2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright (C) 2001-2003 Red Hat, Inc.
5 * Copyright (C) 2004 Thomas Gleixner <tglx@linutronix.de>
7 * Created by David Woodhouse <dwmw2@infradead.org>
8 * Modified debugged and enhanced by Thomas Gleixner <tglx@linutronix.de>
10 * For licensing information, see the file 'LICENCE' in this directory.
12 * $Id: wbuf.c,v 1.100 2005/09/30 13:59:13 dedekind Exp $
16 #include <linux/kernel.h>
17 #include <linux/slab.h>
18 #include <linux/mtd/mtd.h>
19 #include <linux/crc32.h>
20 #include <linux/mtd/nand.h>
21 #include <linux/jiffies.h>
25 /* For testing write failures */
30 static unsigned char *brokenbuf;
33 #define PAGE_DIV(x) ( ((unsigned long)(x) / (unsigned long)(c->wbuf_pagesize)) * (unsigned long)(c->wbuf_pagesize) )
34 #define PAGE_MOD(x) ( (unsigned long)(x) % (unsigned long)(c->wbuf_pagesize) )
36 /* max. erase failures before we mark a block bad */
37 #define MAX_ERASE_FAILURES 2
39 struct jffs2_inodirty {
41 struct jffs2_inodirty *next;
44 static struct jffs2_inodirty inodirty_nomem;
46 static int jffs2_wbuf_pending_for_ino(struct jffs2_sb_info *c, uint32_t ino)
48 struct jffs2_inodirty *this = c->wbuf_inodes;
50 /* If a malloc failed, consider _everything_ dirty */
51 if (this == &inodirty_nomem)
54 /* If ino == 0, _any_ non-GC writes mean 'yes' */
58 /* Look to see if the inode in question is pending in the wbuf */
67 static void jffs2_clear_wbuf_ino_list(struct jffs2_sb_info *c)
69 struct jffs2_inodirty *this;
71 this = c->wbuf_inodes;
73 if (this != &inodirty_nomem) {
75 struct jffs2_inodirty *next = this->next;
80 c->wbuf_inodes = NULL;
83 static void jffs2_wbuf_dirties_inode(struct jffs2_sb_info *c, uint32_t ino)
85 struct jffs2_inodirty *new;
87 /* Mark the superblock dirty so that kupdated will flush... */
88 jffs2_erase_pending_trigger(c);
90 if (jffs2_wbuf_pending_for_ino(c, ino))
93 new = kmalloc(sizeof(*new), GFP_KERNEL);
95 D1(printk(KERN_DEBUG "No memory to allocate inodirty. Fallback to all considered dirty\n"));
96 jffs2_clear_wbuf_ino_list(c);
97 c->wbuf_inodes = &inodirty_nomem;
101 new->next = c->wbuf_inodes;
102 c->wbuf_inodes = new;
106 static inline void jffs2_refile_wbuf_blocks(struct jffs2_sb_info *c)
108 struct list_head *this, *next;
111 if (list_empty(&c->erasable_pending_wbuf_list))
114 list_for_each_safe(this, next, &c->erasable_pending_wbuf_list) {
115 struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
117 D1(printk(KERN_DEBUG "Removing eraseblock at 0x%08x from erasable_pending_wbuf_list...\n", jeb->offset));
119 if ((jiffies + (n++)) & 127) {
120 /* Most of the time, we just erase it immediately. Otherwise we
121 spend ages scanning it on mount, etc. */
122 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
123 list_add_tail(&jeb->list, &c->erase_pending_list);
124 c->nr_erasing_blocks++;
125 jffs2_erase_pending_trigger(c);
127 /* Sometimes, however, we leave it elsewhere so it doesn't get
128 immediately reused, and we spread the load a bit. */
129 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
130 list_add_tail(&jeb->list, &c->erasable_list);
135 #define REFILE_NOTEMPTY 0
136 #define REFILE_ANYWAY 1
138 static void jffs2_block_refile(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, int allow_empty)
140 D1(printk("About to refile bad block at %08x\n", jeb->offset));
142 /* File the existing block on the bad_used_list.... */
143 if (c->nextblock == jeb)
145 else /* Not sure this should ever happen... need more coffee */
146 list_del(&jeb->list);
147 if (jeb->first_node) {
148 D1(printk("Refiling block at %08x to bad_used_list\n", jeb->offset));
149 list_add(&jeb->list, &c->bad_used_list);
151 BUG_ON(allow_empty == REFILE_NOTEMPTY);
152 /* It has to have had some nodes or we couldn't be here */
153 D1(printk("Refiling block at %08x to erase_pending_list\n", jeb->offset));
154 list_add(&jeb->list, &c->erase_pending_list);
155 c->nr_erasing_blocks++;
156 jffs2_erase_pending_trigger(c);
159 if (!jffs2_prealloc_raw_node_refs(c, jeb, 1)) {
160 uint32_t oldfree = jeb->free_size;
162 jffs2_link_node_ref(c, jeb,
163 (jeb->offset+c->sector_size-oldfree) | REF_OBSOLETE,
165 /* convert to wasted */
166 c->wasted_size += oldfree;
167 jeb->wasted_size += oldfree;
168 c->dirty_size -= oldfree;
169 jeb->dirty_size -= oldfree;
172 jffs2_dbg_dump_block_lists_nolock(c);
