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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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/rawnand.h>
21 #include <linux/jiffies.h>
22 #include <linux/sched.h>
23 #include <linux/writeback.h>
27 /* For testing write failures */
32 static unsigned char *brokenbuf;
35 #define PAGE_DIV(x) ( ((unsigned long)(x) / (unsigned long)(c->wbuf_pagesize)) * (unsigned long)(c->wbuf_pagesize) )
36 #define PAGE_MOD(x) ( (unsigned long)(x) % (unsigned long)(c->wbuf_pagesize) )
38 /* max. erase failures before we mark a block bad */
39 #define MAX_ERASE_FAILURES 2
41 struct jffs2_inodirty {
43 struct jffs2_inodirty *next;
46 static struct jffs2_inodirty inodirty_nomem;
48 static int jffs2_wbuf_pending_for_ino(struct jffs2_sb_info *c, uint32_t ino)
50 struct jffs2_inodirty *this = c->wbuf_inodes;
52 /* If a malloc failed, consider _everything_ dirty */
53 if (this == &inodirty_nomem)
56 /* If ino == 0, _any_ non-GC writes mean 'yes' */
60 /* Look to see if the inode in question is pending in the wbuf */
69 static void jffs2_clear_wbuf_ino_list(struct jffs2_sb_info *c)
71 struct jffs2_inodirty *this;
73 this = c->wbuf_inodes;
75 if (this != &inodirty_nomem) {
77 struct jffs2_inodirty *next = this->next;
82 c->wbuf_inodes = NULL;
85 static void jffs2_wbuf_dirties_inode(struct jffs2_sb_info *c, uint32_t ino)
87 struct jffs2_inodirty *new;
89 /* Schedule delayed write-buffer write-out */
90 jffs2_dirty_trigger(c);
92 if (jffs2_wbuf_pending_for_ino(c, ino))
95 new = kmalloc(sizeof(*new), GFP_KERNEL);
97 jffs2_dbg(1, "No memory to allocate inodirty. Fallback to all considered dirty\n");
98 jffs2_clear_wbuf_ino_list(c);
99 c->wbuf_inodes = &inodirty_nomem;
103 new->next = c->wbuf_inodes;
104 c->wbuf_inodes = new;
108 static inline void jffs2_refile_wbuf_blocks(struct jffs2_sb_info *c)
110 struct list_head *this, *next;
113 if (list_empty(&c->erasable_pending_wbuf_list))
116 list_for_each_safe(this, next, &c->erasable_pending_wbuf_list) {
117 struct jffs2_eraseblock *jeb = list_entry(this, struct jffs2_eraseblock, list);
119 jffs2_dbg(1, "Removing eraseblock at 0x%08x from erasable_pending_wbuf_list...\n",
122 if ((jiffies + (n++)) & 127) {
123 /* Most of the time, we just erase it immediately. Otherwise we
124 spend ages scanning it on mount, etc. */
125 jffs2_dbg(1, "...and adding to erase_pending_list\n");
126 list_add_tail(&jeb->list, &c->erase_pending_list);
127 c->nr_erasing_blocks++;
128 jffs2_garbage_collect_trigger(c);
130 /* Sometimes, however, we leave it elsewhere so it doesn't get
131 immediately reused, and we spread the load a bit. */
132 jffs2_dbg(1, "...and adding to erasable_list\n");
133 list_add_tail(&jeb->list, &c->erasable_list);
138 #define REFILE_NOTEMPTY 0
139 #define REFILE_ANYWAY 1
141 static void jffs2_block_refile(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, int allow_empty)
143 jffs2_dbg(1, "About to refile bad block at %08x\n", jeb->offset);
145 /* File the existing block on the bad_used_list.... */
146 if (c->nextblock == jeb)
148 else /* Not sure this should ever happen... need more coffee */
149 list_del(&jeb->list);
150 if (jeb->first_node) {
151 jffs2_dbg(1, "Refiling block at %08x to bad_used_list\n",
153 list_add(&jeb->list, &c->bad_used_list);
155 BUG_ON(allow_empty == REFILE_NOTEMPTY);
156 /* It has to have had some nodes or we couldn't be here */
157 jffs2_dbg(1, "Refiling block at %08x to erase_pending_list\n",
159 list_add(&jeb->list, &c->erase_pending_list);
160 c->nr_erasing_blocks++;
161 jffs2_garbage_collect_trigger(c);
164 if (!jffs2_prealloc_raw_node_refs(c, jeb, 1)) {
165 uint32_t oldfree = jeb->free_size;
167 jffs2_link_node_ref(c, jeb,
168 (jeb->offset+c->sector_size-oldfree) | REF_OBSOLETE,
170 /* convert to wasted */
171 c->wasted_size += oldfree;
172 jeb->wasted_size += oldfree;
173 c->dirty_size -= oldfree;
174 jeb->dirty_size -= oldfree;
177 jffs2_dbg_dump_block_lists_nolock(c);
178 jffs2_dbg_acct_sanity_check_nolock(c,jeb);
179 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
182 static struct jffs2_raw_node_ref **jffs2_incore_replace_raw(struct jffs2_sb_info *c,
