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 /* Adjust its size counts accordingly */
160 c->wasted_size += jeb->free_size;
161 c->free_size -= jeb->free_size;
162 jeb->wasted_size += jeb->free_size;
165 jffs2_dbg_dump_block_lists_nolock(c);
166 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
167 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
170 /* Recover from failure to write wbuf. Recover the nodes up to the
171 * wbuf, not the one which we were starting to try to write. */
173 static void jffs2_wbuf_recover(struct jffs2_sb_info *c)
175 struct jffs2_eraseblock *jeb, *new_jeb;
176 struct jffs2_raw_node_ref **first_raw, **raw;
180 uint32_t start, end, ofs, len;
182 spin_lock(&c->erase_completion_lock);
184 jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
186 jffs2_block_refile(c, jeb, REFILE_NOTEMPTY);
188 /* Find the first node to be recovered, by skipping over every
189 node which ends before the wbuf starts, or which is obsolete. */
190 first_raw = &jeb->first_node;
192 (ref_obsolete(*first_raw) ||
193 (ref_offset(*first_raw)+ref_totlen(c, jeb, *first_raw)) < c->wbuf_ofs)) {
194 D1(printk(KERN_DEBUG "Skipping node at 0x%08x(%d)-0x%08x which is either before 0x%08x or obsolete\n",
195 ref_offset(*first_raw), ref_flags(*first_raw),
196 (ref_offset(*first_raw) + ref_totlen(c, jeb, *first_raw)),
198 first_raw = &(*first_raw)->next_phys;
202 /* All nodes were obsolete. Nothing to recover. */
203 D1(printk(KERN_DEBUG "No non-obsolete nodes to be recovered. Just filing block bad\n"));
204 spin_unlock(&c->erase_completion_lock);
208 start = ref_offset(*first_raw);
209 end = ref_offset(*first_raw) + ref_totlen(c, jeb, *first_raw);
211 /* Find the last node to be recovered */
214 if (!ref_obsolete(*raw))
215 end = ref_offset(*raw) + ref_totlen(c, jeb, *raw);
217 raw = &(*raw)->next_phys;
219 spin_unlock(&c->erase_completion_lock);
221 D1(printk(KERN_DEBUG "wbuf recover %08x-%08x\n", start, end));
224 if (start < c->wbuf_ofs) {
225 /* First affected node was already partially written.
226 * Attempt to reread the old data into our buffer. */
228 buf = kmalloc(end - start, GFP_KERNEL);
230 printk(KERN_CRIT "Malloc failure in wbuf recovery. Data loss ensues.\n");
236 if (jffs2_cleanmarker_oob(c))
237 ret = c->mtd->read_ecc(c->mtd, start, c->wbuf_ofs - start, &retlen, buf, NULL, c->oobinfo);
239 ret = c->mtd->read(c->mtd, start, c->wbuf_ofs - start, &retlen, buf);
241 if (ret == -EBADMSG && retlen == c->wbuf_ofs - start) {
245 if (ret || retlen != c->wbuf_ofs - start) {
246 printk(KERN_CRIT "Old data are already lost in wbuf recovery. Data loss ensues.\n");
251 first_raw = &(*first_raw)->next_phys;
252 /* If this was the only node to be recovered, give up */
256 /* It wasn't. Go on and try to recover nodes complete in the wbuf */
257 start = ref_offset(*first_raw);
259 /* Read succeeded. Copy the remaining data from the wbuf */
260 memcpy(buf + (c->wbuf_ofs - start), c->wbuf, end - c->wbuf_ofs);
263 /* OK... we're to rewrite (end-start) bytes of data from first_raw onwards.
