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 if (!jffs2_is_writebuffered(c))
43 /* Calculate how many bytes were already checked */
44 ofs = ref_offset(ref) + sizeof(struct jffs2_raw_inode);
45 len = ofs % c->wbuf_pagesize;
47 len = c->wbuf_pagesize - len;
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);
56 len = tn->csize - len;
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);
62 /* TODO: instead, incapsulate point() stuff to jffs2_flash_read(),
63 * adding and jffs2_flash_read_end() interface. */
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);
70 JFFS2_WARNING("MTD point failed: error code %d.\n", err);
72 pointed = 1; /* succefully pointed to device */
77 buffer = kmalloc(len, GFP_KERNEL);
78 if (unlikely(!buffer))
81 /* TODO: this is very frequent pattern, make it a separate
83 err = jffs2_flash_read(c, ofs, len, &retlen, buffer);
85 JFFS2_ERROR("can not read %d bytes from 0x%08x, error code: %d.\n", len, ofs, err);
90 JFFS2_ERROR("short read at %#08x: %zd instead of %d.\n", ofs, retlen, len);
96 /* Continue calculating CRC */
97 crc = crc32(tn->partial_crc, buffer, len);
102 c->mtd->unpoint(c->mtd, buffer, ofs, len);
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);
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;
120 * Mark the node as having been checked and fix the
121 * accounting accordingly.
123 spin_lock(&c->erase_completion_lock);
124 jeb->used_size += len;
125 jeb->unchecked_size -= len;
127 c->unchecked_size -= len;
128 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
129 spin_unlock(&c->erase_completion_lock);
138 c->mtd->unpoint(c->mtd, buffer, ofs, len);
144 * Helper function for jffs2_add_older_frag_to_fragtree().
146 * Checks the node if we are in the checking stage.
148 static int check_tn_node(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
152 BUG_ON(ref_obsolete(tn->fn->raw));
154 /* We only check the data CRC of unchecked nodes */
155 if (ref_flags(tn->fn->raw) != REF_UNCHECKED)
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));
161 ret = check_node_data(c, tn);
162 if (unlikely(ret < 0)) {
163 JFFS2_ERROR("check_node_data() returned error: %d.\n",
165 } else if (unlikely(ret > 0)) {
166 dbg_readinode("CRC error, mark it obsolete.\n");
167 jffs2_mark_node_obsolete(c, tn->fn->raw);
173 static struct jffs2_tmp_dnode_info *jffs2_lookup_tn(struct rb_root *tn_root, uint32_t offset)
175 struct rb_node *next;
176 struct jffs2_tmp_dnode_info *tn = NULL;
178 dbg_readinode("root %p, offset %d\n", tn_root, offset);
180 next = tn_root->rb_node;
183 tn = rb_entry(next, struct jffs2_tmp_dnode_info, rb);
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;
197 static void jffs2_kill_tn(struct jffs2_sb_info *c, struct jffs2_tmp_dnode_info *tn)
199 jffs2_mark_node_obsolete(c, tn->fn->raw);
200 jffs2_free_full_dnode(tn->fn);
201 jffs2_free_tmp_dnode_info(tn);
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
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
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 /* 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);
237 dbg_readinode("keep new mdata with ver %d\n", tn->version);
241 /* Find the earliest node which _may_ be relevant to this one */
242 this = jffs2_lookup_tn(&rii->tn_root, tn->fn->ofs);
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");
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)
256 dbg_readinode("'this' found %#04x-%#04x (%s)\n", this->fn->ofs, this->fn->ofs + this->fn->size, this->fn ? "data" : "hole");
259 if (this->fn->ofs > fn_end)
261 dbg_readinode("Ponder this ver %d, 0x%x-0x%x\n",
262 this->version, this->fn->ofs, this->fn->size);
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);
273 /* Who cares if the new one is good; keep it for now anyway. */
274 dbg_readinode("Like new node. Throw away old\n");
275 rb_replace_node(&this->rb, &tn->rb, &rii->tn_root);
276 jffs2_kill_tn(c, this);
277 /* Same overlapping from in front and behind */
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);
290 /* ... and is good. Kill 'this' and any subsequent nodes which are also overlapped */
291 while (this && this->fn->ofs + this->fn->size < fn_end) {
292 struct jffs2_tmp_dnode_info *next = tn_next(this);
293 if (this->version < tn->version) {
294 tn_erase(this, &rii->tn_root);
295 dbg_readinode("Kill overlapped ver %d, 0x%x-0x%x\n",
296 this->version, this->fn->ofs,
297 this->fn->ofs+this->fn->size);
298 jffs2_kill_tn(c, this);
