break;
507                                                 eat_last(&ver_root, &vers_next->rb);
508                                         }
509                                         return ret;
510                                 }
511                                 jffs2_free_tmp_dnode_info(this);
512                         }
513                         this = vers_next;
514                 }
515         }
516         return 0;
517 }
518
519 static void jffs2_free_tmp_dnode_info_list(struct rb_root *list)
520 {
521         struct rb_node *this;
522         struct jffs2_tmp_dnode_info *tn;
523
524         this = list->rb_node;
525
526         /* Now at bottom of tree */
527         while (this) {
528                 if (this->rb_left)
529                         this = this->rb_left;
530                 else if (this->rb_right)
531                         this = this->rb_right;
532                 else {
533                         tn = rb_entry(this, struct jffs2_tmp_dnode_info, rb);
534                         jffs2_free_full_dnode(tn->fn);
535                         jffs2_free_tmp_dnode_info(tn);
536
537                         this = rb_parent(this);
538                         if (!this)
539                                 break;
540
541                         if (this->rb_left == &tn->rb)
542                                 this->rb_left = NULL;
543                         else if (this->rb_right == &tn->rb)
544                                 this->rb_right = NULL;
545                         else BUG();
546                 }
547         }
548         list->rb_node = NULL;
549 }
550
551 static void jffs2_free_full_dirent_list(struct jffs2_full_dirent *fd)
552 {
553         struct jffs2_full_dirent *next;
554
555         while (fd) {
556                 next = fd->next;
557                 jffs2_free_full_dirent(fd);
558                 fd = next;
559         }
560 }
561
562 /* Returns first valid node after 'ref'. May return 'ref' */
563 static struct jffs2_raw_node_ref *jffs2_first_valid_node(struct jffs2_raw_node_ref *ref)
564 {
565         while (ref && ref->next_in_ino) {
566                 if (!ref_obsolete(ref))
567                         return ref;
568                 dbg_noderef("node at 0x%08x is obsoleted. Ignoring.\n", ref_offset(ref));
569                 ref = ref->next_in_ino;
570         }
571         return NULL;
572 }
573
574 /*
575  * Helper function for jffs2_get_inode_nodes().
576  * It is called every time an directory entry node is found.
577  *
578  * Returns: 0 on succes;
579  *          1 if the node should be marked obsolete;
580  *          negative error code on failure.
581  */
582 static inline int read_direntry(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
583                                 struct jffs2_raw_dirent *rd, size_t read,
584                                 struct jffs2_readinode_info *rii)
585 {
586         struct jffs2_full_dirent *fd;
587         uint32_t crc;
588
589         /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
590         BUG_ON(ref_obsolete(ref));
591
592         crc = crc32(0, rd, sizeof(*rd) - 8);
593         if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
594                 JFFS2_NOTICE("header CRC failed on dirent node at %#08x: read %#08x, calculated %#08x\n",
595                              ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
596                 jffs2_mark_node_obsolete(c, ref);
597                 return 0;
598         }
599
600         /* If we've never checked the CRCs on this node, check them now */
601         if (ref_flags(ref) == REF_UNCHECKED) {
602                 struct jffs2_eraseblock *jeb;
603                 int len;
604
605                 /* Sanity check */
606                 if (unlikely(PAD((rd->nsize + sizeof(*rd))) != PAD(je32_to_cpu(rd->totlen)))) {
607                         JFFS2_ERROR("illegal nsize in node at %#08x: nsize %#02x, totlen %#04x\n",
608                                     ref_offset(ref), rd->nsize, je32_to_cpu(rd->totlen));
609                         jffs2_mark_node_obsolete(c, ref);
610                         return 0;
611                 }
612
613                 jeb = &c->blocks[ref->flash_offset / c->sector_size];
614                 len = ref_totlen(c, jeb, ref);
615
616                 spin_lock(&c->erase_completion_lock);
617                 jeb->used_size += len;
618                 jeb->unchecked_size -= len;
619                 c->used_size += len;
620                 c->unchecked_size -= len;
621                 ref->flash_offset = ref_offset(ref) | REF_PRISTINE;
622                 spin_unlock(&c->erase_completion_lock);
623         }
624
625         fd = jffs2_alloc_full_dirent(rd->nsize + 1);
626         if (unlikely(!fd))
627                 return -ENOMEM;
628
629         fd->raw = ref;
630         fd->version = je32_to_cpu(rd->version);
631         fd->ino = je32_to_cpu(rd->ino);
632         fd->type = rd->type;
633
634         if (fd->version > rii->highest_version)
635                 rii->highest_version = fd->version;
636
637         /* Pick out the mctime of the latest dirent */
638         if(fd->version > rii->mctime_ver && je32_to_cpu(rd->mctime)) {
639                 rii->mctime_ver = fd->version;
640                 rii->latest_mctime = je32_to_cpu(rd->mctime);
641         }
642
643         /*
644          * Copy as much of the name as possible from the raw
645          * dirent we've already read from the flash.
