2 * JFFS2 -- Journalling Flash File System, Version 2.
4 * Copyright © 2001-2007 Red Hat, Inc.
5 * Copyright © 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.
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/mtd/mtd.h>
17 #include <linux/crc32.h>
18 #include <linux/mtd/nand.h>
19 #include <linux/jiffies.h>
20 #include <linux/sched.h>
24 /* For testing write failures */
29 static unsigned char *brokenbuf;
32 #define PAGE_DIV(x) ( ((unsigned long)(x) / (unsigned long)(c->wbuf_pagesize)) * (unsigned long)(c->wbuf_pagesize) )
33 #define PAGE_MOD(x) ( (unsigned long)(x) % (unsigned long)(c->wbuf_pagesize) )
35 /* max. erase failures before we mark a block bad */
36 #define MAX_ERASE_FAILURES 2
38 struct jffs2_inodirty {
40 struct jffs2_inodirty *next;
43 static struct jffs2_inodirty inodirty_nomem;
45 static int jffs2_wbuf_pending_for_ino(struct jffs2_sb_info *c, uint32_t ino)
47 struct jffs2_inodirty *this = c->wbuf_inodes;
49 /* If a malloc failed, consider _everything_ dirty */
50 if (this == &inodirty_nomem)
53 /* If ino == 0, _any_ non-GC writes mean 'yes' */
57 /* Look to see if the inode in question is pending in the wbuf */
66 static void jffs2_clear_wbuf_ino_list(struct jffs2_sb_info *c)
68 struct jffs2_inodirty *this;
70 this = c->wbuf_inodes;
72 if (this != &inodirty_nomem) {
74 struct jffs2_inodirty *next = this->next;
79 c->wbuf_inodes = NULL;
82 static void jffs2_wbuf_dirties_inode(struct jffs2_sb_info *c, uint32_t ino)
84 struct jffs2_inodirty *new;
86 /* Mark the superblock dirty so that kupdated will flush... */
87 jffs2_erase_pending_trigger(c);
89 if (jffs2_wbuf_pending_for_ino(c, ino))
92 new = kmalloc(sizeof(*new), GFP_KERNEL);
94 D1(printk(KERN_DEBUG "No memory to allocate inodirty. Fallback to all considered dirty\n"));
95 jffs2_clear_wbuf_ino_list(c);
96 c->wbuf_inodes = &inodirty_nomem;
100 new->next = c->wbuf_inodes;
101 c->wbuf_inodes = new;
105 static inline void jffs2_refile_wbuf_blocks(struct jffs2_sb_info *c)
107 struct list_head *this, *next;
110 if (list_empty(&c->erasable_pending_wbuf_list))
113 list_for_each_safe(this, next, &c->erasable_pending_wbuf_list) {
114 struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
116 D1(printk(KERN_DEBUG "Removing eraseblock at 0x%08x from erasable_pending_wbuf_list...\n", jeb->offset));
118 if ((jiffies + (n++)) & 127) {
119 /* Most of the time, we just erase it immediately. Otherwise we
120 spend ages scanning it on mount, etc. */
121 D1(printk(KERN_DEBUG "...and adding to erase_pending_list\n"));
122 list_add_tail(&jeb->list, &c->erase_pending_list);
123 c->nr_erasing_blocks++;
124 jffs2_erase_pending_trigger(c);
126 /* Sometimes, however, we leave it elsewhere so it doesn't get
127 immediately reused, and we spread the load a bit. */
128 D1(printk(KERN_DEBUG "...and adding to erasable_list\n"));
129 list_add_tail(&jeb->list, &c->erasable_list);
134 #define REFILE_NOTEMPTY 0
135 #define REFILE_ANYWAY 1
137 static void jffs2_block_refile(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, int allow_empty)
139 D1(printk("About to refile bad block at %08x\n", jeb->offset));
141 /* File the existing block on the bad_used_list.... */
142 if (c->nextblock == jeb)
144 else /* Not sure this should ever happen... need more coffee */
145 list_del(&jeb->list);
146 if (jeb->first_node) {
147 D1(printk("Refiling block at %08x to bad_used_list\n", jeb->offset));
148 list_add(&jeb->list, &c->bad_used_list);
150 BUG_ON(allow_empty == REFILE_NOTEMPTY);
151 /* It has to have had some nodes or we couldn't be here */
152 D1(printk("Refiling block at %08x to erase_pending_list\n", jeb->offset));
153 list_add(&jeb->list, &c->erase_pending_list);
154 c->nr_erasing_blocks++;
155 jffs2_erase_pending_trigger(c);
158 if (!jffs2_prealloc_raw_node_refs(c, jeb, 1)) {
159 uint32_t oldfree = jeb->free_size;
161 jffs2_link_node_ref(c, jeb,
162 (jeb->offset+c->sector_size-oldfree) | REF_OBSOLETE,
164 /* convert to wasted */
165 c->wasted_size += oldfree;
166 jeb->wasted_size += oldfree;
167 c->dirty_size -= oldfree;
168 jeb->dirty_size -= oldfree;
171 jffs2_dbg_dump_block_lists_nolock(c);
172 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
173 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