302 dbg_readinode("Done killing overlapped nodes\n");
305 if (this->version > tn->version &&
306 this->fn->ofs <= tn->fn->ofs &&
307 this->fn->ofs+this->fn->size >= fn_end) {
308 /* New node entirely overlapped by 'this' */
309 if (!check_tn_node(c, this)) {
310 dbg_readinode("Good CRC on old node. Kill new\n");
311 jffs2_kill_tn(c, tn);
314 /* ... but 'this' was bad. Replace it... */
315 dbg_readinode("Bad CRC on old overlapping node. Kill it\n");
316 tn_erase(this, &rii->tn_root);
317 jffs2_kill_tn(c, this);
321 this = tn_next(this);
324 /* We neither completely obsoleted nor were completely
325 obsoleted by an earlier node. Insert into the tree */
327 struct rb_node *parent;
328 struct rb_node **link = &rii->tn_root.rb_node;
329 struct jffs2_tmp_dnode_info *insert_point;
333 insert_point = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
334 if (tn->fn->ofs > insert_point->fn->ofs)
335 link = &insert_point->rb.rb_right;
336 else if (tn->fn->ofs < insert_point->fn->ofs ||
337 tn->fn->size < insert_point->fn->size)
338 link = &insert_point->rb.rb_left;
340 link = &insert_point->rb.rb_right;
342 rb_link_node(&tn->rb, &insert_point->rb, link);
343 rb_insert_color(&tn->rb, &rii->tn_root);
347 /* If there's anything behind that overlaps us, note it */
351 if (this->fn->ofs + this->fn->size > tn->fn->ofs) {
352 dbg_readinode("Node is overlapped by %p (v %d, 0x%x-0x%x)\n",
353 this, this->version, this->fn->ofs,
354 this->fn->ofs+this->fn->size);
358 if (!this->overlapped)
360 this = tn_prev(this);
364 /* If the new node overlaps anything ahead, note it */
366 while (this && this->fn->ofs < fn_end) {
367 this->overlapped = 1;
368 dbg_readinode("Node ver %d, 0x%x-0x%x is overlapped\n",
369 this->version, this->fn->ofs,
370 this->fn->ofs+this->fn->size);
371 this = tn_next(this);
376 /* Trivial function to remove the last node in the tree. Which by definition
377 has no right-hand -- so can be removed just by making its only child (if
378 any) take its place under its parent. */
379 static void eat_last(struct rb_root *root, struct rb_node *node)
381 struct rb_node *parent = rb_parent(node);
382 struct rb_node **link;
385 BUG_ON(node->rb_right);
388 link = &root->rb_node;
389 else if (node == parent->rb_left)
390 link = &parent->rb_left;
392 link = &parent->rb_right;
394 *link = node->rb_left;
395 /* Colour doesn't matter now. Only the parent pointer. */
397 node->rb_left->rb_parent_color = node->rb_parent_color;
400 /* We put this in reverse order, so we can just use eat_last */
401 static void ver_insert(struct rb_root *ver_root, struct jffs2_tmp_dnode_info *tn)
403 struct rb_node **link = &ver_root->rb_node;
404 struct rb_node *parent = NULL;
405 struct jffs2_tmp_dnode_info *this_tn;
409 this_tn = rb_entry(parent, struct jffs2_tmp_dnode_info, rb);
411 if (tn->version > this_tn->version)
412 link = &parent->rb_left;
414 link = &parent->rb_right;
416 dbg_readinode("Link new node at %p (root is %p)\n", link, ver_root);
417 rb_link_node(&tn->rb, parent, link);
418 rb_insert_color(&tn->rb, ver_root);
421 /* Build final, normal fragtree from tn tree. It doesn't matter which order
422 we add nodes to the real fragtree, as long as they don't overlap. And
423 having thrown away the majority of overlapped nodes as we went, there
424 really shouldn't be many sets of nodes which do overlap. If we start at
425 the end, we can use the overlap markers -- we can just eat nodes which
426 aren't overlapped, and when we encounter nodes which _do_ overlap we
427 sort them all into a temporary tree in version order before replaying them. */
428 static int jffs2_build_inode_fragtree(struct jffs2_sb_info *c,
429 struct jffs2_inode_info *f,
430 struct jffs2_readinode_info *rii)
432 struct jffs2_tmp_dnode_info *pen, *last, *this;
433 struct rb_root ver_root = RB_ROOT;
434 uint32_t high_ver = 0;
437 dbg_readinode("potential mdata is ver %d at %p\n", rii->mdata_tn->version, rii->mdata_tn);
438 high_ver = rii->mdata_tn->version;
439 rii->latest_ref = rii->mdata_tn->fn->raw;
441 #ifdef JFFS2_DBG_READINODE_MESSAGES
442 this = tn_last(&rii->tn_root);
444 dbg_readinode("tn %p ver %d range 0x%x-0x%x ov %d\n", this, this->version, this->fn->ofs,
445 this->fn->ofs+this->fn->size, this->overlapped);
446 this = tn_prev(this);
449 pen = tn_last(&rii->tn_root);
450 while ((last = pen)) {
453 eat_last(&rii->tn_root, &last->rb);
454 ver_insert(&ver_root, last);
456 if (unlikely(last->overlapped))