264 Either 'buf' contains the data, or we find it in the wbuf */
267 /* ... and get an allocation of space from a shiny new block instead */
268 ret = jffs2_reserve_space_gc(c, end-start, &ofs, &len, JFFS2_SUMMARY_NOSUM_SIZE);
270 printk(KERN_WARNING "Failed to allocate space for wbuf recovery. Data loss ensues.\n");
274 if (end-start >= c->wbuf_pagesize) {
275 /* Need to do another write immediately, but it's possible
276 that this is just because the wbuf itself is completely
277 full, and there's nothing earlier read back from the
278 flash. Hence 'buf' isn't necessarily what we're writing
280 unsigned char *rewrite_buf = buf?:c->wbuf;
281 uint32_t towrite = (end-start) - ((end-start)%c->wbuf_pagesize);
283 D1(printk(KERN_DEBUG "Write 0x%x bytes at 0x%08x in wbuf recover\n",
288 if (breakme++ == 20) {
289 printk(KERN_NOTICE "Faking write error at 0x%08x\n", ofs);
291 c->mtd->write_ecc(c->mtd, ofs, towrite, &retlen,
292 brokenbuf, NULL, c->oobinfo);
296 if (jffs2_cleanmarker_oob(c))
297 ret = c->mtd->write_ecc(c->mtd, ofs, towrite, &retlen,
298 rewrite_buf, NULL, c->oobinfo);
300 ret = c->mtd->write(c->mtd, ofs, towrite, &retlen, rewrite_buf);
302 if (ret || retlen != towrite) {
303 /* Argh. We tried. Really we did. */
304 printk(KERN_CRIT "Recovery of wbuf failed due to a second write error\n");
308 struct jffs2_raw_node_ref *raw2;
310 raw2 = jffs2_alloc_raw_node_ref();
314 raw2->flash_offset = ofs | REF_OBSOLETE;
315 raw2->next_in_ino = NULL;
317 jffs2_add_physical_node_ref(c, raw2, ref_totlen(c, jeb, *first_raw));
321 printk(KERN_NOTICE "Recovery of wbuf succeeded to %08x\n", ofs);
323 c->wbuf_len = (end - start) - towrite;
324 c->wbuf_ofs = ofs + towrite;
325 memmove(c->wbuf, rewrite_buf + towrite, c->wbuf_len);
326 /* Don't muck about with c->wbuf_inodes. False positives are harmless. */
329 /* OK, now we're left with the dregs in whichever buffer we're using */
331 memcpy(c->wbuf, buf, end-start);
334 memmove(c->wbuf, c->wbuf + (start - c->wbuf_ofs), end - start);
337 c->wbuf_len = end - start;
340 /* Now sort out the jffs2_raw_node_refs, moving them from the old to the next block */
341 new_jeb = &c->blocks[ofs / c->sector_size];
343 spin_lock(&c->erase_completion_lock);
344 if (new_jeb->first_node) {
345 /* Odd, but possible with ST flash later maybe */
346 new_jeb->last_node->next_phys = *first_raw;
348 new_jeb->first_node = *first_raw;
353 uint32_t rawlen = ref_totlen(c, jeb, *raw);
355 D1(printk(KERN_DEBUG "Refiling block of %08x at %08x(%d) to %08x\n",
356 rawlen, ref_offset(*raw), ref_flags(*raw), ofs));
358 if (ref_obsolete(*raw)) {
359 /* Shouldn't really happen much */
360 new_jeb->dirty_size += rawlen;
361 new_jeb->free_size -= rawlen;
362 c->dirty_size += rawlen;
364 new_jeb->used_size += rawlen;
365 new_jeb->free_size -= rawlen;
366 jeb->dirty_size += rawlen;
367 jeb->used_size -= rawlen;
368 c->dirty_size += rawlen;
370 c->free_size -= rawlen;
371 (*raw)->flash_offset = ofs | ref_flags(*raw);
373 new_jeb->last_node = *raw;
375 raw = &(*raw)->next_phys;
378 /* Fix up the original jeb now it's on the bad_list */
380 if (first_raw == &jeb->first_node) {
381 jeb->last_node = NULL;
382 D1(printk(KERN_DEBUG "Failing block at %08x is now empty. Moving to erase_pending_list\n", jeb->offset));
383 list_del(&jeb->list);
384 list_add(&jeb->list, &c->erase_pending_list);
385 c->nr_erasing_blocks++;
386 jffs2_erase_pending_trigger(c);
389 jeb->last_node = container_of(first_raw, struct jffs2_raw_node_ref, next_phys);
391 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
392 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
394 jffs2_dbg_acct_sanity_check_nolock(c, new_jeb);
395 jffs2_dbg_acct_paranoia_check_nolock(c, new_jeb);
397 spin_unlock(&c->erase_completion_lock);
399 D1(printk(KERN_DEBUG "wbuf recovery completed OK\n"));
402 /* Meaning of pad argument:
403 0: Do not pad. Probably pointless - we only ever use this when we can't pad anyway.