183 struct jffs2_inode_info *f,
184 struct jffs2_raw_node_ref *raw,
185 union jffs2_node_union *node)
187 struct jffs2_node_frag *frag;
188 struct jffs2_full_dirent *fd;
190 dbg_noderef("incore_replace_raw: node at %p is {%04x,%04x}\n",
191 node, je16_to_cpu(node->u.magic), je16_to_cpu(node->u.nodetype));
193 BUG_ON(je16_to_cpu(node->u.magic) != 0x1985 &&
194 je16_to_cpu(node->u.magic) != 0);
196 switch (je16_to_cpu(node->u.nodetype)) {
197 case JFFS2_NODETYPE_INODE:
198 if (f->metadata && f->metadata->raw == raw) {
199 dbg_noderef("Will replace ->raw in f->metadata at %p\n", f->metadata);
200 return &f->metadata->raw;
202 frag = jffs2_lookup_node_frag(&f->fragtree, je32_to_cpu(node->i.offset));
204 /* Find a frag which refers to the full_dnode we want to modify */
205 while (!frag->node || frag->node->raw != raw) {
206 frag = frag_next(frag);
209 dbg_noderef("Will replace ->raw in full_dnode at %p\n", frag->node);
210 return &frag->node->raw;
212 case JFFS2_NODETYPE_DIRENT:
213 for (fd = f->dents; fd; fd = fd->next) {
214 if (fd->raw == raw) {
215 dbg_noderef("Will replace ->raw in full_dirent at %p\n", fd);
222 dbg_noderef("Don't care about replacing raw for nodetype %x\n",
223 je16_to_cpu(node->u.nodetype));
229 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
230 static int jffs2_verify_write(struct jffs2_sb_info *c, unsigned char *buf,
237 ret = mtd_read(c->mtd, ofs, c->wbuf_pagesize, &retlen, c->wbuf_verify);
238 if (ret && ret != -EUCLEAN && ret != -EBADMSG) {
239 pr_warn("%s(): Read back of page at %08x failed: %d\n",
240 __func__, c->wbuf_ofs, ret);
242 } else if (retlen != c->wbuf_pagesize) {
243 pr_warn("%s(): Read back of page at %08x gave short read: %zd not %d\n",
244 __func__, ofs, retlen, c->wbuf_pagesize);
247 if (!memcmp(buf, c->wbuf_verify, c->wbuf_pagesize))
251 eccstr = "corrected";
252 else if (ret == -EBADMSG)
253 eccstr = "correction failed";
255 eccstr = "OK or unused";
257 pr_warn("Write verify error (ECC %s) at %08x. Wrote:\n",
258 eccstr, c->wbuf_ofs);
259 print_hex_dump(KERN_WARNING, "", DUMP_PREFIX_OFFSET, 16, 1,
260 c->wbuf, c->wbuf_pagesize, 0);
262 pr_warn("Read back:\n");
263 print_hex_dump(KERN_WARNING, "", DUMP_PREFIX_OFFSET, 16, 1,
264 c->wbuf_verify, c->wbuf_pagesize, 0);
269 #define jffs2_verify_write(c,b,o) (0)
272 /* Recover from failure to write wbuf. Recover the nodes up to the
273 * wbuf, not the one which we were starting to try to write. */
275 static void jffs2_wbuf_recover(struct jffs2_sb_info *c)
277 struct jffs2_eraseblock *jeb, *new_jeb;
278 struct jffs2_raw_node_ref *raw, *next, *first_raw = NULL;
283 uint32_t start, end, ofs, len;
285 jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
287 spin_lock(&c->erase_completion_lock);
288 if (c->wbuf_ofs % c->mtd->erasesize)
289 jffs2_block_refile(c, jeb, REFILE_NOTEMPTY);
291 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
292 spin_unlock(&c->erase_completion_lock);
294 BUG_ON(!ref_obsolete(jeb->last_node));
296 /* Find the first node to be recovered, by skipping over every
297 node which ends before the wbuf starts, or which is obsolete. */
298 for (next = raw = jeb->first_node; next; raw = next) {
299 next = ref_next(raw);
301 if (ref_obsolete(raw) ||
302 (next && ref_offset(next) <= c->wbuf_ofs)) {
303 dbg_noderef("Skipping node at 0x%08x(%d)-0x%08x which is either before 0x%08x or obsolete\n",
304 ref_offset(raw), ref_flags(raw),
305 (ref_offset(raw) + ref_totlen(c, jeb, raw)),
309 dbg_noderef("First node to be recovered is at 0x%08x(%d)-0x%08x\n",
310 ref_offset(raw), ref_flags(raw),
311 (ref_offset(raw) + ref_totlen(c, jeb, raw)));
318 /* All nodes were obsolete. Nothing to recover. */
319 jffs2_dbg(1, "No non-obsolete nodes to be recovered. Just filing block bad\n");
324 start = ref_offset(first_raw);
325 end = ref_offset(jeb->last_node);
328 /* Count the number of refs which need to be copied */
329 while ((raw = ref_next(raw)) != jeb->last_node)
332 dbg_noderef("wbuf recover %08x-%08x (%d bytes in %d nodes)\n",
333 start, end, end - start, nr_refile);
336 if (start < c->wbuf_ofs) {
337 /* First affected node was already partially written.
