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;
510 ref->next_in_ino = NULL;
512 spin_lock(&c->erase_completion_lock);
514 jffs2_link_node_ref(c, jeb, ref, waste);
515 /* FIXME: that made it count as dirty. Convert to wasted */
516 jeb->dirty_size -= waste;
517 c->dirty_size -= waste;
518 jeb->wasted_size += waste;
519 c->wasted_size += waste;
521 spin_lock(&c->erase_completion_lock);
523 /* Stick any now-obsoleted blocks on the erase_pending_list */
524 jffs2_refile_wbuf_blocks(c);
525 jffs2_clear_wbuf_ino_list(c);
526 spin_unlock(&c->erase_completion_lock);
528 memset(c->wbuf,0xff,c->wbuf_pagesize);
529 /* adjust write buffer offset, else we get a non contiguous write bug */
530 c->wbuf_ofs += c->wbuf_pagesize;
535 /* Trigger garbage collection to flush the write-buffer.
536 If ino arg is zero, do it if _any_ real (i.e. not GC) writes are
537 outstanding. If ino arg non-zero, do it only if a write for the
538 given inode is outstanding. */
539 int jffs2_flush_wbuf_gc(struct jffs2_sb_info *c, uint32_t ino)
541 uint32_t old_wbuf_ofs;
542 uint32_t old_wbuf_len;
545 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() called for ino #%u...\n", ino));
551 if (!jffs2_wbuf_pending_for_ino(c, ino)) {
552 D1(printk(KERN_DEBUG "Ino #%d not pending in wbuf. Returning\n", ino));
557 old_wbuf_ofs = c->wbuf_ofs;
558 old_wbuf_len = c->wbuf_len;
560 if (c->unchecked_size) {
561 /* GC won't make any progress for a while */
562 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() padding. Not finished checking\n"));
563 down_write(&c->wbuf_sem);
564 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
565 /* retry flushing wbuf in case jffs2_wbuf_recover
566 left some data in the wbuf */
568 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
569 up_write(&c->wbuf_sem);
570 } else while (old_wbuf_len &&
571 old_wbuf_ofs == c->wbuf_ofs) {
575 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() calls gc pass\n"));
577 ret = jffs2_garbage_collect_pass(c);
579 /* GC failed. Flush it with padding instead */
581 down_write(&c->wbuf_sem);
582 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
583 /* retry flushing wbuf in case jffs2_wbuf_recover
584 left some data in the wbuf */
586 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
587 up_write(&c->wbuf_sem);
593 D1(printk(KERN_DEBUG "jffs2_flush_wbuf_gc() ends...\n"));
599 /* Pad write-buffer to end and write it, wasting space. */
600 int jffs2_flush_wbuf_pad(struct jffs2_sb_info *c)
607 down_write(&c->wbuf_sem);
608 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
609 /* retry - maybe wbuf recover left some data in wbuf. */
611 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
612 up_write(&c->wbuf_sem);
616 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)
618 struct kvec outvecs[3];
620 uint32_t split_ofs = 0;
622 int ret, splitvec = -1;
625 unsigned char *wbuf_ptr;
627 uint32_t outvec_to = to;
629 /* If not NAND flash, don't bother */
630 if (!jffs2_is_writebuffered(c))
631 return jffs2_flash_direct_writev(c, invecs, count, to, retlen);
633 down_write(&c->wbuf_sem);
635 /* If wbuf_ofs is not initialized, set it to target address */
636 if (c->wbuf_ofs == 0xFFFFFFFF) {
637 c->wbuf_ofs = PAGE_DIV(to);
638 c->wbuf_len = PAGE_MOD(to);
639 memset(c->wbuf,0xff,c->wbuf_pagesize);
642 /* Fixup the wbuf if we are moving to a new eraseblock. The checks below
643 fail for ECC'd NOR because cleanmarker == 16, so a block starts at
645 if (jffs2_nor_ecc(c)) {
646 if (((c->wbuf_ofs % c->sector_size) == 0) && !c->wbuf_len) {
647 c->wbuf_ofs = PAGE_DIV(to);
648 c->wbuf_len = PAGE_MOD(to);
649 memset(c->wbuf,0xff,c->wbuf_pagesize);
653 /* Sanity checks on target address.