404 1: Pad, do not adjust nextblock free_size
405 2: Pad, adjust nextblock free_size
408 #define PAD_NOACCOUNT 1
409 #define PAD_ACCOUNTING 2
411 static int __jffs2_flush_wbuf(struct jffs2_sb_info *c, int pad)
416 /* Nothing to do if not write-buffering the flash. In particular, we shouldn't
417 del_timer() the timer we never initialised. */
418 if (!jffs2_is_writebuffered(c))
421 if (!down_trylock(&c->alloc_sem)) {
423 printk(KERN_CRIT "jffs2_flush_wbuf() called with alloc_sem not locked!\n");
427 if (!c->wbuf_len) /* already checked c->wbuf above */
430 /* claim remaining space on the page
431 this happens, if we have a change to a new block,
432 or if fsync forces us to flush the writebuffer.
433 if we have a switch to next page, we will not have
434 enough remaining space for this.
437 c->wbuf_len = PAD(c->wbuf_len);
439 /* Pad with JFFS2_DIRTY_BITMASK initially. this helps out ECC'd NOR
440 with 8 byte page size */
441 memset(c->wbuf + c->wbuf_len, 0, c->wbuf_pagesize - c->wbuf_len);
443 if ( c->wbuf_len + sizeof(struct jffs2_unknown_node) < c->wbuf_pagesize) {
444 struct jffs2_unknown_node *padnode = (void *)(c->wbuf + c->wbuf_len);
445 padnode->magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
446 padnode->nodetype = cpu_to_je16(JFFS2_NODETYPE_PADDING);
447 padnode->totlen = cpu_to_je32(c->wbuf_pagesize - c->wbuf_len);
448 padnode->hdr_crc = cpu_to_je32(crc32(0, padnode, sizeof(*padnode)-4));
451 /* else jffs2_flash_writev has actually filled in the rest of the
452 buffer for us, and will deal with the node refs etc. later. */
456 if (breakme++ == 20) {
457 printk(KERN_NOTICE "Faking write error at 0x%08x\n", c->wbuf_ofs);
459 c->mtd->write_ecc(c->mtd, c->wbuf_ofs, c->wbuf_pagesize,
460 &retlen, brokenbuf, NULL, c->oobinfo);
465 if (jffs2_cleanmarker_oob(c))
466 ret = c->mtd->write_ecc(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen, c->wbuf, NULL, c->oobinfo);
468 ret = c->mtd->write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen, c->wbuf);
470 if (ret || retlen != c->wbuf_pagesize) {
472 printk(KERN_WARNING "jffs2_flush_wbuf(): Write failed with %d\n",ret);
474 printk(KERN_WARNING "jffs2_flush_wbuf(): Write was short: %zd instead of %d\n",
475 retlen, c->wbuf_pagesize);
479 jffs2_wbuf_recover(c);
484 /* Adjust free size of the block if we padded. */
486 struct jffs2_eraseblock *jeb;
487 struct jffs2_raw_node_ref *ref;
488 uint32_t waste = c->wbuf_pagesize - c->wbuf_len;
490 jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
492 D1(printk(KERN_DEBUG "jffs2_flush_wbuf() adjusting free_size of %sblock at %08x\n",
493 (jeb==c->nextblock)?"next":"", jeb->offset));
495 /* wbuf_pagesize - wbuf_len is the amount of space that's to be
496 padded. If there is less free space in the block than that,
497 something screwed up */
498 if (jeb->free_size < waste) {
499 printk(KERN_CRIT "jffs2_flush_wbuf(): Accounting error. wbuf at 0x%08x has 0x%03x bytes, 0x%03x left.\n",
500 c->wbuf_ofs, c->wbuf_len, waste);
501 printk(KERN_CRIT "jffs2_flush_wbuf(): But free_size for block at 0x%08x is only 0x%08x\n",
502 jeb->offset, jeb->free_size);
505 ref = jffs2_alloc_raw_node_ref();
508 ref->flash_offset = c->wbuf_ofs + c->wbuf_len;
509 ref->flash_offset |= REF_OBSOLETE;
511 spin_lock(&c->erase_completion_lock);
513 jffs2_link_node_ref(c, jeb, ref, waste);
514 /* FIXME: that made it count as dirty. Convert to wasted */
515 jeb->dirty_size -= waste;
516 c->dirty_size -= waste;
517 jeb->wasted_size += waste;
518 c->wasted_size += waste;
520 spin_lock(&c->erase_completion_lock);
522 /* Stick any now-obsoleted blocks on the erase_pending_list */
523 jffs2_refile_wbuf_blocks(c);
524 jffs2_clear_wbuf_ino_list(c);
525 spin_unlock(&c->erase_completion_lock);
527 memset(c->wbuf,0xff,c->wbuf_pagesize);
528 /* adjust write buffer offset, else we get a non contiguous write bug */
529 c->wbuf_ofs += c->wbuf_pagesize;
534 /* Trigger garbage collection to flush the write-buffer.