338 * Attempt to reread the old data into our buffer. */
340 buf = kmalloc(end - start, GFP_KERNEL);
342 pr_crit("Malloc failure in wbuf recovery. Data loss ensues.\n");
348 ret = mtd_read(c->mtd, start, c->wbuf_ofs - start, &retlen,
351 /* ECC recovered ? */
352 if ((ret == -EUCLEAN || ret == -EBADMSG) &&
353 (retlen == c->wbuf_ofs - start))
356 if (ret || retlen != c->wbuf_ofs - start) {
357 pr_crit("Old data are already lost in wbuf recovery. Data loss ensues.\n");
362 first_raw = ref_next(first_raw);
364 while (first_raw && ref_obsolete(first_raw)) {
365 first_raw = ref_next(first_raw);
369 /* If this was the only node to be recovered, give up */
375 /* It wasn't. Go on and try to recover nodes complete in the wbuf */
376 start = ref_offset(first_raw);
377 dbg_noderef("wbuf now recover %08x-%08x (%d bytes in %d nodes)\n",
378 start, end, end - start, nr_refile);
381 /* Read succeeded. Copy the remaining data from the wbuf */
382 memcpy(buf + (c->wbuf_ofs - start), c->wbuf, end - c->wbuf_ofs);
385 /* OK... we're to rewrite (end-start) bytes of data from first_raw onwards.
386 Either 'buf' contains the data, or we find it in the wbuf */
388 /* ... and get an allocation of space from a shiny new block instead */
389 ret = jffs2_reserve_space_gc(c, end-start, &len, JFFS2_SUMMARY_NOSUM_SIZE);
391 pr_warn("Failed to allocate space for wbuf recovery. Data loss ensues.\n");
396 /* The summary is not recovered, so it must be disabled for this erase block */
397 jffs2_sum_disable_collecting(c->summary);
399 ret = jffs2_prealloc_raw_node_refs(c, c->nextblock, nr_refile);
401 pr_warn("Failed to allocate node refs for wbuf recovery. Data loss ensues.\n");
408 if (end-start >= c->wbuf_pagesize) {
409 /* Need to do another write immediately, but it's possible
410 that this is just because the wbuf itself is completely
411 full, and there's nothing earlier read back from the
412 flash. Hence 'buf' isn't necessarily what we're writing
414 unsigned char *rewrite_buf = buf?:c->wbuf;
415 uint32_t towrite = (end-start) - ((end-start)%c->wbuf_pagesize);
417 jffs2_dbg(1, "Write 0x%x bytes at 0x%08x in wbuf recover\n",
422 if (breakme++ == 20) {
423 pr_notice("Faking write error at 0x%08x\n", ofs);
425 mtd_write(c->mtd, ofs, towrite, &retlen, brokenbuf);
429 ret = mtd_write(c->mtd, ofs, towrite, &retlen,
432 if (ret || retlen != towrite || jffs2_verify_write(c, rewrite_buf, ofs)) {
433 /* Argh. We tried. Really we did. */
434 pr_crit("Recovery of wbuf failed due to a second write error\n");
438 jffs2_add_physical_node_ref(c, ofs | REF_OBSOLETE, ref_totlen(c, jeb, first_raw), NULL);
442 pr_notice("Recovery of wbuf succeeded to %08x\n", ofs);
444 c->wbuf_len = (end - start) - towrite;
445 c->wbuf_ofs = ofs + towrite;
446 memmove(c->wbuf, rewrite_buf + towrite, c->wbuf_len);
447 /* Don't muck about with c->wbuf_inodes. False positives are harmless. */
449 /* OK, now we're left with the dregs in whichever buffer we're using */
451 memcpy(c->wbuf, buf, end-start);
453 memmove(c->wbuf, c->wbuf + (start - c->wbuf_ofs), end - start);
456 c->wbuf_len = end - start;
459 /* Now sort out the jffs2_raw_node_refs, moving them from the old to the next block */
460 new_jeb = &c->blocks[ofs / c->sector_size];
462 spin_lock(&c->erase_completion_lock);
463 for (raw = first_raw; raw != jeb->last_node; raw = ref_next(raw)) {
464 uint32_t rawlen = ref_totlen(c, jeb, raw);
465 struct jffs2_inode_cache *ic;
466 struct jffs2_raw_node_ref *new_ref;
467 struct jffs2_raw_node_ref **adjust_ref = NULL;
468 struct jffs2_inode_info *f = NULL;
470 jffs2_dbg(1, "Refiling block of %08x at %08x(%d) to %08x\n",
471 rawlen, ref_offset(raw), ref_flags(raw), ofs);
473 ic = jffs2_raw_ref_to_ic(raw);
475 /* Ick. This XATTR mess should be fixed shortly... */
476 if (ic && ic->class == RAWNODE_CLASS_XATTR_DATUM) {
477 struct jffs2_xattr_datum *xd = (void *)ic;
478 BUG_ON(xd->node != raw);
479 adjust_ref = &xd->node;
480 raw->next_in_ino = NULL;
482 } else if (ic && ic->class == RAWNODE_CLASS_XATTR_REF) {
483 struct jffs2_xattr_datum *xr = (void *)ic;
484 BUG_ON(xr->node != raw);
485 adjust_ref = &xr->node;
486 raw->next_in_ino = NULL;
488 } else if (ic && ic->class == RAWNODE_CLASS_INODE_CACHE) {
489 struct jffs2_raw_node_ref **p = &ic->nodes;
491 /* Remove the old node from the per-inode list */