459 /* Now we have a bunch of nodes in reverse version
460 order, in the tree at ver_root. Most of the time,
461 there'll actually be only one node in the 'tree',
463 this = tn_last(&ver_root);
466 struct jffs2_tmp_dnode_info *vers_next;
468 vers_next = tn_prev(this);
469 eat_last(&ver_root, &this->rb);
470 if (check_tn_node(c, this)) {
471 dbg_readinode("node ver %d, 0x%x-0x%x failed CRC\n",
472 this->version, this->fn->ofs,
473 this->fn->ofs+this->fn->size);
474 jffs2_kill_tn(c, this);
476 if (this->version > high_ver) {
477 /* Note that this is different from the other
478 highest_version, because this one is only
479 counting _valid_ nodes which could give the
480 latest inode metadata */
481 high_ver = this->version;
482 rii->latest_ref = this->fn->raw;
484 dbg_readinode("Add %p (v %d, 0x%x-0x%x, ov %d) to fragtree\n",
485 this, this->version, this->fn->ofs,
486 this->fn->ofs+this->fn->size, this->overlapped);
488 ret = jffs2_add_full_dnode_to_inode(c, f, this->fn);
490 /* Free the nodes in vers_root; let the caller
491 deal with the rest */
492 JFFS2_ERROR("Add node to tree failed %d\n", ret);
494 vers_next = tn_prev(this);
495 if (check_tn_node(c, this))
496 jffs2_mark_node_obsolete(c, this->fn->raw);
497 jffs2_free_full_dnode(this->fn);
498 jffs2_free_tmp_dnode_info(this);
502 eat_last(&ver_root, &vers_next->rb);
506 jffs2_free_tmp_dnode_info(this);
514 static void jffs2_free_tmp_dnode_info_list(struct rb_root *list)
516 struct rb_node *this;
517 struct jffs2_tmp_dnode_info *tn;
519 this = list->rb_node;
521 /* Now at bottom of tree */
524 this = this->rb_left;
525 else if (this->rb_right)
526 this = this->rb_right;
528 tn = rb_entry(this, struct jffs2_tmp_dnode_info, rb);
529 jffs2_free_full_dnode(tn->fn);
530 jffs2_free_tmp_dnode_info(tn);
532 this = rb_parent(this);
536 if (this->rb_left == &tn->rb)
537 this->rb_left = NULL;
538 else if (this->rb_right == &tn->rb)
539 this->rb_right = NULL;
543 list->rb_node = NULL;
546 static void jffs2_free_full_dirent_list(struct jffs2_full_dirent *fd)
548 struct jffs2_full_dirent *next;
552 jffs2_free_full_dirent(fd);
557 /* Returns first valid node after 'ref'. May return 'ref' */
558 static struct jffs2_raw_node_ref *jffs2_first_valid_node(struct jffs2_raw_node_ref *ref)
560 while (ref && ref->next_in_ino) {
561 if (!ref_obsolete(ref))
563 dbg_noderef("node at 0x%08x is obsoleted. Ignoring.\n", ref_offset(ref));
564 ref = ref->next_in_ino;
570 * Helper function for jffs2_get_inode_nodes().
571 * It is called every time an directory entry node is found.
573 * Returns: 0 on succes;
574 * 1 if the node should be marked obsolete;
575 * negative error code on failure.
577 static inline int read_direntry(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
578 struct jffs2_raw_dirent *rd, size_t read,
579 struct jffs2_readinode_info *rii)
581 struct jffs2_full_dirent *fd;
584 /* Obsoleted. This cannot happen, surely? dwmw2 20020308 */
585 BUG_ON(ref_obsolete(ref));
587 crc = crc32(0, rd, sizeof(*rd) - 8);
588 if (unlikely(crc != je32_to_cpu(rd->node_crc))) {
589 JFFS2_NOTICE("header CRC failed on dirent node at %#08x: read %#08x, calculated %#08x\n",
590 ref_offset(ref), je32_to_cpu(rd->node_crc), crc);
591 jffs2_mark_node_obsolete(c, ref);
595 /* If we've never checked the CRCs on this node, check them now */
596 if (ref_flags(ref) == REF_UNCHECKED) {
597 struct jffs2_eraseblock *jeb;
601 if (unlikely(PAD((rd->nsize + sizeof(*rd))) != PAD(je32_to_cpu(rd->totlen)))) {
602 JFFS2_ERROR("illegal nsize in node at %#08x: nsize %#02x, totlen %#04x\n",
603 ref_offset(ref), rd->nsize, je32_to_cpu(rd->totlen));
604 jffs2_mark_node_obsolete(c, ref);
608 jeb = &c->blocks[ref->flash_offset / c->sector_size];
609 len = ref_totlen(c, jeb, ref);
611 spin_lock(&c->erase_completion_lock);
612 jeb->used_size += len;
613 jeb->unchecked_size -= len;
615 c->unchecked_size -= len;
616 ref->flash_offset = ref_offset(ref) | REF_PRISTINE;
617 spin_unlock(&c->erase_completion_lock);
620 fd = jffs2_alloc_full_dirent(rd->nsize + 1);
625 fd->version = je32_to_cpu(rd->version);
626 fd->ino = je32_to_cpu(rd->ino);
629 if (fd->version > rii->highest_version)
630 rii->highest_version = fd->version;
632 /* Pick out the mctime of the latest dirent */
633 if(fd->version > rii->mctime_ver && je32_to_cpu(rd->mctime)) {
634 rii->mctime_ver = fd->version;
635 rii->latest_mctime = je32_to_cpu(rd->mctime);
639 * Copy as much of the name as possible from the raw
640 * dirent we've already read from the flash.