176 static struct jffs2_raw_node_ref **jffs2_incore_replace_raw(struct jffs2_sb_info *c,
177 struct jffs2_inode_info *f,
178 struct jffs2_raw_node_ref *raw,
179 union jffs2_node_union *node)
181 struct jffs2_node_frag *frag;
182 struct jffs2_full_dirent *fd;
184 dbg_noderef("incore_replace_raw: node at %p is {%04x,%04x}\n",
185 node, je16_to_cpu(node->u.magic), je16_to_cpu(node->u.nodetype));
187 BUG_ON(je16_to_cpu(node->u.magic) != 0x1985 &&
188 je16_to_cpu(node->u.magic) != 0);
190 switch (je16_to_cpu(node->u.nodetype)) {
191 case JFFS2_NODETYPE_INODE:
192 if (f->metadata && f->metadata->raw == raw) {
193 dbg_noderef("Will replace ->raw in f->metadata at %p\n", f->metadata);
194 return &f->metadata->raw;
196 frag = jffs2_lookup_node_frag(&f->fragtree, je32_to_cpu(node->i.offset));
198 /* Find a frag which refers to the full_dnode we want to modify */
199 while (!frag->node || frag->node->raw != raw) {
200 frag = frag_next(frag);
203 dbg_noderef("Will replace ->raw in full_dnode at %p\n", frag->node);
204 return &frag->node->raw;
206 case JFFS2_NODETYPE_DIRENT:
207 for (fd = f->dents; fd; fd = fd->next) {
208 if (fd->raw == raw) {
209 dbg_noderef("Will replace ->raw in full_dirent at %p\n", fd);
216 dbg_noderef("Don't care about replacing raw for nodetype %x\n",
217 je16_to_cpu(node->u.nodetype));
223 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
224 static int jffs2_verify_write(struct jffs2_sb_info *c, unsigned char *buf,
231 ret = c->mtd->read(c->mtd, ofs, c->wbuf_pagesize, &retlen, c->wbuf_verify);
232 if (ret && ret != -EUCLEAN && ret != -EBADMSG) {
233 printk(KERN_WARNING "jffs2_verify_write(): Read back of page at %08x failed: %d\n", c->wbuf_ofs, ret);
235 } else if (retlen != c->wbuf_pagesize) {
236 printk(KERN_WARNING "jffs2_verify_write(): Read back of page at %08x gave short read: %zd not %d.\n", ofs, retlen, c->wbuf_pagesize);
239 if (!memcmp(buf, c->wbuf_verify, c->wbuf_pagesize))
243 eccstr = "corrected";
244 else if (ret == -EBADMSG)
245 eccstr = "correction failed";
247 eccstr = "OK or unused";
249 printk(KERN_WARNING "Write verify error (ECC %s) at %08x. Wrote:\n",
250 eccstr, c->wbuf_ofs);
251 print_hex_dump(KERN_WARNING, "", DUMP_PREFIX_OFFSET, 16, 1,
252 c->wbuf, c->wbuf_pagesize, 0);
254 printk(KERN_WARNING "Read back:\n");
255 print_hex_dump(KERN_WARNING, "", DUMP_PREFIX_OFFSET, 16, 1,
256 c->wbuf_verify, c->wbuf_pagesize, 0);
261 #define jffs2_verify_write(c,b,o) (0)
264 /* Recover from failure to write wbuf. Recover the nodes up to the
265 * wbuf, not the one which we were starting to try to write. */
267 static void jffs2_wbuf_recover(struct jffs2_sb_info *c)
269 struct jffs2_eraseblock *jeb, *new_jeb;
270 struct jffs2_raw_node_ref *raw, *next, *first_raw = NULL;
275 uint32_t start, end, ofs, len;
277 jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
279 spin_lock(&c->erase_completion_lock);
280 if (c->wbuf_ofs % c->mtd->erasesize)
281 jffs2_block_refile(c, jeb, REFILE_NOTEMPTY);
283 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
284 spin_unlock(&c->erase_completion_lock);
286 BUG_ON(!ref_obsolete(jeb->last_node));
288 /* Find the first node to be recovered, by skipping over every
289 node which ends before the wbuf starts, or which is obsolete. */
290 for (next = raw = jeb->first_node; next; raw = next) {
291 next = ref_next(raw);
293 if (ref_obsolete(raw) ||
294 (next && ref_offset(next) <= c->wbuf_ofs)) {
295 dbg_noderef("Skipping node at 0x%08x(%d)-0x%08x which is either before 0x%08x or obsolete\n",
296 ref_offset(raw), ref_flags(raw),
297 (ref_offset(raw) + ref_totlen(c, jeb, raw)),
301 dbg_noderef("First node to be recovered is at 0x%08x(%d)-0x%08x\n",
302 ref_offset(raw), ref_flags(raw),
303 (ref_offset(raw) + ref_totlen(c, jeb, raw)));
310 /* All nodes were obsolete. Nothing to recover. */
311 D1(printk(KERN_DEBUG "No non-obsolete nodes to be recovered. Just filing block bad\n"));
316 start = ref_offset(first_raw);
317 end = ref_offset(jeb->last_node);
320 /* Count the number of refs which need to be copied */
321 while ((raw = ref_next(raw)) != jeb->last_node)
324 dbg_noderef("wbuf recover %08x-%08x (%d bytes in %d nodes)\n",
325 start, end, end - start, nr_refile);
328 if (start < c->wbuf_ofs) {
329 /* First affected node was already partially written.