492 while (*p && *p != (void *)ic) {
494 (*p) = (raw->next_in_ino);
495 raw->next_in_ino = NULL;
498 p = &((*p)->next_in_ino);
501 if (ic->state == INO_STATE_PRESENT && !ref_obsolete(raw)) {
502 /* If it's an in-core inode, then we have to adjust any
503 full_dirent or full_dnode structure to point to the
504 new version instead of the old */
505 f = jffs2_gc_fetch_inode(c, ic->ino, !ic->pino_nlink);
507 /* Should never happen; it _must_ be present */
508 JFFS2_ERROR("Failed to iget() ino #%u, err %ld\n",
509 ic->ino, PTR_ERR(f));
512 /* We don't lock f->sem. There's a number of ways we could
513 end up in here with it already being locked, and nobody's
514 going to modify it on us anyway because we hold the
515 alloc_sem. We're only changing one ->raw pointer too,
516 which we can get away with without upsetting readers. */
517 adjust_ref = jffs2_incore_replace_raw(c, f, raw,
518 (void *)(buf?:c->wbuf) + (ref_offset(raw) - start));
519 } else if (unlikely(ic->state != INO_STATE_PRESENT &&
520 ic->state != INO_STATE_CHECKEDABSENT &&
521 ic->state != INO_STATE_GC)) {
522 JFFS2_ERROR("Inode #%u is in strange state %d!\n", ic->ino, ic->state);
527 new_ref = jffs2_link_node_ref(c, new_jeb, ofs | ref_flags(raw), rawlen, ic);
530 BUG_ON(*adjust_ref != raw);
531 *adjust_ref = new_ref;
534 jffs2_gc_release_inode(c, f);
536 if (!ref_obsolete(raw)) {
537 jeb->dirty_size += rawlen;
538 jeb->used_size -= rawlen;
539 c->dirty_size += rawlen;
540 c->used_size -= rawlen;
541 raw->flash_offset = ref_offset(raw) | REF_OBSOLETE;
542 BUG_ON(raw->next_in_ino);
549 /* Fix up the original jeb now it's on the bad_list */
550 if (first_raw == jeb->first_node) {
551 jffs2_dbg(1, "Failing block at %08x is now empty. Moving to erase_pending_list\n",
553 list_move(&jeb->list, &c->erase_pending_list);
554 c->nr_erasing_blocks++;
555 jffs2_garbage_collect_trigger(c);
558 jffs2_dbg_acct_sanity_check_nolock(c, jeb);
559 jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
561 jffs2_dbg_acct_sanity_check_nolock(c, new_jeb);
562 jffs2_dbg_acct_paranoia_check_nolock(c, new_jeb);
564 spin_unlock(&c->erase_completion_lock);
566 jffs2_dbg(1, "wbuf recovery completed OK. wbuf_ofs 0x%08x, len 0x%x\n",
567 c->wbuf_ofs, c->wbuf_len);
571 /* Meaning of pad argument:
572 0: Do not pad. Probably pointless - we only ever use this when we can't pad anyway.
573 1: Pad, do not adjust nextblock free_size
574 2: Pad, adjust nextblock free_size
577 #define PAD_NOACCOUNT 1
578 #define PAD_ACCOUNTING 2
580 static int __jffs2_flush_wbuf(struct jffs2_sb_info *c, int pad)
582 struct jffs2_eraseblock *wbuf_jeb;
586 /* Nothing to do if not write-buffering the flash. In particular, we shouldn't
587 del_timer() the timer we never initialised. */
588 if (!jffs2_is_writebuffered(c))
591 if (!mutex_is_locked(&c->alloc_sem)) {
592 pr_crit("jffs2_flush_wbuf() called with alloc_sem not locked!\n");
596 if (!c->wbuf_len) /* already checked c->wbuf above */
599 wbuf_jeb = &c->blocks[c->wbuf_ofs / c->sector_size];
600 if (jffs2_prealloc_raw_node_refs(c, wbuf_jeb, c->nextblock->allocated_refs + 1))
603 /* claim remaining space on the page
604 this happens, if we have a change to a new block,
605 or if fsync forces us to flush the writebuffer.
606 if we have a switch to next page, we will not have
607 enough remaining space for this.
610 c->wbuf_len = PAD(c->wbuf_len);
612 /* Pad with JFFS2_DIRTY_BITMASK initially. this helps out ECC'd NOR
613 with 8 byte page size */
614 memset(c->wbuf + c->wbuf_len, 0, c->wbuf_pagesize - c->wbuf_len);
616 if ( c->wbuf_len + sizeof(struct jffs2_unknown_node) < c->wbuf_pagesize) {
617 struct jffs2_unknown_node *padnode = (void *)(c->wbuf + c->wbuf_len);
618 padnode->magic = cpu_to_je16(JFFS2_MAGIC_BITMASK);
619 padnode->nodetype = cpu_to_je16(JFFS2_NODETYPE_PADDING);
620 padnode->totlen = cpu_to_je32(c->wbuf_pagesize - c->wbuf_len);
621 padnode->hdr_crc = cpu_to_je32(crc32(0, padnode, sizeof(*padnode)-4));
624 /* else jffs2_flash_writev has actually filled in the rest of the
625 buffer for us, and will deal with the node refs etc. later. */
629 if (breakme++ == 20) {
630 pr_notice("Faking write error at 0x%08x\n", c->wbuf_ofs);
632 mtd_write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize, &retlen,
638 ret = mtd_write(c->mtd, c->wbuf_ofs, c->wbuf_pagesize,