535 If ino arg is zero, do it if _any_ real (i.e. not GC) writes are
536 outstanding. If ino arg non-zero, do it only if a write for the
537 given inode is outstanding. */
538 int jffs2_flush_wbuf_gc(struct jffs2_sb_info *c, uint32_t ino)
540 uint32_t old_wbuf_ofs;
541 uint32_t old_wbuf_len;
544 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() called for ino #%u...\n", ino));
550 if (!jffs2_wbuf_pending_for_ino(c, ino)) {
551 D1(printk(KERN_DEBUG "Ino #%d not pending in wbuf. Returning\n", ino));
556 old_wbuf_ofs = c->wbuf_ofs;
557 old_wbuf_len = c->wbuf_len;
559 if (c->unchecked_size) {
560 /* GC won't make any progress for a while */
561 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() padding. Not finished checking\n"));
562 down_write(&c->wbuf_sem);
563 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
564 /* retry flushing wbuf in case jffs2_wbuf_recover
565 left some data in the wbuf */
567 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
568 up_write(&c->wbuf_sem);
569 } else while (old_wbuf_len &&
570 old_wbuf_ofs == c->wbuf_ofs) {
574 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() calls gc pass\n"));
576 ret = jffs2_garbage_collect_pass(c);
578 /* GC failed. Flush it with padding instead */
580 down_write(&c->wbuf_sem);
581 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
582 /* retry flushing wbuf in case jffs2_wbuf_recover
583 left some data in the wbuf */
585 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
586 up_write(&c->wbuf_sem);
592 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() ends...\n"));
598 /* Pad write-buffer to end and write it, wasting space. */
599 int jffs2_flush_wbuf_pad(struct jffs2_sb_info *c)
606 down_write(&c->wbuf_sem);
607 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
608 /* retry - maybe wbuf recover left some data in wbuf. */
610 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
611 up_write(&c->wbuf_sem);
615 int jffs2_flash_writev(struct jffs2_sb_info *c, const struct kvec *invecs, unsigned long count, loff_t to, size_t *retlen, uint32_t ino)
617 struct kvec outvecs[3];
619 uint32_t split_ofs = 0;
621 int ret, splitvec = -1;
624 unsigned char *wbuf_ptr;
626 uint32_t outvec_to = to;
628 /* If not NAND flash, don't bother */
629 if (!jffs2_is_writebuffered(c))
630 return jffs2_flash_direct_writev(c, invecs, count, to, retlen);
632 down_write(&c->wbuf_sem);
634 /* If wbuf_ofs is not initialized, set it to target address */
635 if (c->wbuf_ofs == 0xFFFFFFFF) {
636 c->wbuf_ofs = PAGE_DIV(to);
637 c->wbuf_len = PAGE_MOD(to);
638 memset(c->wbuf,0xff,c->wbuf_pagesize);
641 /* Fixup the wbuf if we are moving to a new eraseblock. The checks below
642 fail for ECC'd NOR because cleanmarker == 16, so a block starts at
644 if (jffs2_nor_ecc(c)) {
645 if (((c->wbuf_ofs % c->sector_size) == 0) && !c->wbuf_len) {
646 c->wbuf_ofs = PAGE_DIV(to);
647 c->wbuf_len = PAGE_MOD(to);
648 memset(c->wbuf,0xff,c->wbuf_pagesize);
652 /* Sanity checks on target address.
653 It's permitted to write at PAD(c->wbuf_len+c->wbuf_ofs),
654 and it's permitted to write at the beginning of a new
655 erase block. Anything else, and you die.