330 * Attempt to reread the old data into our buffer. */
332 buf = kmalloc(end - start, GFP_KERNEL);
334 printk(KERN_CRIT "Malloc failure in wbuf recovery. Data loss ensues.\n");
340 ret = c->mtd->read(c->mtd, start, c->wbuf_ofs - start, &retlen, buf);
342 /* ECC recovered ? */
343 if ((ret == -EUCLEAN || ret == -EBADMSG) &&
344 (retlen == c->wbuf_ofs - start))
347 if (ret || retlen != c->wbuf_ofs - start) {
348 printk(KERN_CRIT "Old data are already lost in wbuf recovery. Data loss ensues.\n");
353 first_raw = ref_next(first_raw);
355 while (first_raw && ref_obsolete(first_raw)) {
356 first_raw = ref_next(first_raw);
360 /* If this was the only node to be recovered, give up */
366 /* It wasn't. Go on and try to recover nodes complete in the wbuf */
367 start = ref_offset(first_raw);
368 dbg_noderef("wbuf now recover %08x-%08x (%d bytes in %d nodes)\n",
369 start, end, end - start, nr_refile);
372 /* Read succeeded. Copy the remaining data from the wbuf */
373 memcpy(buf + (c->wbuf_ofs - start), c->wbuf, end - c->wbuf_ofs);
376 /* OK... we're to rewrite (end-start) bytes of data from first_raw onwards.
377 Either 'buf' contains the data, or we find it in the wbuf */
379 /* ... and get an allocation of space from a shiny new block instead */
380 ret = jffs2_reserve_space_gc(c, end-start, &len, JFFS2_SUMMARY_NOSUM_SIZE);
382 printk(KERN_WARNING "Failed to allocate space for wbuf recovery. Data loss ensues.\n");
387 /* The summary is not recovered, so it must be disabled for this erase block */
388 jffs2_sum_disable_collecting(c->summary);
390 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, nr_refile);
392 printk(KERN_WARNING "Failed to allocate node refs for wbuf recovery. Data loss ensues.\n");
399 if (end-start >= c->wbuf_pagesize) {
400 /* Need to do another write immediately, but it's possible
401 that this is just because the wbuf itself is completely
402 full, and there's nothing earlier read back from the
403 flash. Hence 'buf' isn't necessarily what we're writing
405 unsigned char *rewrite_buf = buf?:c->wbuf;
406 uint32_t towrite = (end-start) - ((end-start)%c->wbuf_pagesize);
408 D1(printk(KERN_DEBUG "Write 0x%x bytes at 0x%08x in wbuf recover\n",
413 if (breakme++ == 20) {
414 printk(KERN_NOTICE "Faking write error at 0x%08x\n", ofs);
416 c->mtd->write(c->mtd, ofs, towrite, &retlen,
421 ret = c->mtd->write(c->mtd, ofs, towrite, &retlen,
424 if (ret || retlen != towrite || jffs2_verify_write(c, rewrite_buf, ofs)) {
425 /* Argh. We tried. Really we did. */
426 printk(KERN_CRIT "Recovery of wbuf failed due to a second write error\n");
430 jffs2_add_physical_node_ref(c, ofs | REF_OBSOLETE, ref_totlen(c, jeb, first_raw), NULL);
434 printk(KERN_NOTICE "Recovery of wbuf succeeded to %08x\n", ofs);
436 c->wbuf_len = (end - start) - towrite;
437 c->wbuf_ofs = ofs + towrite;
438 memmove(c->wbuf, rewrite_buf + towrite, c->wbuf_len);
439 /* Don't muck about with c->wbuf_inodes. False positives are harmless. */
441 /* OK, now we're left with the dregs in whichever buffer we're using */
443 memcpy(c->wbuf, buf, end-start);
445 memmove(c->wbuf, c->wbuf + (start - c->wbuf_ofs), end - start);
448 c->wbuf_len = end - start;
451 /* Now sort out the jffs2_raw_node_refs, moving them from the old to the next block */
452 new_jeb = &c->blocks[ofs / c->sector_size];
454 spin_lock(&c->erase_completion_lock);
455 for (raw = first_raw; raw != jeb->last_node; raw = ref_next(raw)) {
456 uint32_t rawlen = ref_totlen(c, jeb, raw);
457 struct jffs2_inode_cache *ic;
458 struct jffs2_raw_node_ref *new_ref;
459 struct jffs2_raw_node_ref **adjust_ref = NULL;
460 struct jffs2_inode_info *f = NULL;
462 D1(printk(KERN_DEBUG "Refiling block of %08x at %08x(%d) to %08x\n",
463 rawlen, ref_offset(raw), ref_flags(raw), ofs));
465 ic = jffs2_raw_ref_to_ic(raw);
467 /* Ick. This XATTR mess should be fixed shortly... */
468 if (ic && ic->class == RAWNODE_CLASS_XATTR_DATUM) {