642 pr_warn("jffs2_flush_wbuf(): Write failed with %d\n", ret);
644 } else if (retlen != c->wbuf_pagesize) {
645 pr_warn("jffs2_flush_wbuf(): Write was short: %zd instead of %d\n",
646 retlen, c->wbuf_pagesize);
649 } else if ((ret = jffs2_verify_write(c, c->wbuf, c->wbuf_ofs))) {
651 jffs2_wbuf_recover(c);
656 /* Adjust free size of the block if we padded. */
658 uint32_t waste = c->wbuf_pagesize - c->wbuf_len;
660 jffs2_dbg(1, "jffs2_flush_wbuf() adjusting free_size of %sblock at %08x\n",
661 (wbuf_jeb == c->nextblock) ? "next" : "",
664 /* wbuf_pagesize - wbuf_len is the amount of space that's to be
665 padded. If there is less free space in the block than that,
666 something screwed up */
667 if (wbuf_jeb->free_size < waste) {
668 pr_crit("jffs2_flush_wbuf(): Accounting error. wbuf at 0x%08x has 0x%03x bytes, 0x%03x left.\n",
669 c->wbuf_ofs, c->wbuf_len, waste);
670 pr_crit("jffs2_flush_wbuf(): But free_size for block at 0x%08x is only 0x%08x\n",
671 wbuf_jeb->offset, wbuf_jeb->free_size);
675 spin_lock(&c->erase_completion_lock);
677 jffs2_link_node_ref(c, wbuf_jeb, (c->wbuf_ofs + c->wbuf_len) | REF_OBSOLETE, waste, NULL);
678 /* FIXME: that made it count as dirty. Convert to wasted */
679 wbuf_jeb->dirty_size -= waste;
680 c->dirty_size -= waste;
681 wbuf_jeb->wasted_size += waste;
682 c->wasted_size += waste;
684 spin_lock(&c->erase_completion_lock);
686 /* Stick any now-obsoleted blocks on the erase_pending_list */
687 jffs2_refile_wbuf_blocks(c);
688 jffs2_clear_wbuf_ino_list(c);
689 spin_unlock(&c->erase_completion_lock);
691 memset(c->wbuf,0xff,c->wbuf_pagesize);
692 /* adjust write buffer offset, else we get a non contiguous write bug */
693 c->wbuf_ofs += c->wbuf_pagesize;
698 /* Trigger garbage collection to flush the write-buffer.
699 If ino arg is zero, do it if _any_ real (i.e. not GC) writes are
700 outstanding. If ino arg non-zero, do it only if a write for the
701 given inode is outstanding. */
702 int jffs2_flush_wbuf_gc(struct jffs2_sb_info *c, uint32_t ino)
704 uint32_t old_wbuf_ofs;
705 uint32_t old_wbuf_len;
708 jffs2_dbg(1, "jffs2_flush_wbuf_gc() called for ino #%u...\n", ino);
713 mutex_lock(&c->alloc_sem);
714 if (!jffs2_wbuf_pending_for_ino(c, ino)) {
715 jffs2_dbg(1, "Ino #%d not pending in wbuf. Returning\n", ino);
716 mutex_unlock(&c->alloc_sem);
720 old_wbuf_ofs = c->wbuf_ofs;
721 old_wbuf_len = c->wbuf_len;
723 if (c->unchecked_size) {
724 /* GC won't make any progress for a while */
725 jffs2_dbg(1, "%s(): padding. Not finished checking\n",
727 down_write(&c->wbuf_sem);
728 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
729 /* retry flushing wbuf in case jffs2_wbuf_recover
730 left some data in the wbuf */
732 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
733 up_write(&c->wbuf_sem);
734 } else while (old_wbuf_len &&
735 old_wbuf_ofs == c->wbuf_ofs) {
737 mutex_unlock(&c->alloc_sem);
739 jffs2_dbg(1, "%s(): calls gc pass\n", __func__);
741 ret = jffs2_garbage_collect_pass(c);
743 /* GC failed. Flush it with padding instead */
744 mutex_lock(&c->alloc_sem);
745 down_write(&c->wbuf_sem);
746 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
747 /* retry flushing wbuf in case jffs2_wbuf_recover
748 left some data in the wbuf */
750 ret = __jffs2_flush_wbuf(c, PAD_ACCOUNTING);
751 up_write(&c->wbuf_sem);
754 mutex_lock(&c->alloc_sem);
757 jffs2_dbg(1, "%s(): ends...\n", __func__);
759 mutex_unlock(&c->alloc_sem);
763 /* Pad write-buffer to end and write it, wasting space. */
764 int jffs2_flush_wbuf_pad(struct jffs2_sb_info *c)
771 down_write(&c->wbuf_sem);
772 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
773 /* retry - maybe wbuf recover left some data in wbuf. */
775 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
776 up_write(&c->wbuf_sem);
781 static size_t jffs2_fill_wbuf(struct jffs2_sb_info *c, const uint8_t *buf,
784 if (len && !c->wbuf_len && (len >= c->wbuf_pagesize))
787 if (len > (c->wbuf_pagesize - c->wbuf_len))
788 len = c->wbuf_pagesize - c->wbuf_len;
789 memcpy(c->wbuf + c->wbuf_len, buf, len);
790 c->wbuf_len += (uint32_t) len;
794 int jffs2_flash_writev(struct jffs2_sb_info *c, const struct kvec *invecs,
795 unsigned long count, loff_t to, size_t *retlen,
798 struct jffs2_eraseblock *jeb;
799 size_t wbuf_retlen, donelen = 0;
800 uint32_t outvec_to = to;