654 It's permitted to write at PAD(c->wbuf_len+c->wbuf_ofs),
655 and it's permitted to write at the beginning of a new
656 erase block. Anything else, and you die.
657 New block starts at xxx000c (0-b = block header)
659 if (SECTOR_ADDR(to) != SECTOR_ADDR(c->wbuf_ofs)) {
660 /* It's a write to a new block */
662 D1(printk(KERN_DEBUG "jffs2_flash_writev() to 0x%lx causes flush of wbuf at 0x%08x\n", (unsigned long)to, c->wbuf_ofs));
663 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
665 /* the underlying layer has to check wbuf_len to do the cleanup */
666 D1(printk(KERN_WARNING "jffs2_flush_wbuf() called from jffs2_flash_writev() failed %d\n", ret));
671 /* set pointer to new block */
672 c->wbuf_ofs = PAGE_DIV(to);
673 c->wbuf_len = PAGE_MOD(to);
676 if (to != PAD(c->wbuf_ofs + c->wbuf_len)) {
677 /* We're not writing immediately after the writebuffer. Bad. */
678 printk(KERN_CRIT "jffs2_flash_writev(): Non-contiguous write to %08lx\n", (unsigned long)to);
680 printk(KERN_CRIT "wbuf was previously %08x-%08x\n",
681 c->wbuf_ofs, c->wbuf_ofs+c->wbuf_len);
685 /* Note outvecs[3] above. We know count is never greater than 2 */
687 printk(KERN_CRIT "jffs2_flash_writev(): count is %ld\n", count);
694 /* Fill writebuffer first, if already in use */
696 uint32_t invec_ofs = 0;
698 /* adjust alignment offset */
699 if (c->wbuf_len != PAGE_MOD(to)) {
700 c->wbuf_len = PAGE_MOD(to);
701 /* take care of alignment to next page */
703 c->wbuf_len = c->wbuf_pagesize;
706 while(c->wbuf_len < c->wbuf_pagesize) {
712 thislen = c->wbuf_pagesize - c->wbuf_len;
714 if (thislen >= invecs[invec].iov_len)
715 thislen = invecs[invec].iov_len;
719 memcpy(c->wbuf + c->wbuf_len, invecs[invec].iov_base, thislen);
720 c->wbuf_len += thislen;
722 /* Get next invec, if actual did not fill the buffer */
723 if (c->wbuf_len < c->wbuf_pagesize)
727 /* write buffer is full, flush buffer */
728 ret = __jffs2_flush_wbuf(c, NOPAD);
730 /* the underlying layer has to check wbuf_len to do the cleanup */
731 D1(printk(KERN_WARNING "jffs2_flush_wbuf() called from jffs2_flash_writev() failed %d\n", ret));
732 /* Retlen zero to make sure our caller doesn't mark the space dirty.