646          */
647         if (read > sizeof(*rd))
648                 memcpy(&fd->name[0], &rd->name[0],
649                        min_t(uint32_t, rd->nsize, (read - sizeof(*rd)) ));
650
651         /* Do we need to copy any more of the name directly from the flash? */
652         if (rd->nsize + sizeof(*rd) > read) {
653                 /* FIXME: point() */
654                 int err;
655                 int already = read - sizeof(*rd);
656
657                 err = jffs2_flash_read(c, (ref_offset(ref)) + read,
658                                 rd->nsize - already, &read, &fd->name[already]);
659                 if (unlikely(read != rd->nsize - already) && likely(!err))
660                         return -EIO;
661
662                 if (unlikely(err)) {
663                         JFFS2_ERROR("read remainder of name: error %d\n", err);
664                         jffs2_free_full_dirent(fd);
665                         return -EIO;
666                 }
667         }
668
669         fd->nhash = full_name_hash(fd->name, rd->nsize);
670         fd->next = NULL;
671         fd->name[rd->nsize] = '\0';
672
673         /*
674          * Wheee. We now have a complete jffs2_full_dirent structure, with
675          * the name in it and everything. Link it into the list
676          */
677         jffs2_add_fd_to_list(c, fd, &rii->fds);
678
679         return 0;
680 }
681
682 /*
683  * Helper function for jffs2_get_inode_nodes().
684  * It is called every time an inode node is found.
685  *
686  * Returns: 0 on success;
687  *          1 if the node should be marked obsolete;
688  *          negative error code on failure.
689  */
690 static inline int read_dnode(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
691                              struct jffs2_raw_inode *rd, int rdlen,
692                              struct jffs2_readinode_info *rii)
693 {
694         struct jffs2_tmp_dnode_info *tn;
695         uint32_t len, csize;
696         int ret = 1;
697         uint32_t crc;
698
699         /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
700         BUG_ON(ref_obsolete(ref));
701
702         crc = crc32(0, rd, sizeof(*rd) - 8);
703         if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
704                 JFFS2_NOTICE("node CRC failed on dnode at %#08x: read %#08x, calculated %#08x\n",
705                              ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
706                 jffs2_mark_node_obsolete(c, ref);
707                 return 0;
708         }
709
710         tn = jffs2_alloc_tmp_dnode_info();
711         if (!tn) {
712                 JFFS2_ERROR("failed to allocate tn (%zu bytes).\n", sizeof(*tn));
713                 return -ENOMEM;
714         }
715
716         tn->partial_crc = 0;
717         csize = je32_to_cpu(rd->csize);
718
719         /* If we've never checked the CRCs on this node, check them now */
720         if (ref_flags(ref) == REF_UNCHECKED) {
721
722                 /* Sanity checks */
723                 if (unlikely(je32_to_cpu(rd->offset) > je32_to_cpu(rd->isize)) ||
724                     unlikely(PAD(je32_to_cpu(rd->csize) + sizeof(*rd)) != PAD(je32_to_cpu(rd->totlen)))) {
725                                 JFFS2_WARNING("inode node header CRC is corrupted at %#08x\n", ref_offset(ref));
726                                 jffs2_dbg_dump_node(c, ref_offset(ref));
727                         goto free_out;
728                 }
729
730                 if (jffs2_is_writebuffered(c) && csize != 0) {
731                         /* At this point we are supposed to check the data CRC
732                          * of our unchecked node. But thus far, we do not
733                          * know whether the node is valid or obsolete. To
734                          * figure this out, we need to walk all the nodes of
735                          * the inode and build the inode fragtree. We don't
736                          * want to spend time checking data of nodes which may
737                          * later be found to be obsolete. So we put off the full
738                          * data CRC checking until we have read all the inode
739                          * nodes and have started building the fragtree.
740                          *
741                          * The fragtree is being built starting with nodes
742                          * having the highest version number, so we'll be able
743                          * to detect whether a node is valid (i.e., it is not
744                          * overlapped by a node with higher version) or not.
745                          * And we'll be able to check only those nodes, which
746                          * are not obsolete.
747                          *
748                          * Of course, this optimization only makes sense in case
749                          * of NAND flashes (or other flashes whith
750                          * !jffs2_can_mark_obsolete()), since on NOR flashes
751                          * nodes are marked obsolete physically.
752                          *
753                          * Since NAND flashes (or other flashes with
754                          * jffs2_is_writebuffered(c)) are anyway read by
755                          * fractions of c->wbuf_pagesize, and we have just read
756                          * the node header, it is likely that the starting part
757                          * of the node data is also read when we read the
758                          * header. So we don't mind to check the CRC of the
759                          * starting part of the data of the node now, and check
760                          * the second part later (in jffs2_check_node_data()).