642 if (read > sizeof(*rd))
643 memcpy(&fd->name[0], &rd->name[0],
644 min_t(uint32_t, rd->nsize, (read - sizeof(*rd)) ));
646 /* Do we need to copy any more of the name directly from the flash? */
647 if (rd->nsize + sizeof(*rd) > read) {
650 int already = read - sizeof(*rd);
652 err = jffs2_flash_read(c, (ref_offset(ref)) + read,
653 rd->nsize - already, &read, &fd->name[already]);
654 if (unlikely(read != rd->nsize - already) && likely(!err))
658 JFFS2_ERROR("read remainder of name: error %d\n", err);
659 jffs2_free_full_dirent(fd);
664 fd->nhash = full_name_hash(fd->name, rd->nsize);
666 fd->name[rd->nsize] = '\0';
669 * Wheee. We now have a complete jffs2_full_dirent structure, with
670 * the name in it and everything. Link it into the list
672 jffs2_add_fd_to_list(c, fd, &rii->fds);
678 * Helper function for jffs2_get_inode_nodes().
679 * It is called every time an inode node is found.
681 * Returns: 0 on success;
682 * 1 if the node should be marked 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));
725 if (jffs2_is_writebuffered(c) && csize != 0) {
726 /* At this point we are supposed to check the data CRC
727 * of our unchecked node. But thus far, we do not
728 * know whether the node is valid or obsolete. To
729 * figure this out, we need to walk all the nodes of
730 * the inode and build the inode fragtree. We don't
731 * want to spend time checking data of nodes which may
732 * later be found to be obsolete. So we put off the full
733 * data CRC checking until we have read all the inode
734 * nodes and have started building the fragtree.
736 * The fragtree is being built starting with nodes
737 * having the highest version number, so we'll be able
738 * to detect whether a node is valid (i.e., it is not
739 * overlapped by a node with higher version) or not.
740 * And we'll be able to check only those nodes, which
743 * Of course, this optimization only makes sense in case
744 * of NAND flashes (or other flashes whith
745 * !jffs2_can_mark_obsolete()), since on NOR flashes
746 * nodes are marked obsolete physically.
748 * Since NAND flashes (or other flashes with
749 * jffs2_is_writebuffered(c)) are anyway read by
750 * fractions of c->wbuf_pagesize, and we have just read
751 * the node header, it is likely that the starting part
752 * of the node data is also read when we read the
753 * header. So we don't mind to check the CRC of the
754 * starting part of the data of the node now, and check
755 * the second part later (in jffs2_check_node_data()).
756 * Of course, we will not need to re-read and re-check
757 * the NAND page which we have just read. This is why we
758 * read the whole NAND page at jffs2_get_inode_nodes(),
759 * while we needed only the node header.
763 /* 'buf' will point to the start of data */
764 buf = (unsigned char *)rd + sizeof(*rd);
765 /* len will be the read data length */
766 len = min_t(uint32_t, rdlen - sizeof(*rd), csize);
767 tn->partial_crc = crc32(0, buf, len);
769 dbg_readinode("Calculates CRC (%#08x) for %d bytes, csize %d\n", tn->partial_crc, len, csize);
771 /* If we actually calculated the whole data CRC
772 * and it is wrong, drop the node. */
773 if (len >= csize && unlikely(tn->partial_crc != je32_to_cpu(rd->data_crc))) {
774 JFFS2_NOTICE("wrong data CRC in data node at 0x%08x: read %#08x, calculated %#08x.\n",
775 ref_offset(ref), tn->partial_crc, je32_to_cpu(rd->data_crc));
779 } else if (csize == 0) {
781 * We checked the header CRC. If the node has no data, adjust
782 * the space accounting now. For other nodes this will be done
783 * later either when the node is marked obsolete or when its
786 struct jffs2_eraseblock *jeb;
788 dbg_readinode("the node has no data.\n");
789 jeb = &c->blocks[ref->flash_offset / c->sector_size];
790 len = ref_totlen(c, jeb, ref);
792 spin_lock(&c->erase_completion_lock);
793 jeb->used_size += len;
794 jeb->unchecked_size -= len;
796 c->unchecked_size -= len;
797 ref->flash_offset = ref_offset(ref) | REF_NORMAL;
798 spin_unlock(&c->erase_completion_lock);
802 tn->fn = jffs2_alloc_full_dnode();
804 JFFS2_ERROR("alloc fn failed\n");
809 tn->version = je32_to_cpu(rd->version);
810 tn->fn->ofs = je32_to_cpu(rd->offset);
811 tn->data_crc = je32_to_cpu(rd->data_crc);
816 if (tn->version > rii->highest_version)
817 rii->highest_version = tn->version;
819 /* There was a bug where we wrote hole nodes out with
820 csize/dsize swapped. Deal with it */
821 if (rd->compr == JFFS2_COMPR_ZERO && !je32_to_cpu(rd->dsize) && csize)
822 tn->fn->size = csize;
823 else // normal case...