733 We've already done everything that's necessary */
737 outvec_to += donelen;
738 c->wbuf_ofs = outvec_to;
740 /* All invecs done ? */
744 /* Set up the first outvec, containing the remainder of the
745 invec we partially used */
746 if (invecs[invec].iov_len > invec_ofs) {
747 outvecs[0].iov_base = invecs[invec].iov_base+invec_ofs;
748 totlen = outvecs[0].iov_len = invecs[invec].iov_len-invec_ofs;
749 if (totlen > c->wbuf_pagesize) {
751 split_ofs = outvecs[0].iov_len - PAGE_MOD(totlen);
758 /* OK, now we've flushed the wbuf and the start of the bits
759 we have been asked to write, now to write the rest.... */
761 /* totlen holds the amount of data still to be written */
763 for ( ; invec < count; invec++,outvec++ ) {
764 outvecs[outvec].iov_base = invecs[invec].iov_base;
765 totlen += outvecs[outvec].iov_len = invecs[invec].iov_len;
766 if (PAGE_DIV(totlen) != PAGE_DIV(old_totlen)) {
768 split_ofs = outvecs[outvec].iov_len - PAGE_MOD(totlen);
773 /* Now the outvecs array holds all the remaining data to write */
774 /* Up to splitvec,split_ofs is to be written immediately. The rest
775 goes into the (now-empty) wbuf */
777 if (splitvec != -1) {
780 remainder = outvecs[splitvec].iov_len - split_ofs;
781 outvecs[splitvec].iov_len = split_ofs;
783 /* We did cross a page boundary, so we write some now */
784 if (jffs2_cleanmarker_oob(c))
785 ret = c->mtd->writev_ecc(c->mtd, outvecs, splitvec+1, outvec_to, &wbuf_retlen, NULL, c->oobinfo);
787 ret = jffs2_flash_direct_writev(c, outvecs, splitvec+1, outvec_to, &wbuf_retlen);
789 if (ret < 0 || wbuf_retlen != PAGE_DIV(totlen)) {
790 /* At this point we have no problem,
791 c->wbuf is empty. However refile nextblock to avoid
792 writing again to same address.
794 struct jffs2_eraseblock *jeb;
796 spin_lock(&c->erase_completion_lock);
798 jeb = &c->blocks[outvec_to / c->sector_size];
799 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
802 spin_unlock(&c->erase_completion_lock);
806 donelen += wbuf_retlen;
807 c->wbuf_ofs = PAGE_DIV(outvec_to) + PAGE_DIV(totlen);
810 outvecs[splitvec].iov_base += split_ofs;
811 outvecs[splitvec].iov_len = remainder;
820 /* Now splitvec points to the start of the bits we have to copy
824 for ( ; splitvec < outvec; splitvec++) {
825 /* Don't copy the wbuf into itself */
826 if (outvecs[splitvec].iov_base == c->wbuf)
828 memcpy(wbuf_ptr, outvecs[splitvec].iov_base, outvecs[splitvec].iov_len);
829 wbuf_ptr += outvecs[splitvec].iov_len;
830 donelen += outvecs[splitvec].iov_len;
832 c->wbuf_len = wbuf_ptr - c->wbuf;
834 /* If there's a remainder in the wbuf and it's a non-GC write,
835 remember that the wbuf affects this ino */
839 if (jffs2_sum_active()) {
840 int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
845 if (c->wbuf_len && ino)
846 jffs2_wbuf_dirties_inode(c, ino);
851 up_write(&c->wbuf_sem);
856 * This is the entry for flash write.
857 * Check, if we work on NAND FLASH, if so build an kvec and write it via vritev
859 int jffs2_flash_write(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, const u_char *buf)
863 if (!jffs2_is_writebuffered(c))
864 return jffs2_flash_direct_write(c, ofs, len, retlen, buf);
866 vecs[0].iov_base = (unsigned char *) buf;
867 vecs[0].iov_len = len;
868 return jffs2_flash_writev(c, vecs, 1, ofs, retlen, 0);
872 Handle readback from writebuffer and ECC failure return
874 int jffs2_flash_read(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, u_char *buf)
876 loff_t orbf = 0, owbf = 0, lwbf = 0;
879 if (!jffs2_is_writebuffered(c))
880 return c->mtd->read(c->mtd, ofs, len, retlen, buf);
883 down_read(&c->wbuf_sem);
884 if (jffs2_cleanmarker_oob(c))
885 ret = c->mtd->read_ecc(c->mtd, ofs, len, retlen, buf, NULL, c->oobinfo);
887 ret = c->mtd->read(c->mtd, ofs, len, retlen, buf);
889 if ( (ret == -EBADMSG) && (*retlen == len) ) {
890 printk(KERN_WARNING "mtd->read(0x%zx bytes from 0x%llx) returned ECC error\n",
893 * We have the raw data without ECC correction in the buffer, maybe
894 * we are lucky and all data or parts are correct. We check the node.
