2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright © 2001-2007 Red Hat, Inc.
6 * Created by David Woodhouse <dwmw2@infradead.org>
8 * For licensing information, see the file 'LICENCE' in this directory.
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
16 #include <linux/crc32.h>
17 #include <linux/pagemap.h>
18 #include <linux/mtd/mtd.h>
19 #include <linux/compiler.h>
23 * Check the data CRC of the node.
25 * Returns: 0 if the data CRC is correct;
27 * error code if an error occured.
29 static int check_node_data(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
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;
38 BUG_ON(tn->csize == 0);
40 /* Calculate how many bytes were already checked */
41 ofs = ref_offset(ref) + sizeof(struct jffs2_raw_inode);
44 if (jffs2_is_writebuffered(c)) {
45 int adj = ofs % c->wbuf_pagesize;
47 adj = c->wbuf_pagesize - adj;
49 if (adj >= 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);
59 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",
60 ref_offset(ref), tn->csize, tn->partial_crc, tn->data_crc, ofs - len, ofs, len);
63 /* TODO: instead, incapsulate point() stuff to jffs2_flash_read(),
64 * adding and jffs2_flash_read_end() interface. */
66 err = c->mtd->point(c->mtd, ofs, len, &retlen, &buffer);
67 if (!err && retlen < len) {
68 JFFS2_WARNING("MTD point returned len too short: %zu instead of %u.\n", retlen, tn->csize);
69 c->mtd->unpoint(c->mtd, buffer, ofs, retlen);
71 JFFS2_WARNING("MTD point failed: error code %d.\n", err);
73 pointed = 1; /* succefully pointed to device */
78 buffer = kmalloc(len, GFP_KERNEL);
79 if (unlikely(!buffer))
82 /* TODO: this is very frequent pattern, make it a separate
84 err = jffs2_flash_read(c, ofs, len, &retlen, buffer);
86 JFFS2_ERROR("can not read %d bytes from 0x%08x, error code: %d.\n", len, ofs, err);
91 JFFS2_ERROR("short read at %#08x: %zd instead of %d.\n", ofs, retlen, len);
97 /* Continue calculating CRC */
98 crc = crc32(tn->partial_crc, buffer, len);
103 c->mtd->unpoint(c->mtd, buffer, ofs, len);
106 if (crc != tn->data_crc) {
107 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
108 ref_offset(ref), tn->data_crc, crc);
113 jeb = &c->blocks[ref->flash_offset / c->sector_size];
114 len = ref_totlen(c, jeb, ref);
115 /* If it should be REF_NORMAL, it'll get marked as such when
116 we build the fragtree, shortly. No need to worry about GC
117 moving it while it's marked REF_PRISTINE -- GC won't happen
118 till we've finished checking every inode anyway. */
119 ref->flash_offset |= REF_PRISTINE;
121 * Mark the node as having been checked and fix the
122 * accounting accordingly.
124 spin_lock(&c->erase_completion_lock);
125 jeb->used_size += len;
126 jeb->unchecked_size -= len;
128 c->unchecked_size -= len;
129 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
130 spin_unlock(&c->erase_completion_lock);
139 c->mtd->unpoint(c->mtd, buffer, ofs, len);
145 * Helper function for jffs2_add_older_frag_to_fragtree().
147 * Checks the node if we are in the checking stage.
149 static int check_tn_node(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
153 BUG_ON(ref_obsolete(tn->fn->raw));
155 /* We only check the data CRC of unchecked nodes */
156 if (ref_flags(tn->fn->raw) != REF_UNCHECKED)
159 dbg_readinode("check node %#04x-%#04x, phys offs %#08x\n",
160 tn->fn->ofs, tn->fn->ofs + tn->fn->size, ref_offset(tn->fn->raw));
162 ret = check_node_data(c, tn);
163 if (unlikely(ret < 0)) {
164 JFFS2_ERROR("check_node_data() returned error: %d.\n",
166 } else if (unlikely(ret > 0)) {
167 dbg_readinode("CRC error, mark it obsolete.\n");
168 jffs2_mark_node_obsolete(c, tn->fn->raw);
174 static struct jffs2_tmp_dnode_info *jffs2_lookup_tn(struct rb_root *tn_root, uint32_t offset)
176 struct rb_node *next;
177 struct jffs2_tmp_dnode_info *tn = NULL;
179 dbg_readinode("root %p, offset %d\n", tn_root, offset);
181 next = tn_root->rb_node;
184 tn = rb_entry(next, struct jffs2_tmp_dnode_info, rb);
186 if (tn->fn->ofs < offset)
187 next = tn->rb.rb_right;
188 else if (tn->fn->ofs >= offset)
189 next = tn->rb.rb_left;
198 static void jffs2_kill_tn(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
200 jffs2_mark_node_obsolete(c, tn->fn->raw);
201 jffs2_free_full_dnode(tn->fn);
202 jffs2_free_tmp_dnode_info(tn);
205 * This function is used when we read an inode. Data nodes arrive in
206 * arbitrary order -- they may be older or newer than the nodes which
207 * are already in the tree. Where overlaps occur, the older node can
208 * be discarded as long as the newer passes the CRC check. We don't
209 * bother to keep track of holes in this rbtree, and neither do we deal
210 * with frags -- we can have multiple entries starting at the same
211 * offset, and the one with the smallest length will come first in the
214 * Returns 0 if the node was handled (including marking it obsolete)