173 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
174 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
177 static struct jffs2_raw_node_ref **jffs2_incore_replace_raw(struct jffs2_sb_info *c,
178 struct jffs2_inode_info *f,
179 struct jffs2_raw_node_ref *raw,
180 union jffs2_node_union *node)
182 struct jffs2_node_frag *frag;
183 struct jffs2_full_dirent *fd;
185 dbg_noderef("incore_replace_raw: node at %p is {%04x,%04x}\n",
186 node, je16_to_cpu(node->u.magic), je16_to_cpu(node->u.nodetype));
188 BUG_ON(je16_to_cpu(node->u.magic) != 0x1985 &&
189 je16_to_cpu(node->u.magic) != 0);
191 switch (je16_to_cpu(node->u.nodetype)) {
192 case JFFS2_NODETYPE_INODE:
193 if (f->metadata && f->metadata->raw == raw) {
194 dbg_noderef("Will replace ->raw in f->metadata at %p\n", f->metadata);
195 return &f->metadata->raw;
197 frag = jffs2_lookup_node_frag(&f->fragtree, je32_to_cpu(node->i.offset));
199 /* Find a frag which refers to the full_dnode we want to modify */
200 while (!frag->node || frag->node->raw != raw) {
201 frag = frag_next(frag);
204 dbg_noderef("Will replace ->raw in full_dnode at %p\n", frag->node);
205 return &frag->node->raw;
207 case JFFS2_NODETYPE_DIRENT:
208 for (fd = f->dents; fd; fd = fd->next) {
209 if (fd->raw == raw) {
210 dbg_noderef("Will replace ->raw in full_dirent at %p\n", fd);
217 dbg_noderef("Don't care about replacing raw for nodetype %x\n",
218 je16_to_cpu(node->u.nodetype));
224 /* Recover from failure to write wbuf. Recover the nodes up to the
225 * wbuf, not the one which we were starting to try to write. */
227 static void jffs2_wbuf_recover(struct jffs2_sb_info *c)
229 struct jffs2_eraseblock *jeb, *new_jeb;
230 struct jffs2_raw_node_ref *raw, *next, *first_raw = NULL;
235 uint32_t start, end, ofs, len;
237 jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
239 spin_lock(&c->erase_completion_lock);
240 jffs2_block_refile(c, jeb, REFILE_NOTEMPTY);
241 spin_unlock(&c->erase_completion_lock);
243 BUG_ON(!ref_obsolete(jeb->last_node));
245 /* Find the first node to be recovered, by skipping over every
246 node which ends before the wbuf starts, or which is obsolete. */
247 for (next = raw = jeb->first_node; next; raw = next) {
248 next = ref_next(raw);
250 if (ref_obsolete(raw) ||
251 (next && ref_offset(next) <= c->wbuf_ofs)) {
252 dbg_noderef("Skipping node at 0x%08x(%d)-0x%08x which is either before 0x%08x or obsolete\n",
253 ref_offset(raw), ref_flags(raw),
254 (ref_offset(raw) + ref_totlen(c, jeb, raw)),
258 dbg_noderef("First node to be recovered is at 0x%08x(%d)-0x%08x\n",
259 ref_offset(raw), ref_flags(raw),
260 (ref_offset(raw) + ref_totlen(c, jeb, raw)));
267 /* All nodes were obsolete. Nothing to recover. */
268 D1(printk(KERN_DEBUG "No non-obsolete nodes to be recovered. Just filing block bad\n"));
273 start = ref_offset(first_raw);
274 end = ref_offset(jeb->last_node);
277 /* Count the number of refs which need to be copied */
278 while ((raw = ref_next(raw)) != jeb->last_node)
281 dbg_noderef("wbuf recover %08x-%08x (%d bytes in %d nodes)\n",
282 start, end, end - start, nr_refile);
285 if (start < c->wbuf_ofs) {
286 /* First affected node was already partially written.
287 * Attempt to reread the old data into our buffer. */
289 buf = kmalloc(end - start, GFP_KERNEL);
291 printk(KERN_CRIT "Malloc failure in wbuf recovery. Data loss ensues.\n");
297 ret = c->mtd->read(c->mtd, start, c->wbuf_ofs - start, &retlen, buf);
299 /* ECC recovered ? */
300 if ((ret == -EUCLEAN || ret == -EBADMSG) &&
301 (retlen == c->wbuf_ofs - start))
304 if (ret || retlen != c->wbuf_ofs - start) {
305 printk(KERN_CRIT "Old data are already lost in wbuf recovery. Data loss ensues.\n");
310 first_raw = ref_next(first_raw);
312 while (first_raw && ref_obsolete(first_raw)) {
313 first_raw = ref_next(first_raw);
317 /* If this was the only node to be recovered, give up */
323 /* It wasn't. Go on and try to recover nodes complete in the wbuf */
324 start = ref_offset(first_raw);
325 dbg_noderef("wbuf now recover %08x-%08x (%d bytes in %d nodes)\n",
326 start, end, end - start, nr_refile);
329 /* Read succeeded. Copy the remaining data from the wbuf */
330 memcpy(buf + (c->wbuf_ofs - start), c->wbuf, end - c->wbuf_ofs);
333 /* OK... we're to rewrite (end-start) bytes of data from first_raw onwards.