656 New block starts at xxx000c (0-b = block header)
658 if (SECTOR_ADDR(to) != SECTOR_ADDR(c->wbuf_ofs)) {
659 /* It's a write to a new block */
661 D1(printk(KERN_DEBUG "jffs2_flash_writev() to 0x%lx causes flush of wbuf at 0x%08x\n", (unsigned long)to, c->wbuf_ofs));
662 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
664 /* the underlying layer has to check wbuf_len to do the cleanup */
665 D1(printk(KERN_WARNING "jffs2_flush_wbuf() called from jffs2_flash_writev() failed %d\n", ret));
670 /* set pointer to new block */
671 c->wbuf_ofs = PAGE_DIV(to);
672 c->wbuf_len = PAGE_MOD(to);
675 if (to != PAD(c->wbuf_ofs + c->wbuf_len)) {
676 /* We're not writing immediately after the writebuffer. Bad. */
677 printk(KERN_CRIT "jffs2_flash_writev(): Non-contiguous write to %08lx\n", (unsigned long)to);
679 printk(KERN_CRIT "wbuf was previously %08x-%08x\n",
680 c->wbuf_ofs, c->wbuf_ofs+c->wbuf_len);
684 /* Note outvecs[3] above. We know count is never greater than 2 */
686 printk(KERN_CRIT "jffs2_flash_writev(): count is %ld\n", count);
693 /* Fill writebuffer first, if already in use */
695 uint32_t invec_ofs = 0;
697 /* adjust alignment offset */
698 if (c->wbuf_len != PAGE_MOD(to)) {
699 c->wbuf_len = PAGE_MOD(to);
700 /* take care of alignment to next page */
702 c->wbuf_len = c->wbuf_pagesize;
705 while(c->wbuf_len < c->wbuf_pagesize) {
711 thislen = c->wbuf_pagesize - c->wbuf_len;
713 if (thislen >= invecs[invec].iov_len)
714 thislen = invecs[invec].iov_len;
718 memcpy(c->wbuf + c->wbuf_len, invecs[invec].iov_base, thislen);
719 c->wbuf_len += thislen;
721 /* Get next invec, if actual did not fill the buffer */
722 if (c->wbuf_len < c->wbuf_pagesize)
726 /* write buffer is full, flush buffer */
727 ret = __jffs2_flush_wbuf(c, NOPAD);
729 /* the underlying layer has to check wbuf_len to do the cleanup */
730 D1(printk(KERN_WARNING "jffs2_flush_wbuf() called from jffs2_flash_writev() failed %d\n", ret));
731 /* Retlen zero to make sure our caller doesn't mark the space dirty.
732 We've already done everything that's necessary */
736 outvec_to += donelen;
737 c->wbuf_ofs = outvec_to;
739 /* All invecs done ? */
743 /* Set up the first outvec, containing the remainder of the
744 invec we partially used */
745 if (invecs[invec].iov_len > invec_ofs) {
746 outvecs[0].iov_base = invecs[invec].iov_base+invec_ofs;
747 totlen = outvecs[0].iov_len = invecs[invec].iov_len-invec_ofs;
748 if (totlen > c->wbuf_pagesize) {
750 split_ofs = outvecs[0].iov_len - PAGE_MOD(totlen);
757 /* OK, now we've flushed the wbuf and the start of the bits
758 we have been asked to write, now to write the rest.... */
760 /* totlen holds the amount of data still to be written */
762 for ( ; invec < count; invec++,outvec++ ) {
763 outvecs[outvec].iov_base = invecs[invec].iov_base;
764 totlen += outvecs[outvec].iov_len = invecs[invec].iov_len;
765 if (PAGE_DIV(totlen) != PAGE_DIV(old_totlen)) {
767 split_ofs = outvecs[outvec].iov_len - PAGE_MOD(totlen);
772 /* Now the outvecs array holds all the remaining data to write */
773 /* Up to splitvec,split_ofs is to be written immediately. The rest
774 goes into the (now-empty) wbuf */
776 if (splitvec != -1) {
779 remainder = outvecs[splitvec].iov_len - split_ofs;
780 outvecs[splitvec].iov_len = split_ofs;
782 /* We did cross a page boundary, so we write some now */
783 if (jffs2_cleanmarker_oob(c))
784 ret = c->mtd->writev_ecc(c->mtd, outvecs, splitvec+1, outvec_to, &wbuf_retlen, NULL, c->oobinfo);
786 ret = jffs2_flash_direct_writev(c, outvecs, splitvec+1, outvec_to, &wbuf_retlen);
788 if (ret < 0 || wbuf_retlen != PAGE_DIV(totlen)) {
789 /* At this point we have no problem,
790 c->wbuf is empty. However refile nextblock to avoid
791 writing again to same address.