803 /* If not writebuffered flash, don't bother */
804 if (!jffs2_is_writebuffered(c))
805 return jffs2_flash_direct_writev(c, invecs, count, to, retlen);
807 down_write(&c->wbuf_sem);
809 /* If wbuf_ofs is not initialized, set it to target address */
810 if (c->wbuf_ofs == 0xFFFFFFFF) {
811 c->wbuf_ofs = PAGE_DIV(to);
812 c->wbuf_len = PAGE_MOD(to);
813 memset(c->wbuf,0xff,c->wbuf_pagesize);
817 * Sanity checks on target address. It's permitted to write
818 * at PAD(c->wbuf_len+c->wbuf_ofs), and it's permitted to
819 * write at the beginning of a new erase block. Anything else,
820 * and you die. New block starts at xxx000c (0-b = block
823 if (SECTOR_ADDR(to) != SECTOR_ADDR(c->wbuf_ofs)) {
824 /* It's a write to a new block */
826 jffs2_dbg(1, "%s(): to 0x%lx causes flush of wbuf at 0x%08x\n",
827 __func__, (unsigned long)to, c->wbuf_ofs);
828 ret = __jffs2_flush_wbuf(c, PAD_NOACCOUNT);
832 /* set pointer to new block */
833 c->wbuf_ofs = PAGE_DIV(to);
834 c->wbuf_len = PAGE_MOD(to);
837 if (to != PAD(c->wbuf_ofs + c->wbuf_len)) {
838 /* We're not writing immediately after the writebuffer. Bad. */
839 pr_crit("%s(): Non-contiguous write to %08lx\n",
840 __func__, (unsigned long)to);
842 pr_crit("wbuf was previously %08x-%08x\n",
843 c->wbuf_ofs, c->wbuf_ofs + c->wbuf_len);
847 /* adjust alignment offset */
848 if (c->wbuf_len != PAGE_MOD(to)) {
849 c->wbuf_len = PAGE_MOD(to);
850 /* take care of alignment to next page */
852 c->wbuf_len = c->wbuf_pagesize;
853 ret = __jffs2_flush_wbuf(c, NOPAD);
859 for (invec = 0; invec < count; invec++) {
860 int vlen = invecs[invec].iov_len;
861 uint8_t *v = invecs[invec].iov_base;
863 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
865 if (c->wbuf_len == c->wbuf_pagesize) {
866 ret = __jffs2_flush_wbuf(c, NOPAD);
871 outvec_to += wbuf_retlen;
872 donelen += wbuf_retlen;
875 if (vlen >= c->wbuf_pagesize) {
876 ret = mtd_write(c->mtd, outvec_to, PAGE_DIV(vlen),
878 if (ret < 0 || wbuf_retlen != PAGE_DIV(vlen))
882 outvec_to += wbuf_retlen;
883 c->wbuf_ofs = outvec_to;
884 donelen += wbuf_retlen;
888 wbuf_retlen = jffs2_fill_wbuf(c, v, vlen);
889 if (c->wbuf_len == c->wbuf_pagesize) {
890 ret = __jffs2_flush_wbuf(c, NOPAD);
895 outvec_to += wbuf_retlen;
896 donelen += wbuf_retlen;
900 * If there's a remainder in the wbuf and it's a non-GC write,
901 * remember that the wbuf affects this ino
905 if (jffs2_sum_active()) {
906 int res = jffs2_sum_add_kvec(c, invecs, count, (uint32_t) to);
911 if (c->wbuf_len && ino)
912 jffs2_wbuf_dirties_inode(c, ino);
915 up_write(&c->wbuf_sem);
920 * At this point we have no problem, c->wbuf is empty. However
921 * refile nextblock to avoid writing again to same address.
924 spin_lock(&c->erase_completion_lock);
926 jeb = &c->blocks[outvec_to / c->sector_size];
927 jffs2_block_refile(c, jeb, REFILE_ANYWAY);
929 spin_unlock(&c->erase_completion_lock);
933 up_write(&c->wbuf_sem);
938 * This is the entry for flash write.
939 * Check, if we work on NAND FLASH, if so build an kvec and write it via vritev
941 int jffs2_flash_write(struct jffs2_sb_info *c, loff_t ofs, size_t len,
942 size_t *retlen, const u_char *buf)
946 if (!jffs2_is_writebuffered(c))
947 return jffs2_flash_direct_write(c, ofs, len, retlen, buf);
949 vecs[0].iov_base = (unsigned char *) buf;
950 vecs[0].iov_len = len;
951 return jffs2_flash_writev(c, vecs, 1, ofs, retlen, 0);
955 Handle readback from writebuffer and ECC failure return
957 int jffs2_flash_read(struct jffs2_sb_info *c, loff_t ofs, size_t len, size_t *retlen, u_char *buf)
959 loff_t orbf = 0, owbf = 0, lwbf = 0;
962 if (!jffs2_is_writebuffered(c))
963 return mtd_read(c->mtd, ofs, len, retlen, buf);
966 down_read(&c->wbuf_sem);
967 ret = mtd_read(c->mtd, ofs, len, retlen, buf);
969 if ( (ret == -EBADMSG || ret == -EUCLEAN) && (*retlen == len) ) {
971 pr_warn("mtd->read(0x%zx bytes from 0x%llx) returned ECC error\n",
974 * We have the raw data without ECC correction in the buffer,
975 * maybe we are lucky and all data or parts are correct. We
976 * check the node. If data are corrupted node check will sort
977 * it out. We keep this block, it will fail on write or erase
978 * and the we mark it bad. Or should we do that now? But we
979 * should give him a chance. Maybe we had a system crash or
980 * power loss before the ecc write or a erase was completed.