334 Either 'buf' contains the data, or we find it in the wbuf */
336 /* ... and get an allocation of space from a shiny new block instead */
337 ret = jffs2_reserve_space_gc(c, end-start, &len, JFFS2_SUMMARY_NOSUM_SIZE);
339 printk(KERN_WARNING "Failed to allocate space for wbuf recovery. Data loss ensues.\n");
344 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, nr_refile);
346 printk(KERN_WARNING "Failed to allocate node refs for wbuf recovery. Data loss ensues.\n");
353 if (end-start >= c->wbuf_pagesize) {
354 /* Need to do another write immediately, but it's possible
355 that this is just because the wbuf itself is completely
356 full, and there's nothing earlier read back from the
357 flash. Hence 'buf' isn't necessarily what we're writing
359 unsigned char *rewrite_buf = buf?:c->wbuf;
360 uint32_t towrite = (end-start) - ((end-start)%c->wbuf_pagesize);
362 D1(printk(KERN_DEBUG "Write 0x%x bytes at 0x%08x in wbuf recover\n",
367 if (breakme++ == 20) {
368 printk(KERN_NOTICE "Faking write error at 0x%08x\n", ofs);
370 c->mtd->write(c->mtd, ofs, towrite, &retlen,
375 ret = c->mtd->write(c->mtd, ofs, towrite, &retlen,
378 if (ret || retlen != towrite) {
379 /* Argh. We tried. Really we did. */
380 printk(KERN_CRIT "Recovery of wbuf failed due to a second write error\n");
384 jffs2_add_physical_node_ref(c, ofs | REF_OBSOLETE, ref_totlen(c, jeb, first_raw), NULL);
388 printk(KERN_NOTICE "Recovery of wbuf succeeded to %08x\n", ofs);
390 c->wbuf_len = (end - start) - towrite;
391 c->wbuf_ofs = ofs + towrite;
392 memmove(c->wbuf, rewrite_buf + towrite, c->wbuf_len);
393 /* Don't muck about with c->wbuf_inodes. False positives are harmless. */
395 /* OK, now we're left with the dregs in whichever buffer we're using */
397 memcpy(c->wbuf, buf, end-start);
399 memmove(c->wbuf, c->wbuf + (start - c->wbuf_ofs), end - start);
402 c->wbuf_len = end - start;
405 /* Now sort out the jffs2_raw_node_refs, moving them from the old to the next block */
406 new_jeb = &c->blocks[ofs / c->sector_size];
408 spin_lock(&c->erase_completion_lock);
409 for (raw = first_raw; raw != jeb->last_node; raw = ref_next(raw)) {
410 uint32_t rawlen = ref_totlen(c, jeb, raw);
411 struct jffs2_inode_cache *ic;
412 struct jffs2_raw_node_ref *new_ref;
413 struct jffs2_raw_node_ref **adjust_ref = NULL;
414 struct jffs2_inode_info *f = NULL;
416 D1(printk(KERN_DEBUG "Refiling block of %08x at %08x(%d) to %08x\n",
417 rawlen, ref_offset(raw), ref_flags(raw), ofs));
419 ic = jffs2_raw_ref_to_ic(raw);
421 /* Ick. This XATTR mess should be fixed shortly... */
422 if (ic && ic->class == RAWNODE_CLASS_XATTR_DATUM) {
423 struct jffs2_xattr_datum *xd = (void *)ic;
424 BUG_ON(xd->node != raw);
425 adjust_ref = &xd->node;
426 raw->next_in_ino = NULL;
428 } else if (ic && ic->class == RAWNODE_CLASS_XATTR_REF) {
429 struct jffs2_xattr_datum *xr = (void *)ic;
430 BUG_ON(xr->node != raw);
431 adjust_ref = &xr->node;
432 raw->next_in_ino = NULL;
434 } else if (ic && ic->class == RAWNODE_CLASS_INODE_CACHE) {
435 struct jffs2_raw_node_ref **p = &ic->nodes;
437 /* Remove the old node from the per-inode list */
438 while (*p && *p != (void *)ic) {
440 (*p) = (raw->next_in_ino);
441 raw->next_in_ino = NULL;
444 p = &((*p)->next_in_ino);
447 if (ic->state == INO_STATE_PRESENT && !ref_obsolete(raw)) {
448 /* If it's an in-core inode, then we have to adjust any
449 full_dirent or full_dnode structure to point to the
450 new version instead of the old */
451 f = jffs2_gc_fetch_inode(c, ic->ino, ic->nlink);
453 /* Should never happen; it _must_ be present */
454 JFFS2_ERROR("Failed to iget() ino #%u, err %ld\n",
455 ic->ino, PTR_ERR(f));
458 /* We don't lock f->sem. There's a number of ways we could
459 end up in here with it already being locked, and nobody's
460 going to modify it on us anyway because we hold the
461 alloc_sem. We're only changing one ->raw pointer too,
462 which we can get away with without upsetting readers. */
463 adjust_ref = jffs2_incore_replace_raw(c, f, raw,
464 (void *)(buf?:c->wbuf) + (ref_offset(raw) - start));