215 * < 0 an if error occurred
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)
221 uint32_t fn_end = tn->fn->ofs + tn->fn->size;
222 struct jffs2_tmp_dnode_info *this;
224 dbg_readinode("insert fragment %#04x-%#04x, ver %u at %08x\n", tn->fn->ofs, fn_end, tn->version, ref_offset(tn->fn->raw));
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
232 if (rii->mdata_tn->version < tn->version) {
233 /* We had a candidate mdata node already */
234 dbg_readinode("kill old mdata with ver %d\n", rii->mdata_tn->version);
235 jffs2_kill_tn(c, rii->mdata_tn);
237 dbg_readinode("kill new mdata with ver %d (older than existing %d\n",
238 tn->version, rii->mdata_tn->version);
239 jffs2_kill_tn(c, tn);
244 dbg_readinode("keep new mdata with ver %d\n", tn->version);
248 /* Find the earliest node which _may_ be relevant to this one */
249 this = jffs2_lookup_tn(&rii->tn_root, tn->fn->ofs);
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 (this->overlapped)
254 this = tn_prev(this);
256 /* First node should never be marked overlapped */
258 dbg_readinode("'this' found %#04x-%#04x (%s)\n", this->fn->ofs, this->fn->ofs + this->fn->size, this->fn ? "data" : "hole");
262 if (this->fn->ofs > fn_end)
264 dbg_readinode("Ponder this ver %d, 0x%x-0x%x\n",
265 this->version, this->fn->ofs, this->fn->size);
267 if (this->version == tn->version) {
268 /* Version number collision means REF_PRISTINE GC. Accept either of them
269 as long as the CRC is correct. Check the one we have already... */
270 if (!check_tn_node(c, this)) {
271 /* The one we already had was OK. Keep it and throw away the new one */
272 dbg_readinode("Like old node. Throw away new\n");
273 jffs2_kill_tn(c, tn);
276 /* Who cares if the new one is good; keep it for now anyway. */
277 dbg_readinode("Like new node. Throw away old\n");
278 rb_replace_node(&this->rb, &tn->rb, &rii->tn_root);
279 jffs2_kill_tn(c, this);
280 /* Same overlapping from in front and behind */
284 if (this->version < tn->version &&
285 this->fn->ofs >= tn->fn->ofs &&
286 this->fn->ofs + this->fn->size <= fn_end) {
287 /* New node entirely overlaps 'this' */
288 if (check_tn_node(c, tn)) {
289 dbg_readinode("new node bad CRC\n");
290 jffs2_kill_tn(c, tn);
293 /* ... and is good. Kill 'this' and any subsequent nodes which are also overlapped */
294 while (this && this->fn->ofs + this->fn->size <= fn_end) {
295 struct jffs2_tmp_dnode_info *next = tn_next(this);
296 if (this->version < tn->version) {
297 tn_erase(this, &rii->tn_root);
298 dbg_readinode("Kill overlapped ver %d, 0x%x-0x%x\n",
299 this->version, this->fn->ofs,
300 this->fn->ofs+this->fn->size);
301 jffs2_kill_tn(c, this);
305 dbg_readinode("Done killing overlapped nodes\n");
308 if (this->version > tn->version &&
309 this->fn->ofs <= tn->fn->ofs &&
310 this->fn->ofs+this->fn->size >= fn_end) {
311 /* New node entirely overlapped by 'this' */
312 if (!check_tn_node(c, this)) {
313 dbg_readinode("Good CRC on old node. Kill new\n");
314 jffs2_kill_tn(c, tn);
317 /* ... but 'this' was bad. Replace it... */
318 dbg_readinode("Bad CRC on old overlapping node. Kill it\n");
319 tn_erase(this, &rii->tn_root);
320 jffs2_kill_tn(c, this);
324 this = tn_next(this);
327 /* We neither completely obsoleted nor were completely
328 obsoleted by an earlier node. Insert into the tree */
330 struct rb_node *parent;
331 struct rb_node **link = &rii->tn_root.rb_node;
332 struct jffs2_tmp_dnode_info *insert_point = NULL;
336 insert_point = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
337 if (tn->fn->ofs > insert_point->fn->ofs)
338 link = &insert_point->rb.rb_right;
339 else if (tn->fn->ofs < insert_point->fn->ofs ||
340 tn->fn->size < insert_point->fn->size)
341 link = &insert_point->rb.rb_left;
343 link = &insert_point->rb.rb_right;
345 rb_link_node(&tn->rb, &insert_point->rb, link);
346 rb_insert_color(&tn->rb, &rii->tn_root);
349 /* If there's anything behind that overlaps us, note it */
353 if (this->fn->ofs + this->fn->size > tn->fn->ofs) {
354 dbg_readinode("Node is overlapped by %p (v %d, 0x%x-0x%x)\n",
355 this, this->version, this->fn->ofs,
356 this->fn->ofs+this->fn->size);
360 if (!this->overlapped)
362 this = tn_prev(this);
366 /* If the new node overlaps anything ahead, note it */
368 while (this && this->fn->ofs < fn_end) {
369 this->overlapped = 1;
370 dbg_readinode("Node ver %d, 0x%x-0x%x is overlapped\n",
371 this->version, this->fn->ofs,
372 this->fn->ofs+this->fn->size);
373 this = tn_next(this);
378 /* Trivial function to remove the last node in the tree. Which by definition
379 has no right-hand -- so can be removed just by making its only child (if
380 any) take its place under its parent. */
381 static void eat_last(struct rb_root *root, struct rb_node *node)
383 struct rb_node *parent = rb_parent(node);
384 struct rb_node **link;
387 BUG_ON(node->rb_right);