469 struct jffs2_xattr_datum *xd = (void *)ic;
470 BUG_ON(xd->node != raw);
471 adjust_ref = &xd->node;
472 raw->next_in_ino = NULL;
474 } else if (ic && ic->class == RAWNODE_CLASS_XATTR_REF) {
475 struct jffs2_xattr_datum *xr = (void *)ic;
476 BUG_ON(xr->node != raw);
477 adjust_ref = &xr->node;
478 raw->next_in_ino = NULL;
480 } else if (ic && ic->class == RAWNODE_CLASS_INODE_CACHE) {
481 struct jffs2_raw_node_ref **p = &ic->nodes;
483 /* Remove the old node from the per-inode list */
484 while (*p && *p != (void *)ic) {
486 (*p) = (raw->next_in_ino);
487 raw->next_in_ino = NULL;
490 p = &((*p)->next_in_ino);
493 if (ic->state == INO_STATE_PRESENT && !ref_obsolete(raw)) {
494 /* If it's an in-core inode, then we have to adjust any
495 full_dirent or full_dnode structure to point to the
496 new version instead of the old */
497 f = jffs2_gc_fetch_inode(c, ic->ino, !ic->pino_nlink);
499 /* Should never happen; it _must_ be present */
500 JFFS2_ERROR("Failed to iget() ino #%u, err %ld\n",
501 ic->ino, PTR_ERR(f));
504 /* We don't lock f->sem. There's a number of ways we could
505 end up in here with it already being locked, and nobody's
506 going to modify it on us anyway because we hold the
507 alloc_sem. We're only changing one ->raw pointer too,
508 which we can get away with without upsetting readers. */
509 adjust_ref = jffs2_incore_replace_raw(c, f, raw,
510 (void *)(buf?:c->wbuf) + (ref_offset(raw) - start));
511 } else if (unlikely(ic->state != INO_STATE_PRESENT &&
512 ic->state != INO_STATE_CHECKEDABSENT &&
513 ic->state != INO_STATE_GC)) {
514 JFFS2_ERROR("Inode #%u is in strange state %d!\n", ic->ino, ic->state);
519 new_ref = jffs2_link_node_ref(c, new_jeb, ofs | ref_flags(raw), rawlen, ic);
522 BUG_ON(*adjust_ref != raw);
523 *adjust_ref = new_ref;
526 jffs2_gc_release_inode(c, f);
528 if (!ref_obsolete(raw)) {
529 jeb->dirty_size += rawlen;
530 jeb->used_size -= rawlen;
531 c->dirty_size += rawlen;
532 c->used_size -= rawlen;
533 raw->flash_offset = ref_offset(raw) | REF_OBSOLETE;
534 BUG_ON(raw->next_in_ino);
541 /* Fix up the original jeb now it's on the bad_list */
542 if (first_raw == jeb->first_node) {
543 D1(printk(KERN_DEBUG "Failing block at %08x is now empty. Moving to erase_pending_list\n", jeb->offset));
544 list_move(&jeb->list, &c->erase_pending_list);
545 c->nr_erasing_blocks++;
546 jffs2_erase_pending_trigger(c);
549 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
550 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
552 jffs2_dbg_acct_sanity_check_nolock(c, new_jeb);
553 jffs2_dbg_acct_paranoia_check_nolock(c, new_jeb);
555 spin_unlock(&c->erase_completion_lock);
557 D1(printk(KERN_DEBUG "wbuf recovery completed OK. wbuf_ofs 0x%08x, len 0x%x\n", c->wbuf_ofs, c->wbuf_len));
561 /* Meaning of pad argument:
562 0: Do not pad. Probably pointless - we only ever use this when we can't pad anyway.
563 1: Pad, do not adjust nextblock free_size
564 2: Pad, adjust nextblock free_size
567 #define PAD_NOACCOUNT 1
568 #define PAD_ACCOUNTING 2
570 static int __jffs2_flush_wbuf(struct jffs2_sb_info *c, int pad)
572 struct jffs2_eraseblock *wbuf_jeb;
576 /* Nothing to do if not write-buffering the flash. In particular, we shouldn't
577 del_timer() the timer we never initialised. */
578 if (!jffs2_is_writebuffered(c))
581 if (mutex_trylock(&c->alloc_sem)) {
582 mutex_unlock(&c->alloc_sem);
583 printk(KERN_CRIT "jffs2_flush_wbuf() called with alloc_sem not locked!\n");
587 if (!c->wbuf_len) /* already checked c->wbuf above */
590 wbuf_jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
591 if (jffs2_prealloc_raw_node_refs(c, wbuf_jeb, c->nextblock->allocated_refs + 1))
594 /* claim remaining space on the page
595 this happens, if we have a change to a new block,
596 or if fsync forces us to flush the writebuffer.
597 if we have a switch to next page, we will not have
598 enough remaining space for this.