793 struct jffs2_eraseblock *jeb;
795 spin_lock(&c->erase_completion_lock);
797 jeb = &c->blocks[outvec_to / c->sector_size];
798 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
801 spin_unlock(&c->erase_completion_lock);
805 donelen += wbuf_retlen;
806 c->wbuf_ofs = PAGE_DIV(outvec_to) + PAGE_DIV(totlen);
809 outvecs[splitvec].iov_base += split_ofs;
810 outvecs[splitvec].iov_len = remainder;
819 /* Now splitvec points to the start of the bits we have to copy
823 for ( ; splitvec < outvec; splitvec++) {
824 /* Don't copy the wbuf into itself */
825 if (outvecs[splitvec].iov_base == c->wbuf)
827 memcpy(wbuf_ptr, outvecs[splitvec].iov_base, outvecs[splitvec].iov_len);
828 wbuf_ptr += outvecs[splitvec].iov_len;
829 donelen += outvecs[splitvec].iov_len;
831 c->wbuf_len = wbuf_ptr - c->wbuf;
833 /* If there's a remainder in the wbuf and it's a non-GC write,
834 remember that the wbuf affects this ino */
838 if (jffs2_sum_active()) {
839 int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
844 if (c->wbuf_len && ino)
845 jffs2_wbuf_dirties_inode(c, ino);
850 up_write(&c->wbuf_sem);
855 * This is the entry for flash write.
856 * Check, if we work on NAND FLASH, if so build an kvec and write it via vritev
858 int jffs2_flash_write(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, const u_char *buf)
862 if (!jffs2_is_writebuffered(c))
863 return jffs2_flash_direct_write(c, ofs, len, retlen, buf);
865 vecs[0].iov_base = (unsigned char *) buf;
866 vecs[0].iov_len = len;
867 return jffs2_flash_writev(c, vecs, 1, ofs, retlen, 0);
871 Handle readback from writebuffer and ECC failure return
873 int jffs2_flash_read(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, u_char *buf)
875 loff_t orbf = 0, owbf = 0, lwbf = 0;
878 if (!jffs2_is_writebuffered(c))
879 return c->mtd->read(c->mtd, ofs, len, retlen, buf);
882 down_read(&c->wbuf_sem);
883 if (jffs2_cleanmarker_oob(c))
884 ret = c->mtd->read_ecc(c->mtd, ofs, len, retlen, buf, NULL, c->oobinfo);
886 ret = c->mtd->read(c->mtd, ofs, len, retlen, buf);
888 if ( (ret == -EBADMSG) && (*retlen == len) ) {
889 printk(KERN_WARNING "mtd->read(0x%zx bytes from 0x%llx) returned ECC error\n",
892 * We have the raw data without ECC correction in the buffer, maybe
893 * we are lucky and all data or parts are correct. We check the node.
894 * If data are corrupted node check will sort it out.
895 * We keep this block, it will fail on write or erase and the we
896 * mark it bad. Or should we do that now? But we should give him a chance.
897 * Maybe we had a system crash or power loss before the ecc write or
898 * a erase was completed.
899 * So we return success. :)
904 /* if no writebuffer available or write buffer empty, return */
905 if (!c->wbuf_pagesize || !c->wbuf_len)
908 /* if we read in a different block, return */
909 if (SECTOR_ADDR(ofs) != SECTOR_ADDR(c->wbuf_ofs))
912 if (ofs >= c->wbuf_ofs) {
913 owbf = (ofs - c->wbuf_ofs); /* offset in write buffer */
914 if (owbf > c->wbuf_len) /* is read beyond write buffer ? */
916 lwbf = c->wbuf_len - owbf; /* number of bytes to copy */
920 orbf = (c->wbuf_ofs - ofs); /* offset in read buffer */
921 if (orbf > len) /* is write beyond write buffer ? */
923 lwbf = len - orbf; /* number of bytes to copy */
924 if (lwbf > c->wbuf_len)
928 memcpy(buf+orbf,c->wbuf+owbf,lwbf);
931 up_read(&c->wbuf_sem);
936 * Check, if the out of band area is empty
938 int jffs2_check_oob_empty( struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, int mode)
946 /* allocate a buffer for all oob data in this sector */
947 oob_size = c->mtd->oobsize;
949 buf = kmalloc(len, GFP_KERNEL);
951 printk(KERN_NOTICE "jffs2_check_oob_empty(): allocation of temporary data buffer for oob check failed\n");
955 * if mode = 0, we scan for a total empty oob area, else we have
956 * to take care of the cleanmarker in the first page of the block
958 ret = jffs2_flash_read_oob(c, jeb->offset, len , &retlen, buf);
960 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB failed %d for block at %08x\n", ret, jeb->offset));
965 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB return short read "
966 "(%zd bytes not %d) for block at %08x\n", retlen, len, jeb->offset));
971 /* Special check for first page */
972 for(i = 0; i < oob_size ; i++) {
973 /* Yeah, we know about the cleanmarker. */
974 if (mode && i >= c->fsdata_pos &&
975 i < c->fsdata_pos + c->fsdata_len)
978 if (buf[i] != 0xFF) {
979 D2(printk(KERN_DEBUG "Found %02x at %x in OOB for %08x\n",
980 buf[i], i, jeb->offset));
986 /* we know, we are aligned :) */
987 for (page = oob_size; page < len; page += sizeof(long)) {
988 unsigned long dat = *(unsigned long *)(&buf[page]);
1002 * Scan for a valid cleanmarker and for bad blocks
1003 * For virtual blocks (concatenated physical blocks) check the cleanmarker
1004 * only in the first page of the first physical block, but scan for bad blocks in all
1007 int jffs2_check_nand_cleanmarker (struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1009 struct jffs2_unknown_node n;
1010 unsigned char buf[2 * NAND_MAX_OOBSIZE];
1012 int ret, i, cnt, retval = 0;
1013 size_t retlen, offset;
1016 offset = jeb->offset;
1017 oob_size = c->mtd->oobsize;
1019 /* Loop through the physical blocks */
1020 for (cnt = 0; cnt < (c->sector_size / c->mtd->erasesize); cnt++) {
1021 /* Check first if the block is bad. */
1022 if (c->mtd->block_isbad (c->mtd, offset)) {
1023 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): Bad block at %08x\n", jeb->offset));
1027 * We read oob data from page 0 and 1 of the block.