390 link = &root->rb_node;
391 else if (node == parent->rb_left)
392 link = &parent->rb_left;
394 link = &parent->rb_right;
396 *link = node->rb_left;
397 /* Colour doesn't matter now. Only the parent pointer. */
399 node->rb_left->rb_parent_color = node->rb_parent_color;
402 /* We put this in reverse order, so we can just use eat_last */
403 static void ver_insert(struct rb_root *ver_root, struct jffs2_tmp_dnode_info *tn)
405 struct rb_node **link = &ver_root->rb_node;
406 struct rb_node *parent = NULL;
407 struct jffs2_tmp_dnode_info *this_tn;
411 this_tn = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
413 if (tn->version > this_tn->version)
414 link = &parent->rb_left;
416 link = &parent->rb_right;
418 dbg_readinode("Link new node at %p (root is %p)\n", link, ver_root);
419 rb_link_node(&tn->rb, parent, link);
420 rb_insert_color(&tn->rb, ver_root);
423 /* Build final, normal fragtree from tn tree. It doesn't matter which order
424 we add nodes to the real fragtree, as long as they don't overlap. And
425 having thrown away the majority of overlapped nodes as we went, there
426 really shouldn't be many sets of nodes which do overlap. If we start at
427 the end, we can use the overlap markers -- we can just eat nodes which
428 aren't overlapped, and when we encounter nodes which _do_ overlap we
429 sort them all into a temporary tree in version order before replaying them. */
430 static int jffs2_build_inode_fragtree(struct jffs2_sb_info *c,
431 struct jffs2_inode_info *f,
432 struct jffs2_readinode_info *rii)
434 struct jffs2_tmp_dnode_info *pen, *last, *this;
435 struct rb_root ver_root = RB_ROOT;
436 uint32_t high_ver = 0;
439 dbg_readinode("potential mdata is ver %d at %p\n", rii->mdata_tn->version, rii->mdata_tn);
440 high_ver = rii->mdata_tn->version;
441 rii->latest_ref = rii->mdata_tn->fn->raw;
443 #ifdef JFFS2_DBG_READINODE_MESSAGES
444 this = tn_last(&rii->tn_root);
446 dbg_readinode("tn %p ver %d range 0x%x-0x%x ov %d\n", this, this->version, this->fn->ofs,
447 this->fn->ofs+this->fn->size, this->overlapped);
448 this = tn_prev(this);
451 pen = tn_last(&rii->tn_root);
452 while ((last = pen)) {
455 eat_last(&rii->tn_root, &last->rb);
456 ver_insert(&ver_root, last);
458 if (unlikely(last->overlapped))
461 /* Now we have a bunch of nodes in reverse version
462 order, in the tree at ver_root. Most of the time,
463 there'll actually be only one node in the 'tree',
465 this = tn_last(&ver_root);
468 struct jffs2_tmp_dnode_info *vers_next;
470 vers_next = tn_prev(this);
471 eat_last(&ver_root, &this->rb);
472 if (check_tn_node(c, this)) {
473 dbg_readinode("node ver %d, 0x%x-0x%x failed CRC\n",
474 this->version, this->fn->ofs,
475 this->fn->ofs+this->fn->size);
476 jffs2_kill_tn(c, this);
478 if (this->version > high_ver) {
479 /* Note that this is different from the other
480 highest_version, because this one is only
481 counting _valid_ nodes which could give the
482 latest inode metadata */
483 high_ver = this->version;
484 rii->latest_ref = this->fn->raw;
486 dbg_readinode("Add %p (v %d, 0x%x-0x%x, ov %d) to fragtree\n",
487 this, this->version, this->fn->ofs,
488 this->fn->ofs+this->fn->size, this->overlapped);
490 ret = jffs2_add_full_dnode_to_inode(c, f, this->fn);
492 /* Free the nodes in vers_root; let the caller
493 deal with the rest */
494 JFFS2_ERROR("Add node to tree failed %d\n", ret);
496 vers_next = tn_prev(this);
497 if (check_tn_node(c, this))
498 jffs2_mark_node_obsolete(c, this->fn->raw);
499 jffs2_free_full_dnode(this->fn);
500 jffs2_free_tmp_dnode_info(this);
504 eat_last(&ver_root, &vers_next->rb);
508 jffs2_free_tmp_dnode_info(this);
516 static void jffs2_free_tmp_dnode_info_list(struct rb_root *list)
518 struct rb_node *this;
519 struct jffs2_tmp_dnode_info *tn;
521 this = list->rb_node;
523 /* Now at bottom of tree */
526 this = this->rb_left;
527 else if (this->rb_right)
528 this = this->rb_right;
530 tn = rb_entry(this, struct jffs2_tmp_dnode_info, rb);
531 jffs2_free_full_dnode(tn->fn);
532 jffs2_free_tmp_dnode_info(tn);
534 this = rb_parent(this);
538 if (this->rb_left == &tn->rb)
539 this->rb_left = NULL;
540 else if (this->rb_right == &tn->rb)
541 this->rb_right = NULL;
545 list->rb_node = NULL;
548 static void jffs2_free_full_dirent_list(struct jffs2_full_dirent *fd)
550 struct jffs2_full_dirent *next;
554 jffs2_free_full_dirent(fd);
559 /* Returns first valid node after 'ref'. May return 'ref' */
560 static struct jffs2_raw_node_ref *jffs2_first_valid_node(struct jffs2_raw_node_ref *ref)
562 while (ref && ref->next_in_ino) {
563 if (!ref_obsolete(ref))
565 dbg_noderef("node at 0x%08x is obsoleted. Ignoring.\n", ref_offset(ref));
566 ref = ref->next_in_ino;
572 * Helper function for jffs2_get_inode_nodes().
573 * It is called every time an directory entry node is found.
575 * Returns: 0 on success;
576 * negative error code on failure.