761                          * Of course, we will not need to re-read and re-check
762                          * the NAND page which we have just read. This is why we
763                          * read the whole NAND page at jffs2_get_inode_nodes(),
764                          * while we needed only the node header.
765                          */
766                         unsigned char *buf;
767
768                         /* 'buf' will point to the start of data */
769                         buf = (unsigned char *)rd + sizeof(*rd);
770                         /* len will be the read data length */
771                         len = min_t(uint32_t, rdlen - sizeof(*rd), csize);
772                         tn->partial_crc = crc32(0, buf, len);
773
774                         dbg_readinode("Calculates CRC (%#08x) for %d bytes, csize %d\n", tn->partial_crc, len, csize);
775
776                         /* If we actually calculated the whole data CRC
777                          * and it is wrong, drop the node. */
778                         if (len >= csize && unlikely(tn->partial_crc != je32_to_cpu(rd->data_crc))) {
779                                 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
780                                         ref_offset(ref), tn->partial_crc, je32_to_cpu(rd->data_crc));
781                                 goto free_out;
782                         }
783
784                 } else if (csize == 0) {
785                         /*
786                          * We checked the header CRC. If the node has no data, adjust
787                          * the space accounting now. For other nodes this will be done
788                          * later either when the node is marked obsolete or when its
789                          * data is checked.
790                          */
791                         struct jffs2_eraseblock *jeb;
792
793                         dbg_readinode("the node has no data.\n");
794                         jeb = &c->blocks[ref->flash_offset / c->sector_size];
795                         len = ref_totlen(c, jeb, ref);
796
797                         spin_lock(&c->erase_completion_lock);
798                         jeb->used_size += len;
799                         jeb->unchecked_size -= len;
800                         c->used_size += len;
801                         c->unchecked_size -= len;
802                         ref->flash_offset = ref_offset(ref) | REF_NORMAL;
803                         spin_unlock(&c->erase_completion_lock);
804                 }
805         }
806
807         tn->fn = jffs2_alloc_full_dnode();
808         if (!tn->fn) {
809                 JFFS2_ERROR("alloc fn failed\n");
810                 ret = -ENOMEM;
811                 goto free_out;
812         }
813
814         tn->version = je32_to_cpu(rd->version);
815         tn->fn->ofs = je32_to_cpu(rd->offset);
816         tn->data_crc = je32_to_cpu(rd->data_crc);
817         tn->csize = csize;
818         tn->fn->raw = ref;
819         tn->overlapped = 0;
820
821         if (tn->version > rii->highest_version)
822                 rii->highest_version = tn->version;
823
824         /* There was a bug where we wrote hole nodes out with
825            csize/dsize swapped. Deal with it */
826         if (rd->compr == JFFS2_COMPR_ZERO && !je32_to_cpu(rd->dsize) && csize)
827                 tn->fn->size = csize;
828         else // normal case...
829                 tn->fn->size = je32_to_cpu(rd->dsize);
830
831         dbg_readinode("dnode @%08x: ver %u, offset %#04x, dsize %#04x, csize %#04x\n",
832                   ref_offset(ref), je32_to_cpu(rd->version), je32_to_cpu(rd->offset), je32_to_cpu(rd->dsize), csize);
833
834         ret = jffs2_add_tn_to_tree(c, rii, tn);
835
836         if (ret) {
837                 jffs2_free_full_dnode(tn->fn);
838         free_out:
839                 jffs2_free_tmp_dnode_info(tn);
840                 return ret;
841         }
842 #ifdef JFFS2_DBG_READINODE_MESSAGES
843         dbg_readinode("After adding ver %d:\n", je32_to_cpu(rd->version));
844         tn = tn_first(&rii->tn_root);
845         while (tn) {
846                 dbg_readinode("%p: v %d r 0x%x-0x%x ov %d\n",
847                              tn, tn->version, tn->fn->ofs,
848                              tn->fn->ofs+tn->fn->size, tn->overlapped);
849                 tn = tn_next(tn);
850         }
851 #endif
852         return 0;
853 }
854
855 /*
856  * Helper function for jffs2_get_inode_nodes().
857  * It is called every time an unknown node is found.
858  *
859  * Returns: 0 on success;
860  *          1 if the node should be marked obsolete;
861  *          negative error code on failure.