1028 * page 0 contains cleanmarker and badblock info
1029 * page 1 contains failure count of this block
1031 ret = c->mtd->read_oob (c->mtd, offset, oob_size << 1, &retlen, buf);
1034 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): Read OOB failed %d for block at %08x\n", ret, jeb->offset));
1037 if (retlen < (oob_size << 1)) {
1038 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): Read OOB return short read (%zd bytes not %d) for block at %08x\n", retlen, oob_size << 1, jeb->offset));
1042 /* Check cleanmarker only on the first physical block */
1044 n.magic = cpu_to_je16 (JFFS2_MAGIC_BITMASK);
1045 n.nodetype = cpu_to_je16 (JFFS2_NODETYPE_CLEANMARKER);
1046 n.totlen = cpu_to_je32 (8);
1047 p = (unsigned char *) &n;
1049 for (i = 0; i < c->fsdata_len; i++) {
1050 if (buf[c->fsdata_pos + i] != p[i]) {
1054 D1(if (retval == 1) {
1055 printk(KERN_WARNING "jffs2_check_nand_cleanmarker(): Cleanmarker node not detected in block at %08x\n", jeb->offset);
1056 printk(KERN_WARNING "OOB at %08x was ", offset);
1057 for (i=0; i < oob_size; i++) {
1058 printk("%02x ", buf[i]);
1063 offset += c->mtd->erasesize;
1068 int jffs2_write_nand_cleanmarker(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1070 struct jffs2_unknown_node n;
1074 n.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1075 n.nodetype = cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER);
1076 n.totlen = cpu_to_je32(8);
1078 ret = jffs2_flash_write_oob(c, jeb->offset + c->fsdata_pos, c->fsdata_len, &retlen, (unsigned char *)&n);
1081 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1084 if (retlen != c->fsdata_len) {
1085 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): Short write for block at %08x: %zd not %d\n", jeb->offset, retlen, c->fsdata_len));
1092 * On NAND we try to mark this block bad. If the block was erased more
1093 * than MAX_ERASE_FAILURES we mark it finaly bad.
1094 * Don't care about failures. This block remains on the erase-pending
1095 * or badblock list as long as nobody manipulates the flash with
1096 * a bootloader or something like that.
1099 int jffs2_write_nand_badblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
1103 /* if the count is < max, we try to write the counter to the 2nd page oob area */
1104 if( ++jeb->bad_count < MAX_ERASE_FAILURES)
1107 if (!c->mtd->block_markbad)
1108 return 1; // What else can we do?