578 static inline int read_direntry(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
579 struct jffs2_raw_dirent *rd, size_t read,
580 struct jffs2_readinode_info *rii)
582 struct jffs2_full_dirent *fd;
585 /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
586 BUG_ON(ref_obsolete(ref));
588 crc = crc32(0, rd, sizeof(*rd) - 8);
589 if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
590 JFFS2_NOTICE("header CRC failed on dirent node at %#08x: read %#08x, calculated %#08x\n",
591 ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
592 jffs2_mark_node_obsolete(c, ref);
596 /* If we've never checked the CRCs on this node, check them now */
597 if (ref_flags(ref) == REF_UNCHECKED) {
598 struct jffs2_eraseblock *jeb;
602 if (unlikely(PAD((rd->nsize + sizeof(*rd))) != PAD(je32_to_cpu(rd->totlen)))) {
603 JFFS2_ERROR("illegal nsize in node at %#08x: nsize %#02x, totlen %#04x\n",
604 ref_offset(ref), rd->nsize, je32_to_cpu(rd->totlen));
605 jffs2_mark_node_obsolete(c, ref);
609 jeb = &c->blocks[ref->flash_offset / c->sector_size];
610 len = ref_totlen(c, jeb, ref);
612 spin_lock(&c->erase_completion_lock);
613 jeb->used_size += len;
614 jeb->unchecked_size -= len;
616 c->unchecked_size -= len;
617 ref->flash_offset = ref_offset(ref) | dirent_node_state(rd);
618 spin_unlock(&c->erase_completion_lock);
621 fd = jffs2_alloc_full_dirent(rd->nsize + 1);
626 fd->version = je32_to_cpu(rd->version);
627 fd->ino = je32_to_cpu(rd->ino);
630 if (fd->version > rii->highest_version)
631 rii->highest_version = fd->version;
633 /* Pick out the mctime of the latest dirent */
634 if(fd->version > rii->mctime_ver && je32_to_cpu(rd->mctime)) {
635 rii->mctime_ver = fd->version;
636 rii->latest_mctime = je32_to_cpu(rd->mctime);
640 * Copy as much of the name as possible from the raw
641 * dirent we've already read from the flash.
643 if (read > sizeof(*rd))
644 memcpy(&fd->name[0], &rd->name[0],
645 min_t(uint32_t, rd->nsize, (read - sizeof(*rd)) ));
647 /* Do we need to copy any more of the name directly from the flash? */
648 if (rd->nsize + sizeof(*rd) > read) {
651 int already = read - sizeof(*rd);
653 err = jffs2_flash_read(c, (ref_offset(ref)) + read,
654 rd->nsize - already, &read, &fd->name[already]);
655 if (unlikely(read != rd->nsize - already) && likely(!err))
659 JFFS2_ERROR("read remainder of name: error %d\n", err);
660 jffs2_free_full_dirent(fd);
665 fd->nhash = full_name_hash(fd->name, rd->nsize);
667 fd->name[rd->nsize] = '\0';
670 * Wheee. We now have a complete jffs2_full_dirent structure, with
671 * the name in it and everything. Link it into the list
673 jffs2_add_fd_to_list(c, fd, &rii->fds);
679 * Helper function for jffs2_get_inode_nodes().
680 * It is called every time an inode node is found.
682 * Returns: 0 on success (possibly after marking a bad node obsolete);
683 * negative error code on failure.
685 static inline int read_dnode(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
686 struct jffs2_raw_inode *rd, int rdlen,
687 struct jffs2_readinode_info *rii)
689 struct jffs2_tmp_dnode_info *tn;
694 /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
695 BUG_ON(ref_obsolete(ref));
697 crc = crc32(0, rd, sizeof(*rd) - 8);
698 if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
699 JFFS2_NOTICE("node CRC failed on dnode at %#08x: read %#08x, calculated %#08x\n",
700 ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
701 jffs2_mark_node_obsolete(c, ref);
705 tn = jffs2_alloc_tmp_dnode_info();
707 JFFS2_ERROR("failed to allocate tn (%zu bytes).\n", sizeof(*tn));
712 csize = je32_to_cpu(rd->csize);
714 /* If we've never checked the CRCs on this node, check them now */
715 if (ref_flags(ref) == REF_UNCHECKED) {
718 if (unlikely(je32_to_cpu(rd->offset) > je32_to_cpu(rd->isize)) ||
719 unlikely(PAD(je32_to_cpu(rd->csize) + sizeof(*rd)) != PAD(je32_to_cpu(rd->totlen)))) {
720 JFFS2_WARNING("inode node header CRC is corrupted at %#08x\n", ref_offset(ref));
721 jffs2_dbg_dump_node(c, ref_offset(ref));
722 jffs2_mark_node_obsolete(c, ref);
726 if (jffs2_is_writebuffered(c) && csize != 0) {
727 /* At this point we are supposed to check the data CRC
728 * of our unchecked node. But thus far, we do not
729 * know whether the node is valid or obsolete. To
730 * figure this out, we need to walk all the nodes of
731 * the inode and build the inode fragtree. We don't
732 * want to spend time checking data of nodes which may
733 * later be found to be obsolete. So we put off the full
734 * data CRC checking until we have read all the inode
735 * nodes and have started building the fragtree.
737 * The fragtree is being built starting with nodes
738 * having the highest version number, so we'll be able
739 * to detect whether a node is valid (i.e., it is not
740 * overlapped by a node with higher version) or not.
741 * And we'll be able to check only those nodes, which
744 * Of course, this optimization only makes sense in case
745 * of NAND flashes (or other flashes with
746 * !jffs2_can_mark_obsolete()), since on NOR flashes
747 * nodes are marked obsolete physically.
749 * Since NAND flashes (or other flashes with
750 * jffs2_is_writebuffered(c)) are anyway read by
751 * fractions of c->wbuf_pagesize, and we have just read
752 * the node header, it is likely that the starting part
753 * of the node data is also read when we read the
754 * header. So we don't mind to check the CRC of the
755 * starting part of the data of the node now, and check
756 * the second part later (in jffs2_check_node_data()).