862  */
863 static inline int read_unknown(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref, struct jffs2_unknown_node *un)
864 {
865         /* We don't mark unknown nodes as REF_UNCHECKED */
866         if (ref_flags(ref) == REF_UNCHECKED) {
867                 JFFS2_ERROR("REF_UNCHECKED but unknown node at %#08x\n",
868                             ref_offset(ref));
869                 JFFS2_ERROR("Node is {%04x,%04x,%08x,%08x}. Please report this error.\n",
870                             je16_to_cpu(un->magic), je16_to_cpu(un->nodetype),
871                             je32_to_cpu(un->totlen), je32_to_cpu(un->hdr_crc));
872                 jffs2_mark_node_obsolete(c, ref);
873                 return 0;
874         }
875
876         un->nodetype = cpu_to_je16(JFFS2_NODE_ACCURATE | je16_to_cpu(un->nodetype));
877
878         switch(je16_to_cpu(un->nodetype) & JFFS2_COMPAT_MASK) {
879
880         case JFFS2_FEATURE_INCOMPAT:
881                 JFFS2_ERROR("unknown INCOMPAT nodetype %#04X at %#08x\n",
882                             je16_to_cpu(un->nodetype), ref_offset(ref));
883                 /* EEP */
884                 BUG();
885                 break;
886
887         case JFFS2_FEATURE_ROCOMPAT:
888                 JFFS2_ERROR("unknown ROCOMPAT nodetype %#04X at %#08x\n",
889                             je16_to_cpu(un->nodetype), ref_offset(ref));
890                 BUG_ON(!(c->flags & JFFS2_SB_FLAG_RO));
891                 break;
892
893         case JFFS2_FEATURE_RWCOMPAT_COPY:
894                 JFFS2_NOTICE("unknown RWCOMPAT_COPY nodetype %#04X at %#08x\n",
895                              je16_to_cpu(un->nodetype), ref_offset(ref));
896                 break;
897
898         case JFFS2_FEATURE_RWCOMPAT_DELETE:
899                 JFFS2_NOTICE("unknown RWCOMPAT_DELETE nodetype %#04X at %#08x\n",
900                              je16_to_cpu(un->nodetype), ref_offset(ref));
901                 jffs2_mark_node_obsolete(c, ref);
902                 return 0;
903         }
904
905         return 0;
906 }
907
908 /*
909  * Helper function for jffs2_get_inode_nodes().
910  * The function detects whether more data should be read and reads it if yes.
911  *
912  * Returns: 0 on succes;
913  *          negative error code on failure.
914  */
915 static int read_more(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
916                      int needed_len, int *rdlen, unsigned char *buf)
917 {
918         int err, to_read = needed_len - *rdlen;
919         size_t retlen;
920         uint32_t offs;
921
922         if (jffs2_is_writebuffered(c)) {
923                 int rem = to_read % c->wbuf_pagesize;
924
925                 if (rem)
926                         to_read += c->wbuf_pagesize - rem;
927         }
928
929         /* We need to read more data */
930         offs = ref_offset(ref) + *rdlen;
931
932         dbg_readinode("read more %d bytes\n", to_read);
933
934         err = jffs2_flash_read(c, offs, to_read, &retlen, buf + *rdlen);
935         if (err) {
936                 JFFS2_ERROR("can not read %d bytes from 0x%08x, "
937                         "error code: %d.\n", to_read, offs, err);
938                 return err;
939         }
940
941         if (retlen < to_read) {
942                 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n",
943                                 offs, retlen, to_read);
944                 return -EIO;
945         }
946
947         *rdlen += to_read;
948         return 0;
949 }
950
951 /* Get tmp_dnode_info and full_dirent for all non-obsolete nodes associated
952    with this ino. Perform a preliminary ordering on data nodes, throwing away
953    those which are completely obsoleted by newer ones. The naïve approach we
954    use to take of just returning them _all_ in version order will cause us to
955    run out of memory in certain degenerate cases. */
956 static int jffs2_get_inode_nodes(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
957                                  struct jffs2_readinode_info *rii)
958 {
959         struct jffs2_raw_node_ref *ref, *valid_ref;
960         unsigned char *buf = NULL;
961         union jffs2_node_union *node;
962         size_t retlen;
963         int len, err;
964
965         rii->mctime_ver = 0;
966
967         dbg_readinode("ino #%u\n", f->inocache->ino);
968
969         /* FIXME: in case of NOR and available ->point() this
970          * needs to be fixed. */
971         len = sizeof(union jffs2_node_union) + c->wbuf_pagesize;
972         buf = kmalloc(len, GFP_KERNEL);
973         if (!buf)
974                 return -ENOMEM;
975
976         spin_lock(&c->erase_completion_lock);
977         valid_ref = jffs2_first_valid_node(f->inocache->nodes);
978         if (!valid_ref && f->inocache->ino != 1)
979                 JFFS2_WARNING("Eep. No valid nodes for ino #%u.\n", f->inocache->ino);
980         while (valid_ref) {
981                 /* We can hold a pointer to a non-obsolete node without the spinlock,
982                    but _obsolete_ nodes may disappear at any time, if the block
983                    they're in gets erased. So if we mark 'ref' obsolete while we're
984                    not holding the lock, it can go away immediately. For that reason,
985                    we find the next valid node first, before processing 'ref'.