981 * So we return success. :)
986 /* if no writebuffer available or write buffer empty, return */
987 if (!c->wbuf_pagesize || !c->wbuf_len)
990 /* if we read in a different block, return */
991 if (SECTOR_ADDR(ofs) != SECTOR_ADDR(c->wbuf_ofs))
994 if (ofs >= c->wbuf_ofs) {
995 owbf = (ofs - c->wbuf_ofs); /* offset in write buffer */
996 if (owbf > c->wbuf_len) /* is read beyond write buffer ? */
998 lwbf = c->wbuf_len - owbf; /* number of bytes to copy */
1002 orbf = (c->wbuf_ofs - ofs); /* offset in read buffer */
1003 if (orbf > len) /* is write beyond write buffer ? */
1005 lwbf = len - orbf; /* number of bytes to copy */
1006 if (lwbf > c->wbuf_len)
1010 memcpy(buf+orbf,c->wbuf+owbf,lwbf);
1013 up_read(&c->wbuf_sem);
1017 #define NR_OOB_SCAN_PAGES 4
1019 /* For historical reasons we use only 8 bytes for OOB clean marker */
1020 #define OOB_CM_SIZE 8
1022 static const struct jffs2_unknown_node oob_cleanmarker =
1024 .magic = constant_cpu_to_je16(JFFS2_MAGIC_BITMASK),
1025 .nodetype = constant_cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
1026 .totlen = constant_cpu_to_je32(8)
1030 * Check, if the out of band area is empty. This function knows about the clean
1031 * marker and if it is present in OOB, treats the OOB as empty anyway.
1033 int jffs2_check_oob_empty(struct jffs2_sb_info *c,
1034 struct jffs2_eraseblock *jeb, int mode)
1037 int cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1038 struct mtd_oob_ops ops;
1040 ops.mode = MTD_OPS_AUTO_OOB;
1041 ops.ooblen = NR_OOB_SCAN_PAGES * c->oobavail;
1042 ops.oobbuf = c->oobbuf;
1043 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1046 ret = mtd_read_oob(c->mtd, jeb->offset, &ops);
1047 if ((ret && !mtd_is_bitflip(ret)) || ops.oobretlen != ops.ooblen) {
1048 pr_err("cannot read OOB for EB at %08x, requested %zd bytes, read %zd bytes, error %d\n",
1049 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1050 if (!ret || mtd_is_bitflip(ret))
1055 for(i = 0; i < ops.ooblen; i++) {
1056 if (mode && i < cmlen)
1057 /* Yeah, we know about the cleanmarker */
1060 if (ops.oobbuf[i] != 0xFF) {
1061 jffs2_dbg(2, "Found %02x at %x in OOB for "
1062 "%08x\n", ops.oobbuf[i], i, jeb->offset);
1071 * Check for a valid cleanmarker.
1072 * Returns: 0 if a valid cleanmarker was found
1073 * 1 if no cleanmarker was found
1074 * negative error code if an error occurred
1076 int jffs2_check_nand_cleanmarker(struct jffs2_sb_info *c,
1077 struct jffs2_eraseblock *jeb)
1079 struct mtd_oob_ops ops;
1080 int ret, cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1082 ops.mode = MTD_OPS_AUTO_OOB;
1084 ops.oobbuf = c->oobbuf;
1085 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1088 ret = mtd_read_oob(c->mtd, jeb->offset, &ops);
1089 if ((ret && !mtd_is_bitflip(ret)) || ops.oobretlen != ops.ooblen) {
1090 pr_err("cannot read OOB for EB at %08x, requested %zd bytes, read %zd bytes, error %d\n",
1091 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1092 if (!ret || mtd_is_bitflip(ret))
1097 return !!memcmp(&oob_cleanmarker, c->oobbuf, cmlen);
1100 int jffs2_write_nand_cleanmarker(struct jffs2_sb_info *c,
1101 struct jffs2_eraseblock *jeb)
1104 struct mtd_oob_ops ops;
1105 int cmlen = min_t(int, c->oobavail, OOB_CM_SIZE);
1107 ops.mode = MTD_OPS_AUTO_OOB;
1109 ops.oobbuf = (uint8_t *)&oob_cleanmarker;
1110 ops.len = ops.ooboffs = ops.retlen = ops.oobretlen = 0;
1113 ret = mtd_write_oob(c->mtd, jeb->offset, &ops);
1114 if (ret || ops.oobretlen != ops.ooblen) {
1115 pr_err("cannot write OOB for EB at %08x, requested %zd bytes, read %zd bytes, error %d\n",
1116 jeb->offset, ops.ooblen, ops.oobretlen, ret);
1126 * On NAND we try to mark this block bad. If the block was erased more
1127 * than MAX_ERASE_FAILURES we mark it finally bad.
1128 * Don't care about failures. This block remains on the erase-pending
1129 * or badblock list as long as nobody manipulates the flash with
1130 * a bootloader or something like that.