465 } else if (unlikely(ic->state != INO_STATE_PRESENT &&
466 ic->state != INO_STATE_CHECKEDABSENT &&
467 ic->state != INO_STATE_GC)) {
468 JFFS2_ERROR("Inode #%u is in strange state %d!\n", ic->ino, ic->state);
473 new_ref = jffs2_link_node_ref(c, new_jeb, ofs | ref_flags(raw), rawlen, ic);
476 BUG_ON(*adjust_ref != raw);
477 *adjust_ref = new_ref;
480 jffs2_gc_release_inode(c, f);
482 if (!ref_obsolete(raw)) {
483 jeb->dirty_size += rawlen;
484 jeb->used_size -= rawlen;
485 c->dirty_size += rawlen;
486 c->used_size -= rawlen;
487 raw->flash_offset = ref_offset(raw) | REF_OBSOLETE;
488 BUG_ON(raw->next_in_ino);
495 /* Fix up the original jeb now it's on the bad_list */
496 if (first_raw == jeb->first_node) {
497 D1(printk(KERN_DEBUG "Failing block at %08x is now empty. Moving to erase_pending_list\n", jeb->offset));
498 list_move(&jeb->list, &c->erase_pending_list);
499 c->nr_erasing_blocks++;
500 jffs2_erase_pending_trigger(c);
503 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
504 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
506 jffs2_dbg_acct_sanity_check_nolock(c, new_jeb);
507 jffs2_dbg_acct_paranoia_check_nolock(c, new_jeb);
509 spin_unlock(&c->erase_completion_lock);
511 D1(printk(KERN_DEBUG "wbuf recovery completed OK. wbuf_ofs 0x%08x, len 0x%x\n", c->wbuf_ofs, c->wbuf_len));
515 /* Meaning of pad argument:
516 0: Do not pad. Probably pointless - we only ever use this when we can't pad anyway.
517 1: Pad, do not adjust nextblock free_size
518 2: Pad, adjust nextblock free_size
521 #define PAD_NOACCOUNT 1
522 #define PAD_ACCOUNTING 2
524 static int __jffs2_flush_wbuf(struct jffs2_sb_info *c, int pad)
526 struct jffs2_eraseblock *wbuf_jeb;
530 /* Nothing to do if not write-buffering the flash. In particular, we shouldn't
531 del_timer() the timer we never initialised. */
532 if (!jffs2_is_writebuffered(c))
535 if (!down_trylock(&c->alloc_sem)) {
537 printk(KERN_CRIT "jffs2_flush_wbuf() called with alloc_sem not locked!\n");
541 if (!c->wbuf_len) /* already checked c->wbuf above */
544 wbuf_jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
545 if (jffs2_prealloc_raw_node_refs(c, wbuf_jeb, c->nextblock->allocated_refs + 1))
548 /* claim remaining space on the page
549 this happens, if we have a change to a new block,
550 or if fsync forces us to flush the writebuffer.
551 if we have a switch to next page, we will not have
552 enough remaining space for this.
555 c->wbuf_len = PAD(c->wbuf_len);
557 /* Pad with JFFS2_DIRTY_BITMASK initially. this helps out ECC'd NOR
558 with 8 byte page size */
559 memset(c->wbuf + c->wbuf_len, 0, c->wbuf_pagesize - c->wbuf_len);
561 if ( c->wbuf_len + sizeof(struct jffs2_unknown_node) < c->wbuf_pagesize) {
562 struct jffs2_unknown_node *padnode = (void *)(c->wbuf + c->wbuf_len);
563 padnode->magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
564 padnode->nodetype = cpu_to_je16(JFFS2_NODETYPE_PADDING);
565 padnode->totlen = cpu_to_je32(c->wbuf_pagesize - c->wbuf_len);
566 padnode->hdr_crc = cpu_to_je32(crc32(0, padnode, sizeof(*padnode)-4));
569 /* else jffs2_flash_writev has actually filled in the rest of the
570 buffer for us, and will deal with the node refs etc. later. */
574 if (breakme++ == 20) {
575 printk(KERN_NOTICE "Faking write error at 0x%08x\n", c->wbuf_ofs);
577 c->mtd->write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen,
583 ret = c->mtd->write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen, c->wbuf);
585 if (ret || retlen != c->wbuf_pagesize) {
587 printk(KERN_WARNING "jffs2_flush_wbuf(): Write failed with %d\n",ret);
589 printk(KERN_WARNING "jffs2_flush_wbuf(): Write was short: %zd instead of %d\n",
590 retlen, c->wbuf_pagesize);
594 jffs2_wbuf_recover(c);
599 /* Adjust free size of the block if we padded. */
601 uint32_t waste = c->wbuf_pagesize - c->wbuf_len;
603 D1(printk(KERN_DEBUG "jffs2_flush_wbuf() adjusting free_size of %sblock at %08x\n",
604 (wbuf_jeb==c->nextblock)?"next":"", wbuf_jeb->offset));
606 /* wbuf_pagesize - wbuf_len is the amount of space that's to be