601 c->wbuf_len = PAD(c->wbuf_len);
603 /* Pad with JFFS2_DIRTY_BITMASK initially. this helps out ECC'd NOR
604 with 8 byte page size */
605 memset(c->wbuf + c->wbuf_len, 0, c->wbuf_pagesize - c->wbuf_len);
607 if ( c->wbuf_len + sizeof(struct jffs2_unknown_node) < c->wbuf_pagesize) {
608 struct jffs2_unknown_node *padnode = (void *)(c->wbuf + c->wbuf_len);
609 padnode->magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
610 padnode->nodetype = cpu_to_je16(JFFS2_NODETYPE_PADDING);
611 padnode->totlen = cpu_to_je32(c->wbuf_pagesize - c->wbuf_len);
612 padnode->hdr_crc = cpu_to_je32(crc32(0, padnode, sizeof(*padnode)-4));
615 /* else jffs2_flash_writev has actually filled in the rest of the
616 buffer for us, and will deal with the node refs etc. later. */
620 if (breakme++ == 20) {
621 printk(KERN_NOTICE "Faking write error at 0x%08x\n", c->wbuf_ofs);
623 c->mtd->write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen,
629 ret = c->mtd->write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen, c->wbuf);
632 printk(KERN_WARNING "jffs2_flush_wbuf(): Write failed with %d\n", ret);
634 } else if (retlen != c->wbuf_pagesize) {
635 printk(KERN_WARNING "jffs2_flush_wbuf(): Write was short: %zd instead of %d\n",
636 retlen, c->wbuf_pagesize);
639 } else if ((ret = jffs2_verify_write(c, c->wbuf, c->wbuf_ofs))) {
641 jffs2_wbuf_recover(c);
646 /* Adjust free size of the block if we padded. */
648 uint32_t waste = c->wbuf_pagesize - c->wbuf_len;
650 D1(printk(KERN_DEBUG "jffs2_flush_wbuf() adjusting free_size of %sblock at %08x\n",
651 (wbuf_jeb==c->nextblock)?"next":"", wbuf_jeb->offset));
653 /* wbuf_pagesize - wbuf_len is the amount of space that's to be
654 padded. If there is less free space in the block than that,
655 something screwed up */
656 if (wbuf_jeb->free_size < waste) {
657 printk(KERN_CRIT "jffs2_flush_wbuf(): Accounting error. wbuf at 0x%08x has 0x%03x bytes, 0x%03x left.\n",
658 c->wbuf_ofs, c->wbuf_len, waste);
659 printk(KERN_CRIT "jffs2_flush_wbuf(): But free_size for block at 0x%08x is only 0x%08x\n",
660 wbuf_jeb->offset, wbuf_jeb->free_size);
664 spin_lock(&c->erase_completion_lock);
666 jffs2_link_node_ref(c, wbuf_jeb, (c->wbuf_ofs + c->wbuf_len) | REF_OBSOLETE, waste, NULL);
667 /* FIXME: that made it count as dirty. Convert to wasted */
668 wbuf_jeb->dirty_size -= waste;
669 c->dirty_size -= waste;
670 wbuf_jeb->wasted_size += waste;
671 c->wasted_size += waste;
673 spin_lock(&c->erase_completion_lock);
675 /* Stick any now-obsoleted blocks on the erase_pending_list */
676 jffs2_refile_wbuf_blocks(c);
677 jffs2_clear_wbuf_ino_list(c);
678 spin_unlock(&c->erase_completion_lock);
680 memset(c->wbuf,0xff,c->wbuf_pagesize);
681 /* adjust write buffer offset, else we get a non contiguous write bug */
682 c->wbuf_ofs += c->wbuf_pagesize;
687 /* Trigger garbage collection to flush the write-buffer.
688 If ino arg is zero, do it if _any_ real (i.e. not GC) writes are
689 outstanding. If ino arg non-zero, do it only if a write for the
690 given inode is outstanding. */
691 int jffs2_flush_wbuf_gc(struct jffs2_sb_info *c, uint32_t ino)
693 uint32_t old_wbuf_ofs;
694 uint32_t old_wbuf_len;
697 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() called for ino #%u...\n", ino));
702 mutex_lock(&c->alloc_sem);
703 if (!jffs2_wbuf_pending_for_ino(c, ino)) {
704 D1(printk(KERN_DEBUG "Ino #%d not pending in wbuf. Returning\n", ino));
705 mutex_unlock(&c->alloc_sem);
709 old_wbuf_ofs = c->wbuf_ofs;
710 old_wbuf_len = c->wbuf_len;
712 if (c->unchecked_size) {
713 /* GC won't make any progress for a while */
714 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() padding. Not finished checking\n"));
715 down_write(&c->wbuf_sem);
716 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
717 /* retry flushing wbuf in case jffs2_wbuf_recover
718 left some data in the wbuf */
720 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
721 up_write(&c->wbuf_sem);
722 } else while (old_wbuf_len &&
723 old_wbuf_ofs == c->wbuf_ofs) {
725 mutex_unlock(&c->alloc_sem);
727 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() calls gc pass\n"));
729 ret = jffs2_garbage_collect_pass(c);
731 /* GC failed. Flush it with padding instead */
732 mutex_lock(&c->alloc_sem);
733 down_write(&c->wbuf_sem);