895 * If data are corrupted node check will sort it out.
896 * We keep this block, it will fail on write or erase and the we
897 * mark it bad. Or should we do that now? But we should give him a chance.
898 * Maybe we had a system crash or power loss before the ecc write or
899 * a erase was completed.
900 * So we return success. :)
905 /* if no writebuffer available or write buffer empty, return */
906 if (!c->wbuf_pagesize || !c->wbuf_len)
909 /* if we read in a different block, return */
910 if (SECTOR_ADDR(ofs) != SECTOR_ADDR(c->wbuf_ofs))
913 if (ofs >= c->wbuf_ofs) {
914 owbf = (ofs - c->wbuf_ofs); /* offset in write buffer */
915 if (owbf > c->wbuf_len) /* is read beyond write buffer ? */
917 lwbf = c->wbuf_len - owbf; /* number of bytes to copy */
921 orbf = (c->wbuf_ofs - ofs); /* offset in read buffer */
922 if (orbf > len) /* is write beyond write buffer ? */
924 lwbf = len - orbf; /* number of bytes to copy */
925 if (lwbf > c->wbuf_len)
929 memcpy(buf+orbf,c->wbuf+owbf,lwbf);
932 up_read(&c->wbuf_sem);
937 * Check, if the out of band area is empty
939 int jffs2_check_oob_empty( struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, int mode)
947 /* allocate a buffer for all oob data in this sector */
948 oob_size = c->mtd->oobsize;
950 buf = kmalloc(len, GFP_KERNEL);
952 printk(KERN_NOTICE "jffs2_check_oob_empty(): allocation of temporary data buffer for oob check failed\n");
956 * if mode = 0, we scan for a total empty oob area, else we have
957 * to take care of the cleanmarker in the first page of the block
959 ret = jffs2_flash_read_oob(c, jeb->offset, len , &retlen, buf);
961 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB failed %d for block at %08x\n", ret, jeb->offset));
966 D1(printk(KERN_WARNING "jffs2_check_oob_empty(): Read OOB return short read "
967 "(%zd bytes not %d) for block at %08x\n", retlen, len, jeb->offset));
972 /* Special check for first page */
973 for(i = 0; i < oob_size ; i++) {
974 /* Yeah, we know about the cleanmarker. */
975 if (mode && i >= c->fsdata_pos &&
976 i < c->fsdata_pos + c->fsdata_len)
979 if (buf[i] != 0xFF) {
980 D2(printk(KERN_DEBUG "Found %02x at %x in OOB for %08x\n",
981 buf[i], i, jeb->offset));
987 /* we know, we are aligned :) */
988 for (page = oob_size; page < len; page += sizeof(long)) {
989 unsigned long dat = *(unsigned long *)(&buf[page]);
1003 * Scan for a valid cleanmarker and for bad blocks
1004 * For virtual blocks (concatenated physical blocks) check the cleanmarker
1005 * only in the first page of the first physical block, but scan for bad blocks in all
1008 int jffs2_check_nand_cleanmarker (struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1010 struct jffs2_unknown_node n;
1011 unsigned char buf[2 * NAND_MAX_OOBSIZE];
1013 int ret, i, cnt, retval = 0;
1014 size_t retlen, offset;
1017 offset = jeb->offset;
1018 oob_size = c->mtd->oobsize;
1020 /* Loop through the physical blocks */
1021 for (cnt = 0; cnt < (c->sector_size / c->mtd->erasesize); cnt++) {
1022 /* Check first if the block is bad. */
1023 if (c->mtd->block_isbad (c->mtd, offset)) {
1024 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): Bad block at %08x\n", jeb->offset));
1028 * We read oob data from page 0 and 1 of the block.