986                 */
987                 ref = valid_ref;
988                 valid_ref = jffs2_first_valid_node(ref->next_in_ino);
989                 spin_unlock(&c->erase_completion_lock);
990
991                 cond_resched();
992
993                 /*
994                  * At this point we don't know the type of the node we're going
995                  * to read, so we do not know the size of its header. In order
996                  * to minimize the amount of flash IO we assume the header is
997                  * of size = JFFS2_MIN_NODE_HEADER.
998                  */
999                 len = JFFS2_MIN_NODE_HEADER;
1000                 if (jffs2_is_writebuffered(c)) {
1001                         int end, rem;
1002
1003                         /*
1004                          * We are about to read JFFS2_MIN_NODE_HEADER bytes,
1005                          * but this flash has some minimal I/O unit. It is
1006                          * possible that we'll need to read more soon, so read
1007                          * up to the next min. I/O unit, in order not to
1008                          * re-read the same min. I/O unit twice.
1009                          */
1010                         end = ref_offset(ref) + len;
1011                         rem = end % c->wbuf_pagesize;
1012                         if (rem)
1013                                 end += c->wbuf_pagesize - rem;
1014                         len = end - ref_offset(ref);
1015                 }
1016
1017                 dbg_readinode("read %d bytes at %#08x(%d).\n", len, ref_offset(ref), ref_flags(ref));
1018
1019                 /* FIXME: point() */
1020                 err = jffs2_flash_read(c, ref_offset(ref), len, &retlen, buf);
1021                 if (err) {
1022                         JFFS2_ERROR("can not read %d bytes from 0x%08x, " "error code: %d.\n", len, ref_offset(ref), err);
1023                         goto free_out;
1024                 }
1025
1026                 if (retlen < len) {
1027                         JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n", ref_offset(ref), retlen, len);
1028                         err = -EIO;
1029                         goto free_out;
1030                 }
1031
1032                 node = (union jffs2_node_union *)buf;
1033
1034                 /* No need to mask in the valid bit; it shouldn't be invalid */
1035                 if (je32_to_cpu(node->u.hdr_crc) != crc32(0, node, sizeof(node->u)-4)) {
1036                         JFFS2_NOTICE("Node header CRC failed at %#08x. {%04x,%04x,%08x,%08x}\n",
1037                                      ref_offset(ref), je16_to_cpu(node->u.magic),
1038                                      je16_to_cpu(node->u.nodetype),
1039                                      je32_to_cpu(node->u.totlen),
1040                                      je32_to_cpu(node->u.hdr_crc));
1041                         jffs2_dbg_dump_node(c, ref_offset(ref));
1042                         jffs2_mark_node_obsolete(c, ref);
1043                         goto cont;
1044                 }
1045                 if (je16_to_cpu(node->u.magic) != JFFS2_MAGIC_BITMASK) {
1046                         /* Not a JFFS2 node, whinge and move on */
1047                         JFFS2_NOTICE("Wrong magic bitmask 0x%04x in node header at %#08x.\n",
1048                                      je16_to_cpu(node->u.magic), ref_offset(ref));
1049                         jffs2_mark_node_obsolete(c, ref);
1050                         goto cont;
1051                 }
1052
1053                 switch (je16_to_cpu(node->u.nodetype)) {
1054
1055                 case JFFS2_NODETYPE_DIRENT:
1056
1057                         if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_dirent)) {
1058                                 err = read_more(c, ref, sizeof(struct jffs2_raw_dirent), &len, buf);
1059                                 if (unlikely(err))
1060                                         goto free_out;
1061                         }
1062
1063                         err = read_direntry(c, ref, &node->d, retlen, rii);
1064                         if (unlikely(err))
1065                                 goto free_out;
1066
1067                         break;
1068
1069                 case JFFS2_NODETYPE_INODE:
1070
1071                         if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_inode)) {
1072                                 err = read_more(c, ref, sizeof(struct jffs2_raw_inode), &len, buf);
1073                                 if (unlikely(err))
1074                                         goto free_out;
1075                         }
1076
1077                         err = read_dnode(c, ref, &node->i, len, rii);
1078                         if (unlikely(err))
1079                                 goto free_out;
1080
1081                         break;
1082
1083                 default:
1084                         if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_unknown_node)) {
1085                                 err = read_more(c, ref, sizeof(struct jffs2_unknown_node), &len, buf);
1086                                 if (unlikely(err))
1087                                         goto free_out;
1088                         }
1089
1090                         err = read_unknown(c, ref, &node->u);
1091                         if (err == 1) {
1092                                 jffs2_mark_node_obsolete(c, ref);
1093                                 break;
1094                         } else if (unlikely(err))
1095                                 goto free_out;
1096
1097                 }
1098         cont:
1099                 spin_lock(&c->erase_completion_lock);
1100         }
1101
1102         spin_unlock(&c->erase_completion_lock);
1103         kfree(buf);
1104
1105         f->highest_version = rii->highest_version;
1106
1107         dbg_readinode("nodes of inode #%u were read, the highest version is %u, latest_mctime %u, mctime_ver %u.\n",
1108                       f->inocache->ino, rii->highest_version, rii->latest_mctime,
1109                       rii->mctime_ver);