757 * Of course, we will not need to re-read and re-check
758 * the NAND page which we have just read. This is why we
759 * read the whole NAND page at jffs2_get_inode_nodes(),
760 * while we needed only the node header.
764 /* 'buf' will point to the start of data */
765 buf = (unsigned char *)rd + sizeof(*rd);
766 /* len will be the read data length */
767 len = min_t(uint32_t, rdlen - sizeof(*rd), csize);
768 tn->partial_crc = crc32(0, buf, len);
770 dbg_readinode("Calculates CRC (%#08x) for %d bytes, csize %d\n", tn->partial_crc, len, csize);
772 /* If we actually calculated the whole data CRC
773 * and it is wrong, drop the node. */
774 if (len >= csize && unlikely(tn->partial_crc != je32_to_cpu(rd->data_crc))) {
775 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
776 ref_offset(ref), tn->partial_crc, je32_to_cpu(rd->data_crc));
777 jffs2_mark_node_obsolete(c, ref);
781 } else if (csize == 0) {
783 * We checked the header CRC. If the node has no data, adjust
784 * the space accounting now. For other nodes this will be done
785 * later either when the node is marked obsolete or when its
788 struct jffs2_eraseblock *jeb;
790 dbg_readinode("the node has no data.\n");
791 jeb = &c->blocks[ref->flash_offset / c->sector_size];
792 len = ref_totlen(c, jeb, ref);
794 spin_lock(&c->erase_completion_lock);
795 jeb->used_size += len;
796 jeb->unchecked_size -= len;
798 c->unchecked_size -= len;
799 ref->flash_offset = ref_offset(ref) | REF_NORMAL;
800 spin_unlock(&c->erase_completion_lock);
804 tn->fn = jffs2_alloc_full_dnode();
806 JFFS2_ERROR("alloc fn failed\n");
811 tn->version = je32_to_cpu(rd->version);
812 tn->fn->ofs = je32_to_cpu(rd->offset);
813 tn->data_crc = je32_to_cpu(rd->data_crc);
818 if (tn->version > rii->highest_version)
819 rii->highest_version = tn->version;
821 /* There was a bug where we wrote hole nodes out with
822 csize/dsize swapped. Deal with it */
823 if (rd->compr == JFFS2_COMPR_ZERO && !je32_to_cpu(rd->dsize) && csize)
824 tn->fn->size = csize;
825 else // normal case...
826 tn->fn->size = je32_to_cpu(rd->dsize);
828 dbg_readinode2("dnode @%08x: ver %u, offset %#04x, dsize %#04x, csize %#04x\n",
829 ref_offset(ref), je32_to_cpu(rd->version),
830 je32_to_cpu(rd->offset), je32_to_cpu(rd->dsize), csize);
832 ret = jffs2_add_tn_to_tree(c, rii, tn);
835 jffs2_free_full_dnode(tn->fn);
837 jffs2_free_tmp_dnode_info(tn);
840 #ifdef JFFS2_DBG_READINODE2_MESSAGES
841 dbg_readinode2("After adding ver %d:\n", je32_to_cpu(rd->version));
842 tn = tn_first(&rii->tn_root);
844 dbg_readinode2("%p: v %d r 0x%x-0x%x ov %d\n",
845 tn, tn->version, tn->fn->ofs,
846 tn->fn->ofs+tn->fn->size, tn->overlapped);
854 * Helper function for jffs2_get_inode_nodes().
855 * It is called every time an unknown node is found.
857 * Returns: 0 on success;
858 * negative error code on failure.
860 static inline int read_unknown(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref, struct jffs2_unknown_node *un)
862 /* We don't mark unknown nodes as REF_UNCHECKED */
863 if (ref_flags(ref) == REF_UNCHECKED) {
864 JFFS2_ERROR("REF_UNCHECKED but unknown node at %#08x\n",
866 JFFS2_ERROR("Node is {%04x,%04x,%08x,%08x}. Please report this error.\n",
867 je16_to_cpu(un->magic), je16_to_cpu(un->nodetype),
868 je32_to_cpu(un->totlen), je32_to_cpu(un->hdr_crc));
869 jffs2_mark_node_obsolete(c, ref);
873 un->nodetype = cpu_to_je16(JFFS2_NODE_ACCURATE | je16_to_cpu(un->nodetype));
875 switch(je16_to_cpu(un->nodetype) & JFFS2_COMPAT_MASK) {
877 case JFFS2_FEATURE_INCOMPAT:
878 JFFS2_ERROR("unknown INCOMPAT nodetype %#04X at %#08x\n",
879 je16_to_cpu(un->nodetype), ref_offset(ref));
884 case JFFS2_FEATURE_ROCOMPAT:
885 JFFS2_ERROR("unknown ROCOMPAT nodetype %#04X at %#08x\n",
886 je16_to_cpu(un->nodetype), ref_offset(ref));
887 BUG_ON(!(c->flags & JFFS2_SB_FLAG_RO));
890 case JFFS2_FEATURE_RWCOMPAT_COPY:
891 JFFS2_NOTICE("unknown RWCOMPAT_COPY nodetype %#04X at %#08x\n",
892 je16_to_cpu(un->nodetype), ref_offset(ref));
895 case JFFS2_FEATURE_RWCOMPAT_DELETE:
896 JFFS2_NOTICE("unknown RWCOMPAT_DELETE nodetype %#04X at %#08x\n",
897 je16_to_cpu(un->nodetype), ref_offset(ref));
898 jffs2_mark_node_obsolete(c, ref);
906 * Helper function for jffs2_get_inode_nodes().
907 * The function detects whether more data should be read and reads it if yes.