607 padded. If there is less free space in the block than that,
608 something screwed up */
609 if (wbuf_jeb->free_size < waste) {
610 printk(KERN_CRIT "jffs2_flush_wbuf(): Accounting error. wbuf at 0x%08x has 0x%03x bytes, 0x%03x left.\n",
611 c->wbuf_ofs, c->wbuf_len, waste);
612 printk(KERN_CRIT "jffs2_flush_wbuf(): But free_size for block at 0x%08x is only 0x%08x\n",
613 wbuf_jeb->offset, wbuf_jeb->free_size);
617 spin_lock(&c->erase_completion_lock);
619 jffs2_link_node_ref(c, wbuf_jeb, (c->wbuf_ofs + c->wbuf_len) | REF_OBSOLETE, waste, NULL);
620 /* FIXME: that made it count as dirty. Convert to wasted */
621 wbuf_jeb->dirty_size -= waste;
622 c->dirty_size -= waste;
623 wbuf_jeb->wasted_size += waste;
624 c->wasted_size += waste;
626 spin_lock(&c->erase_completion_lock);
628 /* Stick any now-obsoleted blocks on the erase_pending_list */
629 jffs2_refile_wbuf_blocks(c);
630 jffs2_clear_wbuf_ino_list(c);
631 spin_unlock(&c->erase_completion_lock);
633 memset(c->wbuf,0xff,c->wbuf_pagesize);
634 /* adjust write buffer offset, else we get a non contiguous write bug */
635 c->wbuf_ofs += c->wbuf_pagesize;
640 /* Trigger garbage collection to flush the write-buffer.
641 If ino arg is zero, do it if _any_ real (i.e. not GC) writes are
642 outstanding. If ino arg non-zero, do it only if a write for the
643 given inode is outstanding. */
644 int jffs2_flush_wbuf_gc(struct jffs2_sb_info *c, uint32_t ino)
646 uint32_t old_wbuf_ofs;
647 uint32_t old_wbuf_len;
650 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() called for ino #%u...\n", ino));
656 if (!jffs2_wbuf_pending_for_ino(c, ino)) {
657 D1(printk(KERN_DEBUG "Ino #%d not pending in wbuf. Returning\n", ino));
662 old_wbuf_ofs = c->wbuf_ofs;
663 old_wbuf_len = c->wbuf_len;
665 if (c->unchecked_size) {
666 /* GC won't make any progress for a while */
667 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() padding. Not finished checking\n"));
668 down_write(&c->wbuf_sem);
669 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
670 /* retry flushing wbuf in case jffs2_wbuf_recover
671 left some data in the wbuf */
673 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
674 up_write(&c->wbuf_sem);
675 } else while (old_wbuf_len &&
676 old_wbuf_ofs == c->wbuf_ofs) {
680 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() calls gc pass\n"));
682 ret = jffs2_garbage_collect_pass(c);
684 /* GC failed. Flush it with padding instead */
686 down_write(&c->wbuf_sem);
687 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
688 /* retry flushing wbuf in case jffs2_wbuf_recover
689 left some data in the wbuf */
691 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
692 up_write(&c->wbuf_sem);
698 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() ends...\n"));
704 /* Pad write-buffer to end and write it, wasting space. */
705 int jffs2_flush_wbuf_pad(struct jffs2_sb_info *c)
712 down_write(&c->wbuf_sem);
713 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
714 /* retry - maybe wbuf recover left some data in wbuf. */
716 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
717 up_write(&c->wbuf_sem);
722 static size_t jffs2_fill_wbuf(struct jffs2_sb_info *c, const uint8_t *buf,
725 if (len && !c->wbuf_len && (len >= c->wbuf_pagesize))
728 if (len > (c->wbuf_pagesize - c->wbuf_len))
729 len = c->wbuf_pagesize - c->wbuf_len;
730 memcpy(c->wbuf + c->wbuf_len, buf, len);
731 c->wbuf_len += (uint32_t) len;
735 int jffs2_flash_writev(struct jffs2_sb_info *c, const struct kvec *invecs,
736 unsigned long count, loff_t to, size_t *retlen,
739 struct jffs2_eraseblock *jeb;
740 size_t wbuf_retlen, donelen = 0;
741 uint32_t outvec_to = to;
744 /* If not writebuffered flash, don't bother */
745 if (!jffs2_is_writebuffered(c))
746 return jffs2_flash_direct_writev(c, invecs, count, to, retlen);
748 down_write(&c->wbuf_sem);
750 /* If wbuf_ofs is not initialized, set it to target address */
751 if (c->wbuf_ofs == 0xFFFFFFFF) {
752 c->wbuf_ofs = PAGE_DIV(to);
753 c->wbuf_len = PAGE_MOD(to);
754 memset(c->wbuf,0xff,c->wbuf_pagesize);