734 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
735 /* retry flushing wbuf in case jffs2_wbuf_recover
736 left some data in the wbuf */
738 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
739 up_write(&c->wbuf_sem);
742 mutex_lock(&c->alloc_sem);
745 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() ends...\n"));
747 mutex_unlock(&c->alloc_sem);
751 /* Pad write-buffer to end and write it, wasting space. */
752 int jffs2_flush_wbuf_pad(struct jffs2_sb_info *c)
759 down_write(&c->wbuf_sem);
760 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
761 /* retry - maybe wbuf recover left some data in wbuf. */
763 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
764 up_write(&c->wbuf_sem);
769 static size_t jffs2_fill_wbuf(struct jffs2_sb_info *c, const uint8_t *buf,
772 if (len && !c->wbuf_len && (len >= c->wbuf_pagesize))
775 if (len > (c->wbuf_pagesize - c->wbuf_len))
776 len = c->wbuf_pagesize - c->wbuf_len;
777 memcpy(c->wbuf + c->wbuf_len, buf, len);
778 c->wbuf_len += (uint32_t) len;
782 int jffs2_flash_writev(struct jffs2_sb_info *c, const struct kvec *invecs,
783 unsigned long count, loff_t to, size_t *retlen,
786 struct jffs2_eraseblock *jeb;
787 size_t wbuf_retlen, donelen = 0;
788 uint32_t outvec_to = to;
791 /* If not writebuffered flash, don't bother */
792 if (!jffs2_is_writebuffered(c))
793 return jffs2_flash_direct_writev(c, invecs, count, to, retlen);
795 down_write(&c->wbuf_sem);
797 /* If wbuf_ofs is not initialized, set it to target address */
798 if (c->wbuf_ofs == 0xFFFFFFFF) {
799 c->wbuf_ofs = PAGE_DIV(to);
800 c->wbuf_len = PAGE_MOD(to);
801 memset(c->wbuf,0xff,c->wbuf_pagesize);
805 * Sanity checks on target address. It's permitted to write
806 * at PAD(c->wbuf_len+c->wbuf_ofs), and it's permitted to
807 * write at the beginning of a new erase block. Anything else,
808 * and you die. New block starts at xxx000c (0-b = block
811 if (SECTOR_ADDR(to) != SECTOR_ADDR(c->wbuf_ofs)) {
812 /* It's a write to a new block */
814 D1(printk(KERN_DEBUG "jffs2_flash_writev() to 0x%lx "
815 "causes flush of wbuf at 0x%08x\n",
816 (unsigned long)to, c->wbuf_ofs));
817 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
821 /* set pointer to new block */
822 c->wbuf_ofs = PAGE_DIV(to);
823 c->wbuf_len = PAGE_MOD(to);
826 if (to != PAD(c->wbuf_ofs + c->wbuf_len)) {
827 /* We're not writing immediately after the writebuffer. Bad. */
828 printk(KERN_CRIT "jffs2_flash_writev(): Non-contiguous write "
829 "to %08lx\n", (unsigned long)to);
831 printk(KERN_CRIT "wbuf was previously %08x-%08x\n",
832 c->wbuf_ofs, c->wbuf_ofs+c->wbuf_len);
836 /* adjust alignment offset */
837 if (c->wbuf_len != PAGE_MOD(to)) {
838 c->wbuf_len = PAGE_MOD(to);
839 /* take care of alignment to next page */
841 c->wbuf_len = c->wbuf_pagesize;
842 ret = __jffs2_flush_wbuf(c, NOPAD);
848 for (invec = 0; invec < count; invec++) {
849 int vlen = invecs[invec].iov_len;
850 uint8_t *v = invecs[invec].iov_base;
852 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
854 if (c->wbuf_len == c->wbuf_pagesize) {
855 ret = __jffs2_flush_wbuf(c, NOPAD);
860 outvec_to += wbuf_retlen;
861 donelen += wbuf_retlen;
864 if (vlen >= c->wbuf_pagesize) {
865 ret = c->mtd->write(c->mtd, outvec_to, PAGE_DIV(vlen),
867 if (ret < 0 || wbuf_retlen != PAGE_DIV(vlen))
871 outvec_to += wbuf_retlen;
872 c->wbuf_ofs = outvec_to;
873 donelen += wbuf_retlen;
877 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
878 if (c->wbuf_len == c->wbuf_pagesize) {
879 ret = __jffs2_flush_wbuf(c, NOPAD);
884 outvec_to += wbuf_retlen;
885 donelen += wbuf_retlen;
889 * If there's a remainder in the wbuf and it's a non-GC write,
890 * remember that the wbuf affects this ino
894 if (jffs2_sum_active()) {
895 int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
900 if (c->wbuf_len && ino)
901 jffs2_wbuf_dirties_inode(c, ino);
904 up_write(&c->wbuf_sem);
909 * At this point we have no problem, c->wbuf is empty. However
910 * refile nextblock to avoid writing again to same address.
913 spin_lock(&c->erase_completion_lock);
915 jeb = &c->blocks[outvec_to / c->sector_size];
916 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
918 spin_unlock(&c->erase_completion_lock);
922 up_write(&c->wbuf_sem);
927 * This is the entry for flash write.