909 * Returns: 0 on succes;
910 * negative error code on failure.
912 static int read_more(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
913 int needed_len, int *rdlen, unsigned char *buf)
915 int err, to_read = needed_len - *rdlen;
919 if (jffs2_is_writebuffered(c)) {
920 int rem = to_read % c->wbuf_pagesize;
923 to_read += c->wbuf_pagesize - rem;
926 /* We need to read more data */
927 offs = ref_offset(ref) + *rdlen;
929 dbg_readinode("read more %d bytes\n", to_read);
931 err = jffs2_flash_read(c, offs, to_read, &retlen, buf + *rdlen);
933 JFFS2_ERROR("can not read %d bytes from 0x%08x, "
934 "error code: %d.\n", to_read, offs, err);
938 if (retlen < to_read) {
939 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n",
940 offs, retlen, to_read);
948 /* Get tmp_dnode_info and full_dirent for all non-obsolete nodes associated
949 with this ino. Perform a preliminary ordering on data nodes, throwing away
950 those which are completely obsoleted by newer ones. The naïve approach we
951 use to take of just returning them _all_ in version order will cause us to
952 run out of memory in certain degenerate cases. */
953 static int jffs2_get_inode_nodes(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
954 struct jffs2_readinode_info *rii)
956 struct jffs2_raw_node_ref *ref, *valid_ref;
957 unsigned char *buf = NULL;
958 union jffs2_node_union *node;
964 dbg_readinode("ino #%u\n", f->inocache->ino);
966 /* FIXME: in case of NOR and available ->point() this
967 * needs to be fixed. */
968 len = sizeof(union jffs2_node_union) + c->wbuf_pagesize;
969 buf = kmalloc(len, GFP_KERNEL);
973 spin_lock(&c->erase_completion_lock);
974 valid_ref = jffs2_first_valid_node(f->inocache->nodes);
975 if (!valid_ref && f->inocache->ino != 1)
976 JFFS2_WARNING("Eep. No valid nodes for ino #%u.\n", f->inocache->ino);
978 /* We can hold a pointer to a non-obsolete node without the spinlock,
979 but _obsolete_ nodes may disappear at any time, if the block
980 they're in gets erased. So if we mark 'ref' obsolete while we're
981 not holding the lock, it can go away immediately. For that reason,
982 we find the next valid node first, before processing 'ref'.
985 valid_ref = jffs2_first_valid_node(ref->next_in_ino);
986 spin_unlock(&c->erase_completion_lock);
991 * At this point we don't know the type of the node we're going
992 * to read, so we do not know the size of its header. In order
993 * to minimize the amount of flash IO we assume the header is
994 * of size = JFFS2_MIN_NODE_HEADER.
996 len = JFFS2_MIN_NODE_HEADER;
997 if (jffs2_is_writebuffered(c)) {
1001 * We are about to read JFFS2_MIN_NODE_HEADER bytes,
1002 * but this flash has some minimal I/O unit. It is
1003 * possible that we'll need to read more soon, so read
1004 * up to the next min. I/O unit, in order not to
1005 * re-read the same min. I/O unit twice.
1007 end = ref_offset(ref) + len;
1008 rem = end % c->wbuf_pagesize;
1010 end += c->wbuf_pagesize - rem;
1011 len = end - ref_offset(ref);
1014 dbg_readinode("read %d bytes at %#08x(%d).\n", len, ref_offset(ref), ref_flags(ref));
1016 /* FIXME: point() */
1017 err = jffs2_flash_read(c, ref_offset(ref), len, &retlen, buf);
1019 JFFS2_ERROR("can not read %d bytes from 0x%08x, " "error code: %d.\n", len, ref_offset(ref), err);
1024 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n", ref_offset(ref), retlen, len);
1029 node = (union jffs2_node_union *)buf;
1031 /* No need to mask in the valid bit; it shouldn't be invalid */
1032 if (je32_to_cpu(node->u.hdr_crc) != crc32(0, node, sizeof(node->u)-4)) {
1033 JFFS2_NOTICE("Node header CRC failed at %#08x. {%04x,%04x,%08x,%08x}\n",
1034 ref_offset(ref), je16_to_cpu(node->u.magic),
1035 je16_to_cpu(node->u.nodetype),
1036 je32_to_cpu(node->u.totlen),
1037 je32_to_cpu(node->u.hdr_crc));
1038 jffs2_dbg_dump_node(c, ref_offset(ref));
1039 jffs2_mark_node_obsolete(c, ref);
1042 if (je16_to_cpu(node->u.magic) != JFFS2_MAGIC_BITMASK) {
1043 /* Not a JFFS2 node, whinge and move on */
1044 JFFS2_NOTICE("Wrong magic bitmask 0x%04x in node header at %#08x.\n",
1045 je16_to_cpu(node->u.magic), ref_offset(ref));
1046 jffs2_mark_node_obsolete(c, ref);
1050 switch (je16_to_cpu(node->u.nodetype)) {
1052 case JFFS2_NODETYPE_DIRENT:
1054 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_dirent) &&
1055 len < sizeof(struct jffs2_raw_dirent)) {
1056 err = read_more(c, ref, sizeof(struct jffs2_raw_dirent), &len, buf);
1061 err = read_direntry(c, ref, &node->d, retlen, rii);
1067 case JFFS2_NODETYPE_INODE:
1069 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_inode) &&