758 * Sanity checks on target address. It's permitted to write
759 * at PAD(c->wbuf_len+c->wbuf_ofs), and it's permitted to
760 * write at the beginning of a new erase block. Anything else,
761 * and you die. New block starts at xxx000c (0-b = block
764 if (SECTOR_ADDR(to) != SECTOR_ADDR(c->wbuf_ofs)) {
765 /* It's a write to a new block */
767 D1(printk(KERN_DEBUG "jffs2_flash_writev() to 0x%lx "
768 "causes flush of wbuf at 0x%08x\n",
769 (unsigned long)to, c->wbuf_ofs));
770 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
774 /* set pointer to new block */
775 c->wbuf_ofs = PAGE_DIV(to);
776 c->wbuf_len = PAGE_MOD(to);
779 if (to != PAD(c->wbuf_ofs + c->wbuf_len)) {
780 /* We're not writing immediately after the writebuffer. Bad. */
781 printk(KERN_CRIT "jffs2_flash_writev(): Non-contiguous write "
782 "to %08lx\n", (unsigned long)to);
784 printk(KERN_CRIT "wbuf was previously %08x-%08x\n",
785 c->wbuf_ofs, c->wbuf_ofs+c->wbuf_len);
789 /* adjust alignment offset */
790 if (c->wbuf_len != PAGE_MOD(to)) {
791 c->wbuf_len = PAGE_MOD(to);
792 /* take care of alignment to next page */
794 c->wbuf_len = c->wbuf_pagesize;
795 ret = __jffs2_flush_wbuf(c, NOPAD);
801 for (invec = 0; invec < count; invec++) {
802 int vlen = invecs[invec].iov_len;
803 uint8_t *v = invecs[invec].iov_base;
805 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
807 if (c->wbuf_len == c->wbuf_pagesize) {
808 ret = __jffs2_flush_wbuf(c, NOPAD);
813 outvec_to += wbuf_retlen;
814 donelen += wbuf_retlen;
817 if (vlen >= c->wbuf_pagesize) {
818 ret = c->mtd->write(c->mtd, outvec_to, PAGE_DIV(vlen),
820 if (ret < 0 || wbuf_retlen != PAGE_DIV(vlen))
824 outvec_to += wbuf_retlen;
825 c->wbuf_ofs = outvec_to;
826 donelen += wbuf_retlen;
830 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
831 if (c->wbuf_len == c->wbuf_pagesize) {
832 ret = __jffs2_flush_wbuf(c, NOPAD);
837 outvec_to += wbuf_retlen;
838 donelen += wbuf_retlen;
842 * If there's a remainder in the wbuf and it's a non-GC write,
843 * remember that the wbuf affects this ino
847 if (jffs2_sum_active()) {
848 int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
853 if (c->wbuf_len && ino)
854 jffs2_wbuf_dirties_inode(c, ino);
857 up_write(&c->wbuf_sem);
862 * At this point we have no problem, c->wbuf is empty. However
863 * refile nextblock to avoid writing again to same address.
866 spin_lock(&c->erase_completion_lock);
868 jeb = &c->blocks[outvec_to / c->sector_size];
869 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
871 spin_unlock(&c->erase_completion_lock);
875 up_write(&c->wbuf_sem);
880 * This is the entry for flash write.
881 * Check, if we work on NAND FLASH, if so build an kvec and write it via vritev
883 int jffs2_flash_write(struct jffs2_sb_info *c, loff_t ofs, size_t len,
884 size_t *retlen, const u_char *buf)
888 if (!jffs2_is_writebuffered(c))
889 return jffs2_flash_direct_write(c, ofs, len, retlen, buf);
891 vecs[0].iov_base = (unsigned char *) buf;
892 vecs[0].iov_len = len;
893 return jffs2_flash_writev(c, vecs, 1, ofs, retlen, 0);
897 Handle readback from writebuffer and ECC failure return
899 int jffs2_flash_read(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, u_char *buf)
901 loff_t orbf = 0, owbf = 0, lwbf = 0;
904 if (!jffs2_is_writebuffered(c))
905 return c->mtd->read(c->mtd, ofs, len, retlen, buf);
908 down_read(&c->wbuf_sem);
909 ret = c->mtd->read(c->mtd, ofs, len, retlen, buf);
911 if ( (ret == -EBADMSG || ret == -EUCLEAN) && (*retlen == len) ) {
913 printk(KERN_WARNING "mtd->read(0x%zx bytes from 0x%llx)"
914 " returned ECC error\n", len, ofs);
916 * We have the raw data without ECC correction in the buffer,
917 * maybe we are lucky and all data or parts are correct. We
918 * check the node. If data are corrupted node check will sort
919 * it out. We keep this block, it will fail on write or erase
920 * and the we mark it bad. Or should we do that now? But we
921 * should give him a chance. Maybe we had a system crash or
922 * power loss before the ecc write or a erase was completed.