928 * Check, if we work on NAND FLASH, if so build an kvec and write it via vritev
930 int jffs2_flash_write(struct jffs2_sb_info *c, loff_t ofs, size_t len,
931 size_t *retlen, const u_char *buf)
935 if (!jffs2_is_writebuffered(c))
936 return jffs2_flash_direct_write(c, ofs, len, retlen, buf);
938 vecs[0].iov_base = (unsigned char *) buf;
939 vecs[0].iov_len = len;
940 return jffs2_flash_writev(c, vecs, 1, ofs, retlen, 0);
944 Handle readback from writebuffer and ECC failure return
946 int jffs2_flash_read(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, u_char *buf)
948 loff_t orbf = 0, owbf = 0, lwbf = 0;
951 if (!jffs2_is_writebuffered(c))
952 return c->mtd->read(c->mtd, ofs, len, retlen, buf);
955 down_read(&c->wbuf_sem);
956 ret = c->mtd->read(c->mtd, ofs, len, retlen, buf);
958 if ( (ret == -EBADMSG || ret == -EUCLEAN) && (*retlen == len) ) {
960 printk(KERN_WARNING "mtd->read(0x%zx bytes from 0x%llx)"
961 " returned ECC error\n", len, ofs);
963 * We have the raw data without ECC correction in the buffer,
964 * maybe we are lucky and all data or parts are correct. We
965 * check the node. If data are corrupted node check will sort
966 * it out. We keep this block, it will fail on write or erase
967 * and the we mark it bad. Or should we do that now? But we
968 * should give him a chance. Maybe we had a system crash or
969 * power loss before the ecc write or a erase was completed.
970 * So we return success. :)
975 /* if no writebuffer available or write buffer empty, return */
976 if (!c->wbuf_pagesize || !c->wbuf_len)
979 /* if we read in a different block, return */
980 if (SECTOR_ADDR(ofs) != SECTOR_ADDR(c->wbuf_ofs))
983 if (ofs >= c->wbuf_ofs) {
984 owbf = (ofs - c->wbuf_ofs); /* offset in write buffer */
985 if (owbf > c->wbuf_len) /* is read beyond write buffer ? */
987 lwbf = c->wbuf_len - owbf; /* number of bytes to copy */
991 orbf = (c->wbuf_ofs - ofs); /* offset in read buffer */
992 if (orbf > len) /* is write beyond write buffer ? */
994 lwbf = len - orbf; /* number of bytes to copy */
995 if (lwbf > c->wbuf_len)
999 memcpy(buf+orbf,c->wbuf+owbf,lwbf);
1002 up_read(&c->wbuf_sem);
1006 #define NR_OOB_SCAN_PAGES 4
1008 /* For historical reasons we use only 8 bytes for OOB clean marker */
1009 #define OOB_CM_SIZE 8
1011 static const struct jffs2_unknown_node oob_cleanmarker =
1013 .magic = constant_cpu_to_je16(JFFS2_MAGIC_BITMASK),
1014 .nodetype = constant_cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
1015 .totlen = constant_cpu_to_je32(8)
1019 * Check, if the out of band area is empty. This function knows about the clean
1020 * marker and if it is present in OOB, treats the OOB as empty anyway.
1022 int jffs2_check_oob_empty(struct jffs2_sb_info *c,
1023 struct jffs2_eraseblock *jeb, int mode)
1026 int cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1027 struct mtd_oob_ops ops;
1029 ops.mode = MTD_OOB_AUTO;
1030 ops.ooblen = NR_OOB_SCAN_PAGES * c->oobavail;
1031 ops.oobbuf = c->oobbuf;
1032 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1035 ret = c->mtd->read_oob(c->mtd, jeb->offset, &ops);
1036 if (ret || ops.oobretlen != ops.ooblen) {
1037 printk(KERN_ERR "cannot read OOB for EB at %08x, requested %zd"
1038 " bytes, read %zd bytes, error %d\n",
1039 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1045 for(i = 0; i < ops.ooblen; i++) {
1046 if (mode && i < cmlen)
1047 /* Yeah, we know about the cleanmarker */
1050 if (ops.oobbuf[i] != 0xFF) {
1051 D2(printk(KERN_DEBUG "Found %02x at %x in OOB for "
1052 "%08x\n", ops.oobbuf[i], i, jeb->offset));
1061 * Check for a valid cleanmarker.
1062 * Returns: 0 if a valid cleanmarker was found
1063 * 1 if no cleanmarker was found
1064 * negative error code if an error occurred
1066 int jffs2_check_nand_cleanmarker(struct jffs2_sb_info *c,
1067 struct jffs2_eraseblock *jeb)
1069 struct mtd_oob_ops ops;
1070 int ret, cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1072 ops.mode = MTD_OOB_AUTO;
1074 ops.oobbuf = c->oobbuf;
1075 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1078 ret = c->mtd->read_oob(c->mtd, jeb->offset, &ops);
1079 if (ret || ops.oobretlen != ops.ooblen) {
1080 printk(KERN_ERR "cannot read OOB for EB at %08x, requested %zd"
1081 " bytes, read %zd bytes, error %d\n",
1082 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1088 return !!memcmp(&oob_cleanmarker, c->oobbuf, cmlen);
1091 int jffs2_write_nand_cleanmarker(struct jffs2_sb_info *c,
1092 struct jffs2_eraseblock *jeb)
1095 struct mtd_oob_ops ops;
1096 int cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1098 ops.mode = MTD_OOB_AUTO;
1100 ops.oobbuf = (uint8_t *)&oob_cleanmarker;
1101 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1104 ret = c->mtd->write_oob(c->mtd, jeb->offset, &ops);
1105 if (ret || ops.oobretlen != ops.ooblen) {
1106 printk(KERN_ERR "cannot write OOB for EB at %08x, requested %zd"
1107 " bytes, read %zd bytes, error %d\n",
1108 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1118 * On NAND we try to mark this block bad. If the block was erased more
1119 * than MAX_ERASE_FAILURES we mark it finaly bad.