1110 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Marking bad block at %08x\n", bad_offset));
1111 ret = c->mtd->block_markbad(c->mtd, bad_offset);
1114 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1120 #define NAND_JFFS2_OOB16_FSDALEN 8
1122 static struct nand_oobinfo jffs2_oobinfo_docecc = {
1123 .useecc = MTD_NANDECC_PLACE,
1125 .eccpos = {0,1,2,3,4,5}
1129 static int jffs2_nand_set_oobinfo(struct jffs2_sb_info *c)
1131 struct nand_oobinfo *oinfo = &c->mtd->oobinfo;
1133 /* Do this only, if we have an oob buffer */
1134 if (!c->mtd->oobsize)
1137 /* Cleanmarker is out-of-band, so inline size zero */
1138 c->cleanmarker_size = 0;
1140 /* Should we use autoplacement ? */
1141 if (oinfo && oinfo->useecc == MTD_NANDECC_AUTOPLACE) {
1142 D1(printk(KERN_DEBUG "JFFS2 using autoplace on NAND\n"));
1143 /* Get the position of the free bytes */
1144 if (!oinfo->oobfree[0][1]) {
1145 printk (KERN_WARNING "jffs2_nand_set_oobinfo(): Eeep. Autoplacement selected and no empty space in oob\n");
1148 c->fsdata_pos = oinfo->oobfree[0][0];
1149 c->fsdata_len = oinfo->oobfree[0][1];
1150 if (c->fsdata_len > 8)
1153 /* This is just a legacy fallback and should go away soon */
1154 switch(c->mtd->ecctype) {
1155 case MTD_ECC_RS_DiskOnChip:
1156 printk(KERN_WARNING "JFFS2 using DiskOnChip hardware ECC without autoplacement. Fix it!\n");
1157 c->oobinfo = &jffs2_oobinfo_docecc;
1159 c->fsdata_len = NAND_JFFS2_OOB16_FSDALEN;
1160 c->badblock_pos = 15;
1164 D1(printk(KERN_DEBUG "JFFS2 on NAND. No autoplacment info found\n"));
1171 int jffs2_nand_flash_setup(struct jffs2_sb_info *c)
1175 /* Initialise write buffer */
1176 init_rwsem(&c->wbuf_sem);
1177 c->wbuf_pagesize = c->mtd->oobblock;
1178 c->wbuf_ofs = 0xFFFFFFFF;
1180 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1184 res = jffs2_nand_set_oobinfo(c);
1188 brokenbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1193 memset(brokenbuf, 0xdb, c->wbuf_pagesize);
1198 void jffs2_nand_flash_cleanup(struct jffs2_sb_info *c)
1203 int jffs2_dataflash_setup(struct jffs2_sb_info *c) {
1204 c->cleanmarker_size = 0; /* No cleanmarkers needed */
1206 /* Initialize write buffer */
1207 init_rwsem(&c->wbuf_sem);
1210 c->wbuf_pagesize = c->mtd->erasesize;
1212 /* Find a suitable c->sector_size
1213 * - Not too much sectors
1214 * - Sectors have to be at least 4 K + some bytes
1215 * - All known dataflashes have erase sizes of 528 or 1056
1216 * - we take at least 8 eraseblocks and want to have at least 8K size
1217 * - The concatenation should be a power of 2
1220 c->sector_size = 8 * c->mtd->erasesize;
1222 while (c->sector_size < 8192) {
1223 c->sector_size *= 2;
1226 /* It may be necessary to adjust the flash size */
1227 c->flash_size = c->mtd->size;
1229 if ((c->flash_size % c->sector_size) != 0) {
1230 c->flash_size = (c->flash_size / c->sector_size) * c->sector_size;
1231 printk(KERN_WARNING "JFFS2 flash size adjusted to %dKiB\n", c->flash_size);
1234 c->wbuf_ofs = 0xFFFFFFFF;
1235 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1239 printk(KERN_INFO "JFFS2 write-buffering enabled buffer (%d) erasesize (%d)\n", c->wbuf_pagesize, c->sector_size);
1244 void jffs2_dataflash_cleanup(struct jffs2_sb_info *c) {
1248 int jffs2_nor_ecc_flash_setup(struct jffs2_sb_info *c) {
1249 /* Cleanmarker is actually larger on the flashes */
1250 c->cleanmarker_size = 16;
1252 /* Initialize write buffer */
1253 init_rwsem(&c->wbuf_sem);
1254 c->wbuf_pagesize = c->mtd->eccsize;
1255 c->wbuf_ofs = 0xFFFFFFFF;
1257 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1264 void jffs2_nor_ecc_flash_cleanup(struct jffs2_sb_info *c) {
1268 int jffs2_nor_wbuf_flash_setup(struct jffs2_sb_info *c) {
1269 /* Cleanmarker currently occupies a whole programming region */
1270 c->cleanmarker_size = MTD_PROGREGION_SIZE(c->mtd);
1272 /* Initialize write buffer */
1273 init_rwsem(&c->wbuf_sem);
1274 c->wbuf_pagesize = MTD_PROGREGION_SIZE(c->mtd);
1275 c->wbuf_ofs = 0xFFFFFFFF;
1277 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1284 void jffs2_nor_wbuf_flash_cleanup(struct jffs2_sb_info *c) {