923 * So we return success. :)
928 /* if no writebuffer available or write buffer empty, return */
929 if (!c->wbuf_pagesize || !c->wbuf_len)
932 /* if we read in a different block, return */
933 if (SECTOR_ADDR(ofs) != SECTOR_ADDR(c->wbuf_ofs))
936 if (ofs >= c->wbuf_ofs) {
937 owbf = (ofs - c->wbuf_ofs); /* offset in write buffer */
938 if (owbf > c->wbuf_len) /* is read beyond write buffer ? */
940 lwbf = c->wbuf_len - owbf; /* number of bytes to copy */
944 orbf = (c->wbuf_ofs - ofs); /* offset in read buffer */
945 if (orbf > len) /* is write beyond write buffer ? */
947 lwbf = len - orbf; /* number of bytes to copy */
948 if (lwbf > c->wbuf_len)
952 memcpy(buf+orbf,c->wbuf+owbf,lwbf);
955 up_read(&c->wbuf_sem);
959 #define NR_OOB_SCAN_PAGES 4
962 * Check, if the out of band area is empty
964 int jffs2_check_oob_empty(struct jffs2_sb_info *c,
965 struct jffs2_eraseblock *jeb, int mode)
968 int oobsize = c->mtd->oobsize;
969 struct mtd_oob_ops ops;
971 ops.len = NR_OOB_SCAN_PAGES * oobsize;
972 ops.ooblen = oobsize;
973 ops.oobbuf = c->oobbuf;
976 ops.mode = MTD_OOB_PLACE;
978 ret = c->mtd->read_oob(c->mtd, jeb->offset, &ops);
980 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB "
981 "failed %d for block at %08x\n", ret, jeb->offset));
985 if (ops.retlen < ops.len) {
986 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB "
987 "returned short read (%zd bytes not %d) for block "
988 "at %08x\n", ops.retlen, ops.len, jeb->offset));
992 /* Special check for first page */
993 for(i = 0; i < oobsize ; i++) {
994 /* Yeah, we know about the cleanmarker. */
995 if (mode && i >= c->fsdata_pos &&
996 i < c->fsdata_pos + c->fsdata_len)
999 if (ops.oobbuf[i] != 0xFF) {
1000 D2(printk(KERN_DEBUG "Found %02x at %x in OOB for "
1001 "%08x\n", ops.oobbuf[i], i, jeb->offset));
1006 /* we know, we are aligned :) */
1007 for (page = oobsize; page < ops.len; page += sizeof(long)) {
1008 long dat = *(long *)(&ops.oobbuf[page]);
1016 * Scan for a valid cleanmarker and for bad blocks
1018 int jffs2_check_nand_cleanmarker (struct jffs2_sb_info *c,
1019 struct jffs2_eraseblock *jeb)
1021 struct jffs2_unknown_node n;
1022 struct mtd_oob_ops ops;
1023 int oobsize = c->mtd->oobsize;
1024 unsigned char *p,*b;
1026 size_t offset = jeb->offset;
1028 /* Check first if the block is bad. */
1029 if (c->mtd->block_isbad(c->mtd, offset)) {
1030 D1 (printk(KERN_WARNING "jffs2_check_nand_cleanmarker()"
1031 ": Bad block at %08x\n", jeb->offset));
1036 ops.ooblen = oobsize;
1037 ops.oobbuf = c->oobbuf;
1040 ops.mode = MTD_OOB_PLACE;
1042 ret = c->mtd->read_oob(c->mtd, offset, &ops);
1044 D1 (printk(KERN_WARNING "jffs2_check_nand_cleanmarker(): "
1045 "Read OOB failed %d for block at %08x\n",
1050 if (ops.retlen < ops.len) {
1051 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): "
1052 "Read OOB return short read (%zd bytes not %d) "
1053 "for block at %08x\n", ops.retlen, ops.len,
1058 n.magic = cpu_to_je16 (JFFS2_MAGIC_BITMASK);
1059 n.nodetype = cpu_to_je16 (JFFS2_NODETYPE_CLEANMARKER);
1060 n.totlen = cpu_to_je32 (8);
1061 p = (unsigned char *) &n;
1062 b = c->oobbuf + c->fsdata_pos;
1064 for (i = c->fsdata_len; i; i--) {
1070 printk(KERN_WARNING "jffs2_check_nand_cleanmarker(): "
1071 "Cleanmarker node not detected in block at %08x\n",
1073 printk(KERN_WARNING "OOB at %08zx was ", offset);
1074 for (i=0; i < oobsize; i++)
1075 printk("%02x ", c->oobbuf[i]);
1081 int jffs2_write_nand_cleanmarker(struct jffs2_sb_info *c,
1082 struct jffs2_eraseblock *jeb)
1084 struct jffs2_unknown_node n;
1086 struct mtd_oob_ops ops;
1088 n.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1089 n.nodetype = cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER);
1090 n.totlen = cpu_to_je32(8);
1092 ops.len = c->fsdata_len;
1093 ops.ooblen = c->fsdata_len;;
1094 ops.oobbuf = (uint8_t *)&n;
1095 ops.ooboffs = c->fsdata_pos;
1097 ops.mode = MTD_OOB_PLACE;
1099 ret = c->mtd->write_oob(c->mtd, jeb->offset, &ops);
1102 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): "
1103 "Write failed for block at %08x: error %d\n",
1107 if (ops.retlen != ops.len) {
1108 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): "
1109 "Short write for block at %08x: %zd not %d\n",
1110 jeb->offset, ops.retlen, ops.len));
1117 * On NAND we try to mark this block bad. If the block was erased more
1118 * than MAX_ERASE_FAILURES we mark it finaly bad.