1029 * page 0 contains cleanmarker and badblock info
1030 * page 1 contains failure count of this block
1032 ret = c->mtd->read_oob (c->mtd, offset, oob_size << 1, &retlen, buf);
1035 D1 (printk (KERN_WARNING "jffs2_check_nand_cleanmarker(): Read OOB failed %d for block at %08x\n", ret, jeb->offset));
1038 if (retlen < (oob_size << 1)) {
1039 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));
1043 /* Check cleanmarker only on the first physical block */
1045 n.magic = cpu_to_je16 (JFFS2_MAGIC_BITMASK);
1046 n.nodetype = cpu_to_je16 (JFFS2_NODETYPE_CLEANMARKER);
1047 n.totlen = cpu_to_je32 (8);
1048 p = (unsigned char *) &n;
1050 for (i = 0; i < c->fsdata_len; i++) {
1051 if (buf[c->fsdata_pos + i] != p[i]) {
1055 D1(if (retval == 1) {
1056 printk(KERN_WARNING "jffs2_check_nand_cleanmarker(): Cleanmarker node not detected in block at %08x\n", jeb->offset);
1057 printk(KERN_WARNING "OOB at %08x was ", offset);
1058 for (i=0; i < oob_size; i++) {
1059 printk("%02x ", buf[i]);
1064 offset += c->mtd->erasesize;
1069 int jffs2_write_nand_cleanmarker(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
1071 struct jffs2_unknown_node n;
1075 n.magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
1076 n.nodetype = cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER);
1077 n.totlen = cpu_to_je32(8);
1079 ret = jffs2_flash_write_oob(c, jeb->offset + c->fsdata_pos, c->fsdata_len, &retlen, (unsigned char *)&n);
1082 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1085 if (retlen != c->fsdata_len) {
1086 D1(printk(KERN_WARNING "jffs2_write_nand_cleanmarker(): Short write for block at %08x: %zd not %d\n", jeb->offset, retlen, c->fsdata_len));
1093 * On NAND we try to mark this block bad. If the block was erased more
1094 * than MAX_ERASE_FAILURES we mark it finaly bad.
1095 * Don't care about failures. This block remains on the erase-pending
1096 * or badblock list as long as nobody manipulates the flash with
1097 * a bootloader or something like that.
1100 int jffs2_write_nand_badblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
1104 /* if the count is < max, we try to write the counter to the 2nd page oob area */
1105 if( ++jeb->bad_count < MAX_ERASE_FAILURES)
1108 if (!c->mtd->block_markbad)
1109 return 1; // What else can we do?
1111 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Marking bad block at %08x\n", bad_offset));
1112 ret = c->mtd->block_markbad(c->mtd, bad_offset);
1115 D1(printk(KERN_WARNING "jffs2_write_nand_badblock(): Write failed for block at %08x: error %d\n", jeb->offset, ret));
1121 #define NAND_JFFS2_OOB16_FSDALEN 8
1123 static struct nand_oobinfo jffs2_oobinfo_docecc = {
1124 .useecc = MTD_NANDECC_PLACE,
1126 .eccpos = {0,1,2,3,4,5}
1130 static int jffs2_nand_set_oobinfo(struct jffs2_sb_info *c)
1132 struct nand_oobinfo *oinfo = &c->mtd->oobinfo;
1134 /* Do this only, if we have an oob buffer */
1135 if (!c->mtd->oobsize)
1138 /* Cleanmarker is out-of-band, so inline size zero */
1139 c->cleanmarker_size = 0;
1141 /* Should we use autoplacement ? */
1142 if (oinfo && oinfo->useecc == MTD_NANDECC_AUTOPLACE) {
1143 D1(printk(KERN_DEBUG "JFFS2 using autoplace on NAND\n"));
1144 /* Get the position of the free bytes */
