From: Thomas Betker <thomas.bet...@rohde-schwarz.com>

jffs2_flash_direct_writev() always invokes jffs2_sum_add_kvec(), even
if mtd_writev() fails. Usually, this results in an extra summary entry
pointing to dirty node space, which should be ignored -- it is a bit of
a waste, but harmless.

When mtd_writev() returns *retlen == 0, though, the node space is not
reserved as dirty, but re-used; the extra summary entry then points
into the space of the next node. After the erase block has been closed,
we get the following messages on remount:

    jffs2: error: (79) jffs2_link_node_ref:
        Adding new ref c3048d18 at (0x00ec5b88-0x00ec6bcc)
        not immediately after previous (0x00ec5b88-0x00ec5b88)
    ...
    jffs2: Checked all inodes but still 0x2088 bytes of unchecked space?
    jffs2: No space for garbage collection. Aborting GC thread

The extra summary entries amount to "unchecked space", so that
jffs2_garbage_collect_pass() returns -ENOSPC. And without garbage
collection, the filesystem becomes unuseable over time as the erase
blocks fill up.

Fix this by skipping jffs2_sum_add_kvec() when the MTD write fails. We
don't need the summary entry anyway, and the behaviour matches that of
jffs2_flash_writev() in wbuf.c (with write buffering enabled).

Signed-off-by: Thomas Betker <thomas.bet...@rohde-schwarz.com>
---
 fs/jffs2/writev.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/fs/jffs2/writev.c b/fs/jffs2/writev.c
index a1bda9d..eec4197 100644
--- a/fs/jffs2/writev.c
+++ b/fs/jffs2/writev.c
@@ -16,9 +16,18 @@
 int jffs2_flash_direct_writev(struct jffs2_sb_info *c, const struct kvec *vecs,
                              unsigned long count, loff_t to, size_t *retlen)
 {
+       int ret;
+
+       ret = mtd_writev(c->mtd, vecs, count, to, retlen);
+
        if (!jffs2_is_writebuffered(c)) {
                if (jffs2_sum_active()) {
                        int res;
+
+                       if (ret ||
+                           *retlen != iov_length((struct iovec *) vecs, count))
+                               return ret;
+
                        res = jffs2_sum_add_kvec(c, vecs, count, (uint32_t) to);
                        if (res) {
                                return res;
@@ -26,19 +35,23 @@ int jffs2_flash_direct_writev(struct jffs2_sb_info *c, 
const struct kvec *vecs,
                }
        }
 
-       return mtd_writev(c->mtd, vecs, count, to, retlen);
+       return ret;
 }
 
 int jffs2_flash_direct_write(struct jffs2_sb_info *c, loff_t ofs, size_t len,
                        size_t *retlen, const u_char *buf)
 {
        int ret;
+
        ret = mtd_write(c->mtd, ofs, len, retlen, buf);
 
        if (jffs2_sum_active()) {
                struct kvec vecs[1];
                int res;
 
+               if (ret || *retlen != len)
+                       return ret;
+
                vecs[0].iov_base = (unsigned char *) buf;
                vecs[0].iov_len = len;
 
@@ -47,5 +60,6 @@ int jffs2_flash_direct_write(struct jffs2_sb_info *c, loff_t 
ofs, size_t len,
                        return res;
                }
        }
+
        return ret;
 }
-- 
2.6.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to