1133 int jffs2_write_nand_badblock(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
1137 /* if the count is < max, we try to write the counter to the 2nd page oob area */
1138 if( ++jeb->bad_count < MAX_ERASE_FAILURES)
1141 pr_warn("marking eraseblock at %08x as bad\n", bad_offset);
1142 ret = mtd_block_markbad(c->mtd, bad_offset);
1145 jffs2_dbg(1, "%s(): Write failed for block at %08x: error %d\n",
1146 __func__, jeb->offset, ret);
1152 static struct jffs2_sb_info *work_to_sb(struct work_struct *work)
1154 struct delayed_work *dwork;
1156 dwork = to_delayed_work(work);
1157 return container_of(dwork, struct jffs2_sb_info, wbuf_dwork);
1160 static void delayed_wbuf_sync(struct work_struct *work)
1162 struct jffs2_sb_info *c = work_to_sb(work);
1163 struct super_block *sb = OFNI_BS_2SFFJ(c);
1165 if (!sb_rdonly(sb)) {
1166 jffs2_dbg(1, "%s()\n", __func__);
1167 jffs2_flush_wbuf_gc(c, 0);
1171 void jffs2_dirty_trigger(struct jffs2_sb_info *c)
1173 struct super_block *sb = OFNI_BS_2SFFJ(c);
1174 unsigned long delay;
1179 delay = msecs_to_jiffies(dirty_writeback_interval * 10);
1180 if (queue_delayed_work(system_long_wq, &c->wbuf_dwork, delay))
1181 jffs2_dbg(1, "%s()\n", __func__);
1184 int jffs2_nand_flash_setup(struct jffs2_sb_info *c)
1186 if (!c->mtd->oobsize)
1189 /* Cleanmarker is out-of-band, so inline size zero */
1190 c->cleanmarker_size = 0;
1192 if (c->mtd->oobavail == 0) {
1193 pr_err("inconsistent device description\n");
1197 jffs2_dbg(1, "using OOB on NAND\n");
1199 c->oobavail = c->mtd->oobavail;
1201 /* Initialise write buffer */
1202 init_rwsem(&c->wbuf_sem);
1203 INIT_DELAYED_WORK(&c->wbuf_dwork, delayed_wbuf_sync);
1204 c->wbuf_pagesize = c->mtd->writesize;
1205 c->wbuf_ofs = 0xFFFFFFFF;
1207 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1211 c->oobbuf = kmalloc_array(NR_OOB_SCAN_PAGES, c->oobavail, GFP_KERNEL);
1217 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1218 c->wbuf_verify = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1219 if (!c->wbuf_verify) {
1228 void jffs2_nand_flash_cleanup(struct jffs2_sb_info *c)
1230 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1231 kfree(c->wbuf_verify);
1237 int jffs2_dataflash_setup(struct jffs2_sb_info *c) {
1238 c->cleanmarker_size = 0; /* No cleanmarkers needed */
1240 /* Initialize write buffer */
1241 init_rwsem(&c->wbuf_sem);
1242 INIT_DELAYED_WORK(&c->wbuf_dwork, delayed_wbuf_sync);
1243 c->wbuf_pagesize = c->mtd->erasesize;
1245 /* Find a suitable c->sector_size
1246 * - Not too much sectors
1247 * - Sectors have to be at least 4 K + some bytes
1248 * - All known dataflashes have erase sizes of 528 or 1056
1249 * - we take at least 8 eraseblocks and want to have at least 8K size
1250 * - The concatenation should be a power of 2
1253 c->sector_size = 8 * c->mtd->erasesize;
1255 while (c->sector_size < 8192) {
1256 c->sector_size *= 2;
1259 /* It may be necessary to adjust the flash size */
1260 c->flash_size = c->mtd->size;
1262 if ((c->flash_size % c->sector_size) != 0) {
1263 c->flash_size = (c->flash_size / c->sector_size) * c->sector_size;
1264 pr_warn("flash size adjusted to %dKiB\n", c->flash_size);
1267 c->wbuf_ofs = 0xFFFFFFFF;
1268 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1272 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1273 c->wbuf_verify = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1274 if (!c->wbuf_verify) {
1280 pr_info("write-buffering enabled buffer (%d) erasesize (%d)\n",
1281 c->wbuf_pagesize, c->sector_size);
1286 void jffs2_dataflash_cleanup(struct jffs2_sb_info *c) {
1287 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1288 kfree(c->wbuf_verify);
1293 int jffs2_nor_wbuf_flash_setup(struct jffs2_sb_info *c) {
1294 /* Cleanmarker currently occupies whole programming regions,
1295 * either one or 2 for 8Byte STMicro flashes. */
1296 c->cleanmarker_size = max(16u, c->mtd->writesize);
1298 /* Initialize write buffer */
1299 init_rwsem(&c->wbuf_sem);
1300 INIT_DELAYED_WORK(&c->wbuf_dwork, delayed_wbuf_sync);
1302 c->wbuf_pagesize = c->mtd->writesize;
1303 c->wbuf_ofs = 0xFFFFFFFF;
1305 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1309 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1310 c->wbuf_verify = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1311 if (!c->wbuf_verify) {
1319 void jffs2_nor_wbuf_flash_cleanup(struct jffs2_sb_info *c) {
1320 #ifdef CONFIG_JFFS2_FS_WBUF_VERIFY
1321 kfree(c->wbuf_verify);
1326 int jffs2_ubivol_setup(struct jffs2_sb_info *c) {
1327 c->cleanmarker_size = 0;
1329 if (c->mtd->writesize == 1)
1330 /* We do not need write-buffer */
1333 init_rwsem(&c->wbuf_sem);
1334 INIT_DELAYED_WORK(&c->wbuf_dwork, delayed_wbuf_sync);
1336 c->wbuf_pagesize = c->mtd->writesize;
1337 c->wbuf_ofs = 0xFFFFFFFF;
1338 c->wbuf = kmalloc(c->wbuf_pagesize, GFP_KERNEL);
1342 pr_info("write-buffering enabled buffer (%d) erasesize (%d)\n",
1343 c->wbuf_pagesize, c->sector_size);
1348 void jffs2_ubivol_cleanup(struct jffs2_sb_info *c) {