1119 * Don't care about failures. This block remains on the erase-pending
1120 * or badblock list as long as nobody manipulates the flash with
1121 * a bootloader or something like that.
1124 int jffs2_write_nand_badblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
1128 /* if the count is < max, we try to write the counter to the 2nd page oob area */
1129 if( ++jeb->bad_count < MAX_ERASE_FAILURES)
1132 if (!c->mtd->block_markbad)
1133 return 1; // What else can we do?
1135 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Marking bad block at %08x\n", bad_offset));
1136 ret = c->mtd->block_markbad(c->mtd, bad_offset);
1139 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1145 static int jffs2_nand_set_oobinfo(struct jffs2_sb_info *c)
1147 struct nand_ecclayout *oinfo = c->mtd->ecclayout;
1149 /* Do this only, if we have an oob buffer */
1150 if (!c->mtd->oobsize)
1153 /* Cleanmarker is out-of-band, so inline size zero */
1154 c->cleanmarker_size = 0;
1156 /* Should we use autoplacement ? */
1158 D1(printk(KERN_DEBUG "JFFS2 on NAND. No autoplacment info found\n"));
1162 D1(printk(KERN_DEBUG "JFFS2 using autoplace on NAND\n"));
1163 /* Get the position of the free bytes */
1164 if (!oinfo->oobfree[0].length) {
1165 printk (KERN_WARNING "jffs2_nand_set_oobinfo(): Eeep."
1166 " Autoplacement selected and no empty space in oob\n");
1169 c->fsdata_pos = oinfo->oobfree[0].offset;
1170 c->fsdata_len = oinfo->oobfree[0].length;
1171 if (c->fsdata_len > 8)
1177 int jffs2_nand_flash_setup(struct jffs2_sb_info *c)
1181 /* Initialise write buffer */
1182 init_rwsem(&c->wbuf_sem);
1183 c->wbuf_pagesize = c->mtd->writesize;
1184 c->wbuf_ofs = 0xFFFFFFFF;
1186 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1190 c->oobbuf = kmalloc(NR_OOB_SCAN_PAGES * c->mtd->oobsize, GFP_KERNEL);
1194 res = jffs2_nand_set_oobinfo(c);
1198 brokenbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1203 memset(brokenbuf, 0xdb, c->wbuf_pagesize);
1208 void jffs2_nand_flash_cleanup(struct jffs2_sb_info *c)
1214 int jffs2_dataflash_setup(struct jffs2_sb_info *c) {
1215 c->cleanmarker_size = 0; /* No cleanmarkers needed */
1217 /* Initialize write buffer */
1218 init_rwsem(&c->wbuf_sem);
1221 c->wbuf_pagesize = c->mtd->erasesize;
1223 /* Find a suitable c->sector_size
1224 * - Not too much sectors
1225 * - Sectors have to be at least 4 K + some bytes
1226 * - All known dataflashes have erase sizes of 528 or 1056
1227 * - we take at least 8 eraseblocks and want to have at least 8K size
1228 * - The concatenation should be a power of 2
1231 c->sector_size = 8 * c->mtd->erasesize;
1233 while (c->sector_size < 8192) {
1234 c->sector_size *= 2;
1237 /* It may be necessary to adjust the flash size */
1238 c->flash_size = c->mtd->size;
1240 if ((c->flash_size % c->sector_size) != 0) {
1241 c->flash_size = (c->flash_size / c->sector_size) * c->sector_size;
1242 printk(KERN_WARNING "JFFS2 flash size adjusted to %dKiB\n", c->flash_size);
1245 c->wbuf_ofs = 0xFFFFFFFF;
1246 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1250 printk(KERN_INFO "JFFS2 write-buffering enabled buffer (%d) erasesize (%d)\n", c->wbuf_pagesize, c->sector_size);
1255 void jffs2_dataflash_cleanup(struct jffs2_sb_info *c) {
1259 int jffs2_nor_wbuf_flash_setup(struct jffs2_sb_info *c) {
1260 /* Cleanmarker currently occupies whole programming regions,
1261 * either one or 2 for 8Byte STMicro flashes. */
1262 c->cleanmarker_size = max(16u, c->mtd->writesize);
1264 /* Initialize write buffer */
1265 init_rwsem(&c->wbuf_sem);
1266 c->wbuf_pagesize = c->mtd->writesize;
1267 c->wbuf_ofs = 0xFFFFFFFF;
1269 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1276 void jffs2_nor_wbuf_flash_cleanup(struct jffs2_sb_info *c) {