1070 len < sizeof(struct jffs2_raw_inode)) {
1071 err = read_more(c, ref, sizeof(struct jffs2_raw_inode), &len, buf);
1076 err = read_dnode(c, ref, &node->i, len, rii);
1083 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_unknown_node) &&
1084 len < sizeof(struct jffs2_unknown_node)) {
1085 err = read_more(c, ref, sizeof(struct jffs2_unknown_node), &len, buf);
1090 err = read_unknown(c, ref, &node->u);
1096 spin_lock(&c->erase_completion_lock);
1099 spin_unlock(&c->erase_completion_lock);
1102 f->highest_version = rii->highest_version;
1104 dbg_readinode("nodes of inode #%u were read, the highest version is %u, latest_mctime %u, mctime_ver %u.\n",
1105 f->inocache->ino, rii->highest_version, rii->latest_mctime,
1110 jffs2_free_tmp_dnode_info_list(&rii->tn_root);
1111 jffs2_free_full_dirent_list(rii->fds);
1117 static int jffs2_do_read_inode_internal(struct jffs2_sb_info *c,
1118 struct jffs2_inode_info *f,
1119 struct jffs2_raw_inode *latest_node)
1121 struct jffs2_readinode_info rii;
1122 uint32_t crc, new_size;
1126 dbg_readinode("ino #%u nlink is %d\n", f->inocache->ino, f->inocache->nlink);
1128 memset(&rii, 0, sizeof(rii));
1130 /* Grab all nodes relevant to this ino */
1131 ret = jffs2_get_inode_nodes(c, f, &rii);
1134 JFFS2_ERROR("cannot read nodes for ino %u, returned error is %d\n", f->inocache->ino, ret);
1135 if (f->inocache->state == INO_STATE_READING)
1136 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1140 ret = jffs2_build_inode_fragtree(c, f, &rii);
1142 JFFS2_ERROR("Failed to build final fragtree for inode #%u: error %d\n",
1143 f->inocache->ino, ret);
1144 if (f->inocache->state == INO_STATE_READING)
1145 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1146 jffs2_free_tmp_dnode_info_list(&rii.tn_root);
1147 /* FIXME: We could at least crc-check them all */
1149 jffs2_free_full_dnode(rii.mdata_tn->fn);
1150 jffs2_free_tmp_dnode_info(rii.mdata_tn);
1151 rii.mdata_tn = NULL;
1157 if (rii.mdata_tn->fn->raw == rii.latest_ref) {
1158 f->metadata = rii.mdata_tn->fn;
1159 jffs2_free_tmp_dnode_info(rii.mdata_tn);
1161 jffs2_kill_tn(c, rii.mdata_tn);
1163 rii.mdata_tn = NULL;
1168 jffs2_dbg_fragtree_paranoia_check_nolock(f);
1170 if (unlikely(!rii.latest_ref)) {
1171 /* No data nodes for this inode. */
1172 if (f->inocache->ino != 1) {
1173 JFFS2_WARNING("no data nodes found for ino #%u\n", f->inocache->ino);
1175 if (f->inocache->state == INO_STATE_READING)
1176 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1179 JFFS2_NOTICE("but it has children so we fake some modes for it\n");
1181 latest_node->mode = cpu_to_jemode(S_IFDIR|S_IRUGO|S_IWUSR|S_IXUGO);
1182 latest_node->version = cpu_to_je32(0);
1183 latest_node->atime = latest_node->ctime = latest_node->mtime = cpu_to_je32(0);
1184 latest_node->isize = cpu_to_je32(0);
1185 latest_node->gid = cpu_to_je16(0);
1186 latest_node->uid = cpu_to_je16(0);
1187 if (f->inocache->state == INO_STATE_READING)
1188 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1192 ret = jffs2_flash_read(c, ref_offset(rii.latest_ref), sizeof(*latest_node), &retlen, (void *)latest_node);
1193 if (ret || retlen != sizeof(*latest_node)) {
1194 JFFS2_ERROR("failed to read from flash: error %d, %zd of %zd bytes read\n",
1195 ret, retlen, sizeof(*latest_node));
1196 /* FIXME: If this fails, there seems to be a memory leak. Find it. */
1197 mutex_unlock(&f->sem);
1198 jffs2_do_clear_inode(c, f);
1199 return ret?ret:-EIO;
1202 crc = crc32(0, latest_node, sizeof(*latest_node)-8);
1203 if (crc != je32_to_cpu(latest_node->node_crc)) {
1204 JFFS2_ERROR("CRC failed for read_inode of inode %u at physical location 0x%x\n",
1205 f->inocache->ino, ref_offset(rii.latest_ref));
1206 mutex_unlock(&f->sem);
1207 jffs2_do_clear_inode(c, f);
1211 switch(jemode_to_cpu(latest_node->mode) & S_IFMT) {
1213 if (rii.mctime_ver > je32_to_cpu(latest_node->version)) {
1214 /* The times in the latest_node are actually older than
1215 mctime in the latest dirent. Cheat. */
1216 latest_node->ctime = latest_node->mtime = cpu_to_je32(rii.latest_mctime);
1222 /* If it was a regular file, truncate it to the latest node's isize */
1223 new_size = jffs2_truncate_fragtree(c, &f->fragtree, je32_to_cpu(latest_node->isize));
1224 if (new_size != je32_to_cpu(latest_node->isize)) {
1225 JFFS2_WARNING("Truncating ino #%u to %d bytes failed because it only had %d bytes to start with!\n",
1226 f->inocache->ino, je32_to_cpu(latest_node->isize), new_size);
1227 latest_node->isize = cpu_to_je32(new_size);
1232 /* Hack to work around broken isize in old symlink code.