824 tn->fn->size = je32_to_cpu(rd->dsize);
826 dbg_readinode("dnode @%08x: ver %u, offset %#04x, dsize %#04x, csize %#04x\n",
827 ref_offset(ref), je32_to_cpu(rd->version), je32_to_cpu(rd->offset), je32_to_cpu(rd->dsize), csize);
829 ret = jffs2_add_tn_to_tree(c, rii, tn);
832 jffs2_free_full_dnode(tn->fn);
834 jffs2_free_tmp_dnode_info(tn);
837 #ifdef JFFS2_DBG_READINODE_MESSAGES
838 dbg_readinode("After adding ver %d:\n", je32_to_cpu(rd->version));
839 tn = tn_first(&rii->tn_root);
841 dbg_readinode("%p: v %d r 0x%x-0x%x ov %d\n",
842 tn, tn->version, tn->fn->ofs,
843 tn->fn->ofs+tn->fn->size, tn->overlapped);
851 * Helper function for jffs2_get_inode_nodes().
852 * It is called every time an unknown node is found.
854 * Returns: 0 on success;
855 * 1 if the node should be marked obsolete;
856 * negative error code on failure.
858 static inline int read_unknown(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref, struct jffs2_unknown_node *un)
860 /* We don't mark unknown nodes as REF_UNCHECKED */
861 if (ref_flags(ref) == REF_UNCHECKED) {
862 JFFS2_ERROR("REF_UNCHECKED but unknown node at %#08x\n",
864 JFFS2_ERROR("Node is {%04x,%04x,%08x,%08x}. Please report this error.\n",
865 je16_to_cpu(un->magic), je16_to_cpu(un->nodetype),
866 je32_to_cpu(un->totlen), je32_to_cpu(un->hdr_crc));
867 jffs2_mark_node_obsolete(c, ref);
871 un->nodetype = cpu_to_je16(JFFS2_NODE_ACCURATE | je16_to_cpu(un->nodetype));
873 switch(je16_to_cpu(un->nodetype) & JFFS2_COMPAT_MASK) {
875 case JFFS2_FEATURE_INCOMPAT:
876 JFFS2_ERROR("unknown INCOMPAT nodetype %#04X at %#08x\n",
877 je16_to_cpu(un->nodetype), ref_offset(ref));
882 case JFFS2_FEATURE_ROCOMPAT:
883 JFFS2_ERROR("unknown ROCOMPAT nodetype %#04X at %#08x\n",
884 je16_to_cpu(un->nodetype), ref_offset(ref));
885 BUG_ON(!(c->flags & JFFS2_SB_FLAG_RO));
888 case JFFS2_FEATURE_RWCOMPAT_COPY:
889 JFFS2_NOTICE("unknown RWCOMPAT_COPY nodetype %#04X at %#08x\n",
890 je16_to_cpu(un->nodetype), ref_offset(ref));
893 case JFFS2_FEATURE_RWCOMPAT_DELETE:
894 JFFS2_NOTICE("unknown RWCOMPAT_DELETE nodetype %#04X at %#08x\n",
895 je16_to_cpu(un->nodetype), ref_offset(ref));
896 jffs2_mark_node_obsolete(c, ref);
904 * Helper function for jffs2_get_inode_nodes().
905 * The function detects whether more data should be read and reads it if yes.
907 * Returns: 0 on succes;
908 * negative error code on failure.
910 static int read_more(struct jffs2_sb_info *c, struct jffs2_raw_node_ref *ref,
911 int needed_len, int *rdlen, unsigned char *buf)
913 int err, to_read = needed_len - *rdlen;
917 if (jffs2_is_writebuffered(c)) {
918 int rem = to_read % c->wbuf_pagesize;
921 to_read += c->wbuf_pagesize - rem;
924 /* We need to read more data */
925 offs = ref_offset(ref) + *rdlen;
927 dbg_readinode("read more %d bytes\n", to_read);
929 err = jffs2_flash_read(c, offs, to_read, &retlen, buf + *rdlen);
931 JFFS2_ERROR("can not read %d bytes from 0x%08x, "
932 "error code: %d.\n", to_read, offs, err);
936 if (retlen < to_read) {
937 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n",
938 offs, retlen, to_read);
946 /* Get tmp_dnode_info and full_dirent for all non-obsolete nodes associated
947 with this ino. Perform a preliminary ordering on data nodes, throwing away
948 those which are completely obsoleted by newer ones. The naïve approach we
949 use to take of just returning them _all_ in version order will cause us to
950 run out of memory in certain degenerate cases. */
951 static int jffs2_get_inode_nodes(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
952 struct jffs2_readinode_info *rii)
954 struct jffs2_raw_node_ref *ref, *valid_ref;
955 unsigned char *buf = NULL;
956 union jffs2_node_union *node;
962 dbg_readinode("ino #%u\n", f->inocache->ino);
964 /* FIXME: in case of NOR and available ->point() this
965 * needs to be fixed. */
966 len = sizeof(union jffs2_node_union) + c->wbuf_pagesize;
967 buf = kmalloc(len, GFP_KERNEL);
971 spin_lock(&c->erase_completion_lock);
972 valid_ref = jffs2_first_valid_node(f->inocache->nodes);
973 if (!valid_ref && f->inocache->ino != 1)
974 JFFS2_WARNING("Eep. No valid nodes for ino #%u.\n", f->inocache->ino);
976 /* We can hold a pointer to a non-obsolete node without the spinlock,
977 but _obsolete_ nodes may disappear at any time, if the block
978 they're in gets erased. So if we mark 'ref' obsolete while we're
979 not holding the lock, it can go away immediately. For that reason,
980 we find the next valid node first, before processing 'ref'.