1145 if (!oinfo->oobfree[0][1]) {
1146 printk (KERN_WARNING "jffs2_nand_set_oobinfo(): Eeep. Autoplacement selected and no empty space in oob\n");
1149 c->fsdata_pos = oinfo->oobfree[0][0];
1150 c->fsdata_len = oinfo->oobfree[0][1];
1151 if (c->fsdata_len > 8)
1154 /* This is just a legacy fallback and should go away soon */
1155 switch(c->mtd->ecctype) {
1156 case MTD_ECC_RS_DiskOnChip:
1157 printk(KERN_WARNING "JFFS2 using DiskOnChip hardware ECC without autoplacement. Fix it!\n");
1158 c->oobinfo = &jffs2_oobinfo_docecc;
1160 c->fsdata_len = NAND_JFFS2_OOB16_FSDALEN;
1161 c->badblock_pos = 15;
1165 D1(printk(KERN_DEBUG "JFFS2 on NAND. No autoplacment info found\n"));
1172 int jffs2_nand_flash_setup(struct jffs2_sb_info *c)
1176 /* Initialise write buffer */
1177 init_rwsem(&c->wbuf_sem);
1178 c->wbuf_pagesize = c->mtd->oobblock;
1179 c->wbuf_ofs = 0xFFFFFFFF;
1181 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1185 res = jffs2_nand_set_oobinfo(c);
1189 brokenbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1194 memset(brokenbuf, 0xdb, c->wbuf_pagesize);
1199 void jffs2_nand_flash_cleanup(struct jffs2_sb_info *c)
1204 int jffs2_dataflash_setup(struct jffs2_sb_info *c) {
1205 c->cleanmarker_size = 0; /* No cleanmarkers needed */
1207 /* Initialize write buffer */
1208 init_rwsem(&c->wbuf_sem);
1211 c->wbuf_pagesize = c->mtd->erasesize;
1213 /* Find a suitable c->sector_size
1214 * - Not too much sectors
1215 * - Sectors have to be at least 4 K + some bytes
1216 * - All known dataflashes have erase sizes of 528 or 1056
1217 * - we take at least 8 eraseblocks and want to have at least 8K size
1218 * - The concatenation should be a power of 2
1221 c->sector_size = 8 * c->mtd->erasesize;
1223 while (c->sector_size < 8192) {
1224 c->sector_size *= 2;
1227 /* It may be necessary to adjust the flash size */
1228 c->flash_size = c->mtd->size;
1230 if ((c->flash_size % c->sector_size) != 0) {
1231 c->flash_size = (c->flash_size / c->sector_size) * c->sector_size;
1232 printk(KERN_WARNING "JFFS2 flash size adjusted to %dKiB\n", c->flash_size);
1235 c->wbuf_ofs = 0xFFFFFFFF;
1236 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1240 printk(KERN_INFO "JFFS2 write-buffering enabled buffer (%d) erasesize (%d)\n", c->wbuf_pagesize, c->sector_size);
1245 void jffs2_dataflash_cleanup(struct jffs2_sb_info *c) {
1249 int jffs2_nor_ecc_flash_setup(struct jffs2_sb_info *c) {
1250 /* Cleanmarker is actually larger on the flashes */
1251 c->cleanmarker_size = 16;
1253 /* Initialize write buffer */
1254 init_rwsem(&c->wbuf_sem);
1255 c->wbuf_pagesize = c->mtd->eccsize;
1256 c->wbuf_ofs = 0xFFFFFFFF;
1258 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1265 void jffs2_nor_ecc_flash_cleanup(struct jffs2_sb_info *c) {
1269 int jffs2_nor_wbuf_flash_setup(struct jffs2_sb_info *c) {
1270 /* Cleanmarker currently occupies a whole programming region */
1271 c->cleanmarker_size = MTD_PROGREGION_SIZE(c->mtd);
1273 /* Initialize write buffer */
1274 init_rwsem(&c->wbuf_sem);
1275 c->wbuf_pagesize = MTD_PROGREGION_SIZE(c->mtd);
1276 c->wbuf_ofs = 0xFFFFFFFF;
1278 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1285 void jffs2_nor_wbuf_flash_cleanup(struct jffs2_sb_info *c) {