1110         return 0;
1111
1112  free_out:
1113         jffs2_free_tmp_dnode_info_list(&rii->tn_root);
1114         jffs2_free_full_dirent_list(rii->fds);
1115         rii->fds = NULL;
1116         kfree(buf);
1117         return err;
1118 }
1119
1120 static int jffs2_do_read_inode_internal(struct jffs2_sb_info *c,
1121                                         struct jffs2_inode_info *f,
1122                                         struct jffs2_raw_inode *latest_node)
1123 {
1124         struct jffs2_readinode_info rii;
1125         uint32_t crc, new_size;
1126         size_t retlen;
1127         int ret;
1128
1129         dbg_readinode("ino #%u nlink is %d\n", f->inocache->ino, f->inocache->nlink);
1130
1131         memset(&rii, 0, sizeof(rii));
1132
1133         /* Grab all nodes relevant to this ino */
1134         ret = jffs2_get_inode_nodes(c, f, &rii);
1135
1136         if (ret) {
1137                 JFFS2_ERROR("cannot read nodes for ino %u, returned error is %d\n", f->inocache->ino, ret);
1138                 if (f->inocache->state == INO_STATE_READING)
1139                         jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1140                 return ret;
1141         }
1142
1143         ret = jffs2_build_inode_fragtree(c, f, &rii);
1144         if (ret) {
1145                 JFFS2_ERROR("Failed to build final fragtree for inode #%u: error %d\n",
1146                             f->inocache->ino, ret);
1147                 if (f->inocache->state == INO_STATE_READING)
1148                         jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1149                 jffs2_free_tmp_dnode_info_list(&rii.tn_root);
1150                 /* FIXME: We could at least crc-check them all */
1151                 if (rii.mdata_tn) {
1152                         jffs2_free_full_dnode(rii.mdata_tn->fn);
1153                         jffs2_free_tmp_dnode_info(rii.mdata_tn);
1154                         rii.mdata_tn = NULL;
1155                 }
1156                 return ret;
1157         }
1158
1159         if (rii.mdata_tn) {
1160                 if (rii.mdata_tn->fn->raw == rii.latest_ref) {
1161                         f->metadata = rii.mdata_tn->fn;
1162                         jffs2_free_tmp_dnode_info(rii.mdata_tn);
1163                 } else {
1164                         jffs2_kill_tn(c, rii.mdata_tn);
1165                 }
1166                 rii.mdata_tn = NULL;
1167         }
1168
1169         f->dents = rii.fds;
1170
1171         jffs2_dbg_fragtree_paranoia_check_nolock(f);
1172
1173         if (unlikely(!rii.latest_ref)) {
1174                 /* No data nodes for this inode. */
1175                 if (f->inocache->ino != 1) {
1176                         JFFS2_WARNING("no data nodes found for ino #%u\n", f->inocache->ino);
1177                         if (!rii.fds) {
1178                                 if (f->inocache->state == INO_STATE_READING)
1179                                         jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1180                                 return -EIO;
1181                         }
1182                         JFFS2_NOTICE("but it has children so we fake some modes for it\n");
1183                 }
1184                 latest_node->mode = cpu_to_jemode(S_IFDIR|S_IRUGO|S_IWUSR|S_IXUGO);
1185                 latest_node->version = cpu_to_je32(0);
1186                 latest_node->atime = latest_node->ctime = latest_node->mtime = cpu_to_je32(0);
1187                 latest_node->isize = cpu_to_je32(0);
1188                 latest_node->gid = cpu_to_je16(0);
1189                 latest_node->uid = cpu_to_je16(0);
1190                 if (f->inocache->state == INO_STATE_READING)
1191                         jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1192                 return 0;
1193         }
1194
1195         ret = jffs2_flash_read(c, ref_offset(rii.latest_ref), sizeof(*latest_node), &retlen, (void *)latest_node);
1196         if (ret || retlen != sizeof(*latest_node)) {
1197                 JFFS2_ERROR("failed to read from flash: error %d, %zd of %zd bytes read\n",
1198                         ret, retlen, sizeof(*latest_node));
1199                 /* FIXME: If this fails, there seems to be a memory leak. Find it. */
1200                 up(&f->sem);
1201                 jffs2_do_clear_inode(c, f);
1202                 return ret?ret:-EIO;
1203         }
1204
1205         crc = crc32(0, latest_node, sizeof(*latest_node)-8);
1206         if (crc != je32_to_cpu(latest_node->node_crc)) {
1207                 JFFS2_ERROR("CRC failed for read_inode of inode %u at physical location 0x%x\n",
1208                         f->inocache->ino, ref_offset(rii.latest_ref));
1209                 up(&f->sem);
1210                 jffs2_do_clear_inode(c, f);
1211                 return -EIO;
1212         }
1213
1214         switch(jemode_to_cpu(latest_node->mode) & S_IFMT) {
1215         case S_IFDIR:
1216                 if (rii.mctime_ver > je32_to_cpu(latest_node->version)) {
1217                         /* The times in the latest_node are actually older than
1218                            mctime in the latest dirent. Cheat. */
1219                         latest_node->ctime = latest_node->mtime = cpu_to_je32(rii.latest_mctime);
1220                 }
1221                 break;
1222
1223
1224         case S_IFREG:
1225                 /* If it was a regular file, truncate it to the latest node's isize */
1226                 new_size = jffs2_truncate_fragtree(c, &f->fragtree, je32_to_cpu(latest_node->isize));
1227                 if (new_size != je32_to_cpu(latest_node->isize)) {
1228                         JFFS2_WARNING("Truncating ino #%u to %d bytes failed because it only had %d bytes to start with!\n",
1229                                       f->inocache->ino, je32_to_cpu(latest_node->isize), new_size);
1230                         latest_node->isize = cpu_to_je32(new_size);
1231                 }
1232                 break;
1233
1234         case S_IFLNK:
1235                 /* Hack to work around broken isize in old symlink code.