983 valid_ref = jffs2_first_valid_node(ref->next_in_ino);
984 spin_unlock(&c->erase_completion_lock);
989 * At this point we don't know the type of the node we're going
990 * to read, so we do not know the size of its header. In order
991 * to minimize the amount of flash IO we assume the header is
992 * of size = JFFS2_MIN_NODE_HEADER.
994 len = JFFS2_MIN_NODE_HEADER;
995 if (jffs2_is_writebuffered(c)) {
999 * We are about to read JFFS2_MIN_NODE_HEADER bytes,
1000 * but this flash has some minimal I/O unit. It is
1001 * possible that we'll need to read more soon, so read
1002 * up to the next min. I/O unit, in order not to
1003 * re-read the same min. I/O unit twice.
1005 end = ref_offset(ref) + len;
1006 rem = end % c->wbuf_pagesize;
1008 end += c->wbuf_pagesize - rem;
1009 len = end - ref_offset(ref);
1012 dbg_readinode("read %d bytes at %#08x(%d).\n", len, ref_offset(ref), ref_flags(ref));
1014 /* FIXME: point() */
1015 err = jffs2_flash_read(c, ref_offset(ref), len, &retlen, buf);
1017 JFFS2_ERROR("can not read %d bytes from 0x%08x, " "error code: %d.\n", len, ref_offset(ref), err);
1022 JFFS2_ERROR("short read at %#08x: %zu instead of %d.\n", ref_offset(ref), retlen, len);
1027 node = (union jffs2_node_union *)buf;
1029 /* No need to mask in the valid bit; it shouldn't be invalid */
1030 if (je32_to_cpu(node->u.hdr_crc) != crc32(0, node, sizeof(node->u)-4)) {
1031 JFFS2_NOTICE("Node header CRC failed at %#08x. {%04x,%04x,%08x,%08x}\n",
1032 ref_offset(ref), je16_to_cpu(node->u.magic),
1033 je16_to_cpu(node->u.nodetype),
1034 je32_to_cpu(node->u.totlen),
1035 je32_to_cpu(node->u.hdr_crc));
1036 jffs2_dbg_dump_node(c, ref_offset(ref));
1037 jffs2_mark_node_obsolete(c, ref);
1040 if (je16_to_cpu(node->u.magic) != JFFS2_MAGIC_BITMASK) {
1041 /* Not a JFFS2 node, whinge and move on */
1042 JFFS2_NOTICE("Wrong magic bitmask 0x%04x in node header at %#08x.\n",
1043 je16_to_cpu(node->u.magic), ref_offset(ref));
1044 jffs2_mark_node_obsolete(c, ref);
1048 switch (je16_to_cpu(node->u.nodetype)) {
1050 case JFFS2_NODETYPE_DIRENT:
1052 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_dirent)) {
1053 err = read_more(c, ref, sizeof(struct jffs2_raw_dirent), &len, buf);
1058 err = read_direntry(c, ref, &node->d, retlen, rii);
1064 case JFFS2_NODETYPE_INODE:
1066 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_raw_inode)) {
1067 err = read_more(c, ref, sizeof(struct jffs2_raw_inode), &len, buf);
1072 err = read_dnode(c, ref, &node->i, len, rii);
1079 if (JFFS2_MIN_NODE_HEADER < sizeof(struct jffs2_unknown_node)) {
1080 err = read_more(c, ref, sizeof(struct jffs2_unknown_node), &len, buf);
1085 err = read_unknown(c, ref, &node->u);
1087 jffs2_mark_node_obsolete(c, ref);
1089 } else if (unlikely(err))
1094 spin_lock(&c->erase_completion_lock);
1097 spin_unlock(&c->erase_completion_lock);
1100 f->highest_version = rii->highest_version;
1102 dbg_readinode("nodes of inode #%u were read, the highest version is %u, latest_mctime %u, mctime_ver %u.\n",
1103 f->inocache->ino, rii->highest_version, rii->latest_mctime,
1108 jffs2_free_tmp_dnode_info_list(&rii->tn_root);
1109 jffs2_free_full_dirent_list(rii->fds);
1115 static int jffs2_do_read_inode_internal(struct jffs2_sb_info *c,
1116 struct jffs2_inode_info *f,
1117 struct jffs2_raw_inode *latest_node)
1119 struct jffs2_readinode_info rii;
1120 uint32_t crc, new_size;
1124 dbg_readinode("ino #%u nlink is %d\n", f->inocache->ino, f->inocache->nlink);