1120 * Don't care about failures. This block remains on the erase-pending
1121 * or badblock list as long as nobody manipulates the flash with
1122 * a bootloader or something like that.
1125 int jffs2_write_nand_badblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
1129 /* if the count is < max, we try to write the counter to the 2nd page oob area */
1130 if( ++jeb->bad_count < MAX_ERASE_FAILURES)
1133 if (!c->mtd->block_markbad)
1134 return 1; // What else can we do?
1136 printk(KERN_WARNING "JFFS2: marking eraseblock at %08x\n as bad", bad_offset);
1137 ret = c->mtd->block_markbad(c->mtd, bad_offset);
1140 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1146 int jffs2_nand_flash_setup(struct jffs2_sb_info *c)
1148 struct nand_ecclayout *oinfo = c->mtd->ecclayout;
1150 if (!c->mtd->oobsize)
1153 /* Cleanmarker is out-of-band, so inline size zero */
1154 c->cleanmarker_size = 0;
1156 if (!oinfo || oinfo->oobavail == 0) {
1157 printk(KERN_ERR "inconsistent device description\n");
1161 D1(printk(KERN_DEBUG "JFFS2 using OOB on NAND\n"));
1163 c->oobavail = oinfo->oobavail;
1165 /* Initialise write buffer */
1166 init_rwsem(&c->wbuf_sem);
1167 c->wbuf_pagesize = c->mtd->writesize;
1168 c->wbuf_ofs = 0xFFFFFFFF;
1170 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1174 c->oobbuf = kmalloc(NR_OOB_SCAN_PAGES * c->oobavail, GFP_KERNEL);
1180 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1181 c->wbuf_verify = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1182 if (!c->wbuf_verify) {
1191 void jffs2_nand_flash_cleanup(struct jffs2_sb_info *c)
1193 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1194 kfree(c->wbuf_verify);
1200 int jffs2_dataflash_setup(struct jffs2_sb_info *c) {
1201 c->cleanmarker_size = 0; /* No cleanmarkers needed */
1203 /* Initialize write buffer */
1204 init_rwsem(&c->wbuf_sem);
1207 c->wbuf_pagesize = c->mtd->erasesize;
1209 /* Find a suitable c->sector_size
1210 * - Not too much sectors
1211 * - Sectors have to be at least 4 K + some bytes
1212 * - All known dataflashes have erase sizes of 528 or 1056
1213 * - we take at least 8 eraseblocks and want to have at least 8K size
1214 * - The concatenation should be a power of 2
1217 c->sector_size = 8 * c->mtd->erasesize;
1219 while (c->sector_size < 8192) {
1220 c->sector_size *= 2;
1223 /* It may be necessary to adjust the flash size */
1224 c->flash_size = c->mtd->size;
1226 if ((c->flash_size % c->sector_size) != 0) {
1227 c->flash_size = (c->flash_size / c->sector_size) * c->sector_size;
1228 printk(KERN_WARNING "JFFS2 flash size adjusted to %dKiB\n", c->flash_size);
1231 c->wbuf_ofs = 0xFFFFFFFF;
1232 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1236 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1237 c->wbuf_verify = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1238 if (!c->wbuf_verify) {
1245 printk(KERN_INFO "JFFS2 write-buffering enabled buffer (%d) erasesize (%d)\n", c->wbuf_pagesize, c->sector_size);
1250 void jffs2_dataflash_cleanup(struct jffs2_sb_info *c) {
1251 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1252 kfree(c->wbuf_verify);
1257 int jffs2_nor_wbuf_flash_setup(struct jffs2_sb_info *c) {
1258 /* Cleanmarker currently occupies whole programming regions,
1259 * either one or 2 for 8Byte STMicro flashes. */
1260 c->cleanmarker_size = max(16u, c->mtd->writesize);
1262 /* Initialize write buffer */
1263 init_rwsem(&c->wbuf_sem);
1264 c->wbuf_pagesize = c->mtd->writesize;
1265 c->wbuf_ofs = 0xFFFFFFFF;
1267 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1274 void jffs2_nor_wbuf_flash_cleanup(struct jffs2_sb_info *c) {
1278 int jffs2_ubivol_setup(struct jffs2_sb_info *c) {
1279 c->cleanmarker_size = 0;
1281 if (c->mtd->writesize == 1)
1282 /* We do not need write-buffer */
1285 init_rwsem(&c->wbuf_sem);
1287 c->wbuf_pagesize = c->mtd->writesize;
1288 c->wbuf_ofs = 0xFFFFFFFF;
1289 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1293 printk(KERN_INFO "JFFS2 write-buffering enabled buffer (%d) erasesize (%d)\n", c->wbuf_pagesize, c->sector_size);
1298 void jffs2_ubivol_cleanup(struct jffs2_sb_info *c) {