1236                    Remove this when dwmw2 comes to his senses and stops
1237                    symlinks from being an entirely gratuitous special
1238                    case. */
1239                 if (!je32_to_cpu(latest_node->isize))
1240                         latest_node->isize = latest_node->dsize;
1241
1242                 if (f->inocache->state != INO_STATE_CHECKING) {
1243                         /* Symlink's inode data is the target path. Read it and
1244                          * keep in RAM to facilitate quick follow symlink
1245                          * operation. */
1246                         f->target = kmalloc(je32_to_cpu(latest_node->csize) + 1, GFP_KERNEL);
1247                         if (!f->target) {
1248                                 JFFS2_ERROR("can't allocate %d bytes of memory for the symlink target path cache\n", je32_to_cpu(latest_node->csize));
1249                                 up(&f->sem);
1250                                 jffs2_do_clear_inode(c, f);
1251                                 return -ENOMEM;
1252                         }
1253
1254                         ret = jffs2_flash_read(c, ref_offset(rii.latest_ref) + sizeof(*latest_node),
1255                                                 je32_to_cpu(latest_node->csize), &retlen, (char *)f->target);
1256
1257                         if (ret  || retlen != je32_to_cpu(latest_node->csize)) {
1258                                 if (retlen != je32_to_cpu(latest_node->csize))
1259                                         ret = -EIO;
1260                                 kfree(f->target);
1261                                 f->target = NULL;
1262                                 up(&f->sem);
1263                                 jffs2_do_clear_inode(c, f);
1264                                 return -ret;
1265                         }
1266
1267                         f->target[je32_to_cpu(latest_node->csize)] = '\0';
1268                         dbg_readinode("symlink's target '%s' cached\n", f->target);
1269                 }
1270
1271                 /* fall through... */
1272
1273         case S_IFBLK:
1274         case S_IFCHR:
1275                 /* Certain inode types should have only one data node, and it's
1276                    kept as the metadata node */
1277                 if (f->metadata) {
1278                         JFFS2_ERROR("Argh. Special inode #%u with mode 0%o had metadata node\n",
1279                                f->inocache->ino, jemode_to_cpu(latest_node->mode));
1280                         up(&f->sem);
1281                         jffs2_do_clear_inode(c, f);
1282                         return -EIO;
1283                 }
1284                 if (!frag_first(&f->fragtree)) {
1285                         JFFS2_ERROR("Argh. Special inode #%u with mode 0%o has no fragments\n",
1286                                f->inocache->ino, jemode_to_cpu(latest_node->mode));
1287                         up(&f->sem);
1288                         jffs2_do_clear_inode(c, f);
1289                         return -EIO;
1290                 }
1291                 /* ASSERT: f->fraglist != NULL */
1292                 if (frag_next(frag_first(&f->fragtree))) {
1293                         JFFS2_ERROR("Argh. Special inode #%u with mode 0x%x had more than one node\n",
1294                                f->inocache->ino, jemode_to_cpu(latest_node->mode));
1295                         /* FIXME: Deal with it - check crc32, check for duplicate node, check times and discard the older one */
1296                         up(&f->sem);
1297                         jffs2_do_clear_inode(c, f);
1298                         return -EIO;
1299                 }
1300                 /* OK. We're happy */
1301                 f->metadata = frag_first(&f->fragtree)->node;
1302                 jffs2_free_node_frag(frag_first(&f->fragtree));
1303                 f->fragtree = RB_ROOT;
1304                 break;
1305         }
1306         if (f->inocache->state == INO_STATE_READING)
1307                 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1308
1309         return 0;
1310 }
1311
1312 /* Scan the list of all nodes present for this ino, build map of versions, etc. */
1313 int jffs2_do_read_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
1314                         uint32_t ino, struct jffs2_raw_inode *latest_node)
1315 {
1316         dbg_readinode("read inode #%u\n", ino);
1317
1318  retry_inocache:
1319         spin_lock(&c->inocache_lock);
1320         f->inocache = jffs2_get_ino_cache(c, ino);
1321
1322         if (f->inocache) {
1323                 /* Check its state. We may need to wait before we can use it */