1126 memset(&rii, 0, sizeof(rii));
1128 /* Grab all nodes relevant to this ino */
1129 ret = jffs2_get_inode_nodes(c, f, &rii);
1132 JFFS2_ERROR("cannot read nodes for ino %u, returned error is %d\n", f->inocache->ino, ret);
1133 if (f->inocache->state == INO_STATE_READING)
1134 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1138 ret = jffs2_build_inode_fragtree(c, f, &rii);
1140 JFFS2_ERROR("Failed to build final fragtree for inode #%u: error %d\n",
1141 f->inocache->ino, ret);
1142 if (f->inocache->state == INO_STATE_READING)
1143 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1144 jffs2_free_tmp_dnode_info_list(&rii.tn_root);
1145 /* FIXME: We could at least crc-check them all */
1147 jffs2_free_full_dnode(rii.mdata_tn->fn);
1148 jffs2_free_tmp_dnode_info(rii.mdata_tn);
1149 rii.mdata_tn = NULL;
1155 if (rii.mdata_tn->fn->raw == rii.latest_ref) {
1156 f->metadata = rii.mdata_tn->fn;
1157 jffs2_free_tmp_dnode_info(rii.mdata_tn);
1159 jffs2_kill_tn(c, rii.mdata_tn);
1161 rii.mdata_tn = NULL;
1166 jffs2_dbg_fragtree_paranoia_check_nolock(f);
1168 if (unlikely(!rii.latest_ref)) {
1169 /* No data nodes for this inode. */
1170 if (f->inocache->ino != 1) {
1171 JFFS2_WARNING("no data nodes found for ino #%u\n", f->inocache->ino);
1173 if (f->inocache->state == INO_STATE_READING)
1174 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1177 JFFS2_NOTICE("but it has children so we fake some modes for it\n");
1179 latest_node->mode = cpu_to_jemode(S_IFDIR|S_IRUGO|S_IWUSR|S_IXUGO);
1180 latest_node->version = cpu_to_je32(0);
1181 latest_node->atime = latest_node->ctime = latest_node->mtime = cpu_to_je32(0);
1182 latest_node->isize = cpu_to_je32(0);
1183 latest_node->gid = cpu_to_je16(0);
1184 latest_node->uid = cpu_to_je16(0);
1185 if (f->inocache->state == INO_STATE_READING)
1186 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1190 ret = jffs2_flash_read(c, ref_offset(rii.latest_ref), sizeof(*latest_node), &retlen, (void *)latest_node);
1191 if (ret || retlen != sizeof(*latest_node)) {
1192 JFFS2_ERROR("failed to read from flash: error %d, %zd of %zd bytes read\n",
1193 ret, retlen, sizeof(*latest_node));
1194 /* FIXME: If this fails, there seems to be a memory leak. Find it. */
1196 jffs2_do_clear_inode(c, f);
1197 return ret?ret:-EIO;
1200 crc = crc32(0, latest_node, sizeof(*latest_node)-8);
1201 if (crc != je32_to_cpu(latest_node->node_crc)) {
1202 JFFS2_ERROR("CRC failed for read_inode of inode %u at physical location 0x%x\n",
1203 f->inocache->ino, ref_offset(rii.latest_ref));
1205 jffs2_do_clear_inode(c, f);
1209 switch(jemode_to_cpu(latest_node->mode) & S_IFMT) {
1211 if (rii.mctime_ver > je32_to_cpu(latest_node->version)) {
1212 /* The times in the latest_node are actually older than
1213 mctime in the latest dirent. Cheat. */
1214 latest_node->ctime = latest_node->mtime = cpu_to_je32(rii.latest_mctime);
1220 /* If it was a regular file, truncate it to the latest node's isize */
1221 new_size = jffs2_truncate_fragtree(c, &f->fragtree, je32_to_cpu(latest_node->isize));
1222 if (new_size != je32_to_cpu(latest_node->isize)) {
1223 JFFS2_WARNING("Truncating ino #%u to %d bytes failed because it only had %d bytes to start with!\n",
1224 f->inocache->ino, je32_to_cpu(latest_node->isize), new_size);
1225 latest_node->isize = cpu_to_je32(new_size);
1230 /* Hack to work around broken isize in old symlink code.