1233 Remove this when dwmw2 comes to his senses and stops
1234 symlinks from being an entirely gratuitous special
1236 if (!je32_to_cpu(latest_node->isize))
1237 latest_node->isize = latest_node->dsize;
1239 if (f->inocache->state != INO_STATE_CHECKING) {
1240 /* Symlink's inode data is the target path. Read it and
1241 * keep in RAM to facilitate quick follow symlink
1243 f->target = kmalloc(je32_to_cpu(latest_node->csize) + 1, GFP_KERNEL);
1245 JFFS2_ERROR("can't allocate %d bytes of memory for the symlink target path cache\n", je32_to_cpu(latest_node->csize));
1246 mutex_unlock(&f->sem);
1247 jffs2_do_clear_inode(c, f);
1251 ret = jffs2_flash_read(c, ref_offset(rii.latest_ref) + sizeof(*latest_node),
1252 je32_to_cpu(latest_node->csize), &retlen, (char *)f->target);
1254 if (ret || retlen != je32_to_cpu(latest_node->csize)) {
1255 if (retlen != je32_to_cpu(latest_node->csize))
1259 mutex_unlock(&f->sem);
1260 jffs2_do_clear_inode(c, f);
1264 f->target[je32_to_cpu(latest_node->csize)] = '\0';
1265 dbg_readinode("symlink's target '%s' cached\n", f->target);
1268 /* fall through... */
1272 /* Certain inode types should have only one data node, and it's
1273 kept as the metadata node */
1275 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o had metadata node\n",
1276 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1277 mutex_unlock(&f->sem);
1278 jffs2_do_clear_inode(c, f);
1281 if (!frag_first(&f->fragtree)) {
1282 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o has no fragments\n",
1283 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1284 mutex_unlock(&f->sem);
1285 jffs2_do_clear_inode(c, f);
1288 /* ASSERT: f->fraglist != NULL */
1289 if (frag_next(frag_first(&f->fragtree))) {
1290 JFFS2_ERROR("Argh. Special inode #%u with mode 0x%x had more than one node\n",
1291 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1292 /* FIXME: Deal with it - check crc32, check for duplicate node, check times and discard the older one */
1293 mutex_unlock(&f->sem);
1294 jffs2_do_clear_inode(c, f);
1297 /* OK. We're happy */
1298 f->metadata = frag_first(&f->fragtree)->node;
1299 jffs2_free_node_frag(frag_first(&f->fragtree));
1300 f->fragtree = RB_ROOT;
1303 if (f->inocache->state == INO_STATE_READING)
1304 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1309 /* Scan the list of all nodes present for this ino, build map of versions, etc. */
1310 int jffs2_do_read_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
1311 uint32_t ino, struct jffs2_raw_inode *latest_node)
1313 dbg_readinode("read inode #%u\n", ino);
1316 spin_lock(&c->inocache_lock);
1317 f->inocache = jffs2_get_ino_cache(c, ino);
1320 /* Check its state. We may need to wait before we can use it */
1321 switch(f->inocache->state) {
1322 case INO_STATE_UNCHECKED:
1323 case INO_STATE_CHECKEDABSENT:
1324 f->inocache->state = INO_STATE_READING;
1327 case INO_STATE_CHECKING:
1329 /* If it's in either of these states, we need
1330 to wait for whoever's got it to finish and
1332 dbg_readinode("waiting for ino #%u in state %d\n", ino, f->inocache->state);
1333 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
1334 goto retry_inocache;
1336 case INO_STATE_READING:
1337 case INO_STATE_PRESENT:
1338 /* Eep. This should never happen. It can
1339 happen if Linux calls read_inode() again
1340 before clear_inode() has finished though. */
1341 JFFS2_ERROR("Eep. Trying to read_inode #%u when it's already in state %d!\n", ino, f->inocache->state);
1342 /* Fail. That's probably better than allowing it to succeed */
1350 spin_unlock(&c->inocache_lock);
1352 if (!f->inocache && ino == 1) {
1353 /* Special case - no root inode on medium */
1354 f->inocache = jffs2_alloc_inode_cache();
1356 JFFS2_ERROR("cannot allocate inocache for root inode\n");
1359 dbg_readinode("creating inocache for root inode\n");
1360 memset(f->inocache, 0, sizeof(struct jffs2_inode_cache));
1361 f->inocache->ino = f->inocache->nlink = 1;
1362 f->inocache->nodes = (struct jffs2_raw_node_ref *)f->inocache;
1363 f->inocache->state = INO_STATE_READING;
1364 jffs2_add_ino_cache(c, f->inocache);
1367 JFFS2_ERROR("requestied to read an nonexistent ino %u\n", ino);
1371 return jffs2_do_read_inode_internal(c, f, latest_node);
1374 int jffs2_do_crccheck_inode(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic)
1376 struct jffs2_raw_inode n;
1377 struct jffs2_inode_info *f = kzalloc(sizeof(*f), GFP_KERNEL);
1383 mutex_init(&f->sem);
1384 mutex_lock(&f->sem);
1387 ret = jffs2_do_read_inode_internal(c, f, &n);
1389 mutex_unlock(&f->sem);
1390 jffs2_do_clear_inode(c, f);
1396 void jffs2_do_clear_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f)
1398 struct jffs2_full_dirent *fd, *fds;
1402 jffs2_xattr_delete_inode(c, f->inocache);
1403 mutex_lock(&f->sem);
1404 deleted = f->inocache && !f->inocache->nlink;
1406 if (f->inocache && f->inocache->state != INO_STATE_CHECKING)
1407 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CLEARING);
1411 jffs2_mark_node_obsolete(c, f->metadata->raw);
1412 jffs2_free_full_dnode(f->metadata);
1415 jffs2_kill_fragtree(&f->fragtree, deleted?c:NULL);
1426 jffs2_free_full_dirent(fd);
1429 if (f->inocache && f->inocache->state != INO_STATE_CHECKING) {
1430 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1431 if (f->inocache->nodes == (void *)f->inocache)
1432 jffs2_del_ino_cache(c, f->inocache);
1435 mutex_unlock(&f->sem);