1324                 switch(f->inocache->state) {
1325                 case INO_STATE_UNCHECKED:
1326                 case INO_STATE_CHECKEDABSENT:
1327                         f->inocache->state = INO_STATE_READING;
1328                         break;
1329
1330                 case INO_STATE_CHECKING:
1331                 case INO_STATE_GC:
1332                         /* If it's in either of these states, we need
1333                            to wait for whoever's got it to finish and
1334                            put it back. */
1335                         dbg_readinode("waiting for ino #%u in state %d\n", ino, f->inocache->state);
1336                         sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
1337                         goto retry_inocache;
1338
1339                 case INO_STATE_READING:
1340                 case INO_STATE_PRESENT:
1341                         /* Eep. This should never happen. It can
1342                         happen if Linux calls read_inode() again
1343                         before clear_inode() has finished though. */
1344                         JFFS2_ERROR("Eep. Trying to read_inode #%u when it's already in state %d!\n", ino, f->inocache->state);
1345                         /* Fail. That's probably better than allowing it to succeed */
1346                         f->inocache = NULL;
1347                         break;
1348
1349                 default:
1350                         BUG();
1351                 }
1352         }
1353         spin_unlock(&c->inocache_lock);
1354
1355         if (!f->inocache && ino == 1) {
1356                 /* Special case - no root inode on medium */
1357                 f->inocache = jffs2_alloc_inode_cache();
1358                 if (!f->inocache) {
1359                         JFFS2_ERROR("cannot allocate inocache for root inode\n");
1360                         return -ENOMEM;
1361                 }
1362                 dbg_readinode("creating inocache for root inode\n");
1363                 memset(f->inocache, 0, sizeof(struct jffs2_inode_cache));
1364                 f->inocache->ino = f->inocache->nlink = 1;
1365                 f->inocache->nodes = (struct jffs2_raw_node_ref *)f->inocache;
1366                 f->inocache->state = INO_STATE_READING;
1367                 jffs2_add_ino_cache(c, f->inocache);
1368         }
1369         if (!f->inocache) {
1370                 JFFS2_ERROR("requestied to read an nonexistent ino %u\n", ino);
1371                 return -ENOENT;
1372         }
1373
1374         return jffs2_do_read_inode_internal(c, f, latest_node);
1375 }
1376
1377 int jffs2_do_crccheck_inode(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic)
1378 {
1379         struct jffs2_raw_inode n;
1380         struct jffs2_inode_info *f = kzalloc(sizeof(*f), GFP_KERNEL);
1381         int ret;
1382
1383         if (!f)
1384                 return -ENOMEM;
1385
1386         init_MUTEX_LOCKED(&f->sem);
1387         f->inocache = ic;
1388
1389         ret = jffs2_do_read_inode_internal(c, f, &n);
1390         if (!ret) {
1391                 up(&f->sem);
1392                 jffs2_do_clear_inode(c, f);
1393         }
1394         kfree (f);
1395         return ret;
1396 }
1397
1398 void jffs2_do_clear_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f)
1399 {
1400         struct jffs2_full_dirent *fd, *fds;
1401         int deleted;
1402
1403         jffs2_clear_acl(f);
1404         jffs2_xattr_delete_inode(c, f->inocache);
1405         down(&f->sem);
1406         deleted = f->inocache && !f->inocache->nlink;
1407
1408         if (f->inocache && f->inocache->state != INO_STATE_CHECKING)
1409                 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CLEARING);
1410
1411         if (f->metadata) {
1412                 if (deleted)
1413                         jffs2_mark_node_obsolete(c, f->metadata->raw);
1414                 jffs2_free_full_dnode(f->metadata);
1415         }
1416
1417         jffs2_kill_fragtree(&f->fragtree, deleted?c:NULL);
1418
1419         if (f->target) {
1420                 kfree(f->target);
1421                 f->target = NULL;
1422         }
1423
1424         fds = f->dents;
1425         while(fds) {
1426                 fd = fds;
1427                 fds = fd->next;
1428                 jffs2_free_full_dirent(fd);
1429         }
1430
1431         if (f->inocache && f->inocache->state != INO_STATE_CHECKING) {
1432                 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1433                 if (f->inocache->nodes == (void *)f->inocache)
1434                         jffs2_del_ino_cache(c, f->inocache);
1435         }
1436
1437         up(&f->sem);
1438 }