1231 Remove this when dwmw2 comes to his senses and stops
1232 symlinks from being an entirely gratuitous special
1234 if (!je32_to_cpu(latest_node->isize))
1235 latest_node->isize = latest_node->dsize;
1237 if (f->inocache->state != INO_STATE_CHECKING) {
1238 /* Symlink's inode data is the target path. Read it and
1239 * keep in RAM to facilitate quick follow symlink
1241 f->target = kmalloc(je32_to_cpu(latest_node->csize) + 1, GFP_KERNEL);
1243 JFFS2_ERROR("can't allocate %d bytes of memory for the symlink target path cache\n", je32_to_cpu(latest_node->csize));
1245 jffs2_do_clear_inode(c, f);
1249 ret = jffs2_flash_read(c, ref_offset(rii.latest_ref) + sizeof(*latest_node),
1250 je32_to_cpu(latest_node->csize), &retlen, (char *)f->target);
1252 if (ret || retlen != je32_to_cpu(latest_node->csize)) {
1253 if (retlen != je32_to_cpu(latest_node->csize))
1258 jffs2_do_clear_inode(c, f);
1262 f->target[je32_to_cpu(latest_node->csize)] = '\0';
1263 dbg_readinode("symlink's target '%s' cached\n", f->target);
1266 /* fall through... */
1270 /* Certain inode types should have only one data node, and it's
1271 kept as the metadata node */
1273 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o had metadata node\n",
1274 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1276 jffs2_do_clear_inode(c, f);
1279 if (!frag_first(&f->fragtree)) {
1280 JFFS2_ERROR("Argh. Special inode #%u with mode 0%o has no fragments\n",
1281 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1283 jffs2_do_clear_inode(c, f);
1286 /* ASSERT: f->fraglist != NULL */
1287 if (frag_next(frag_first(&f->fragtree))) {
1288 JFFS2_ERROR("Argh. Special inode #%u with mode 0x%x had more than one node\n",
1289 f->inocache->ino, jemode_to_cpu(latest_node->mode));
1290 /* FIXME: Deal with it - check crc32, check for duplicate node, check times and discard the older one */
1292 jffs2_do_clear_inode(c, f);
1295 /* OK. We're happy */
1296 f->metadata = frag_first(&f->fragtree)->node;
1297 jffs2_free_node_frag(frag_first(&f->fragtree));
1298 f->fragtree = RB_ROOT;
1301 if (f->inocache->state == INO_STATE_READING)
1302 jffs2_set_inocache_state(c, f->inocache, INO_STATE_PRESENT);
1307 /* Scan the list of all nodes present for this ino, build map of versions, etc. */
1308 int jffs2_do_read_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f,
1309 uint32_t ino, struct jffs2_raw_inode *latest_node)
1311 dbg_readinode("read inode #%u\n", ino);
1314 spin_lock(&c->inocache_lock);
1315 f->inocache = jffs2_get_ino_cache(c, ino);
1318 /* Check its state. We may need to wait before we can use it */
1319 switch(f->inocache->state) {
1320 case INO_STATE_UNCHECKED:
1321 case INO_STATE_CHECKEDABSENT:
1322 f->inocache->state = INO_STATE_READING;
1325 case INO_STATE_CHECKING:
1327 /* If it's in either of these states, we need
1328 to wait for whoever's got it to finish and
1330 dbg_readinode("waiting for ino #%u in state %d\n", ino, f->inocache->state);
1331 sleep_on_spinunlock(&c->inocache_wq, &c->inocache_lock);
1332 goto retry_inocache;
1334 case INO_STATE_READING:
1335 case INO_STATE_PRESENT:
1336 /* Eep. This should never happen. It can
1337 happen if Linux calls read_inode() again
1338 before clear_inode() has finished though. */
1339 JFFS2_ERROR("Eep. Trying to read_inode #%u when it's already in state %d!\n", ino, f->inocache->state);
1340 /* Fail. That's probably better than allowing it to succeed */
1348 spin_unlock(&c->inocache_lock);
1350 if (!f->inocache && ino == 1) {
1351 /* Special case - no root inode on medium */
1352 f->inocache = jffs2_alloc_inode_cache();
1354 JFFS2_ERROR("cannot allocate inocache for root inode\n");
1357 dbg_readinode("creating inocache for root inode\n");
1358 memset(f->inocache, 0, sizeof(struct jffs2_inode_cache));
1359 f->inocache->ino = f->inocache->nlink = 1;
1360 f->inocache->nodes = (struct jffs2_raw_node_ref *)f->inocache;
1361 f->inocache->state = INO_STATE_READING;
1362 jffs2_add_ino_cache(c, f->inocache);
1365 JFFS2_ERROR("requestied to read an nonexistent ino %u\n", ino);
1369 return jffs2_do_read_inode_internal(c, f, latest_node);
1372 int jffs2_do_crccheck_inode(struct jffs2_sb_info *c, struct jffs2_inode_cache *ic)
1374 struct jffs2_raw_inode n;
1375 struct jffs2_inode_info *f = kzalloc(sizeof(*f), GFP_KERNEL);
1381 init_MUTEX_LOCKED(&f->sem);
1384 ret = jffs2_do_read_inode_internal(c, f, &n);
1387 jffs2_do_clear_inode(c, f);
1393 void jffs2_do_clear_inode(struct jffs2_sb_info *c, struct jffs2_inode_info *f)
1395 struct jffs2_full_dirent *fd, *fds;
1399 jffs2_xattr_delete_inode(c, f->inocache);
1401 deleted = f->inocache && !f->inocache->nlink;
1403 if (f->inocache && f->inocache->state != INO_STATE_CHECKING)
1404 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CLEARING);
1408 jffs2_mark_node_obsolete(c, f->metadata->raw);
1409 jffs2_free_full_dnode(f->metadata);
1412 jffs2_kill_fragtree(&f->fragtree, deleted?c:NULL);
1423 jffs2_free_full_dirent(fd);
1426 if (f->inocache && f->inocache->state != INO_STATE_CHECKING) {
1427 jffs2_set_inocache_state(c, f->inocache, INO_STATE_CHECKEDABSENT);
1428 if (f->inocache->nodes == (void *)f->inocache)
1429 jffs2_del_ino_cache(c, f->inocache);