[lkml]   [2002]   [Jan]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
Subject[PATCH] ReiserFS 2.5 Update Patch Set 16 of 25

This set of patches of which this is one will update ReiserFS in 2.5
to contain all bugfixes applied to 2.4 plus allow relocating the journal plus
uuid support plus fix the kdev_t compilation failure.

Fixes a bug when mmap-write to a file tail and subsequent read cause written
data to be lost due to page-cache interacting mistake in low number of free
buffers situation.

The other patches in this set are:

kdev_t fixes to comply with new interface.

Support for relocated journals.

It is possible that knfsd is trying to access inode of a file
that is being removed from the disk by some other thread. As we
update sd on unlink all that is required is to check for nlink
here. This bug was first found by Sizif when debugging
SquidNG/Butterfly, forgotten, and found again after Philippe
Gramoulle <> reproduced it.

More logical fix would require changes in fs/inode.c:iput() to
remove inode from hash-table _after_ fs cleaned disk stuff up and
in iget() to return NULL if I_FREEING inode is found in
hash-table. We await Al Viro doing the more logical fix, and we
provide this fix so that users can work while we wait for the
better fix.

Check that block number are going to free in a bitmap makes sense.
This avoids oops after trying to access bitmap for wild block number.

Patch by Chris Mason <Mason@Suse.COM>.
prepare_for_delete_or_cut() tries to find the unformatted node in
the buffer cache to make sure it isn't in use. Since unformatted
nodes are never in the buffer cache, this check is useless. The
page locking done by mm/vmscan.c:vmtruncate protects us from
truncating away pages that are in use, so it is safe to just remove
the bogus check from our code.

Since the get_hash_table was also the reason for the repeat loop,
this patch removes it as well.

This should make file deletes faster, at the very least it cuts down
on CPU overhead for deletes/truncates.

There is always place for Yet Another Cleanup of Reiserfs Code.

fixes a bug first noticed using a Freebsd nfs testing tool. When writing to
a previously mmaped-filled hole in file, and then writing with write() there
again, page that write() hits loses mmap-written content.

Fixes long-standing problem in reiserfs, when disk space gets leaked
if crash occurred when some process hold a reference to unlinked file.

It's possible to unlink file that is still opened by some
process. In this case, body of file is actually removed at the time
of last close. If crash occurs in between last unlink (when
directory entry for this file is removed) and last close, body
doesn't get unlinked and "disk-space-leak" occurs. To prevent this,
unlink-truncate-opened patch stores in the tree a special record at the
time of last unlink. This record is a form of logical logging and
will be either removed during following close, or replayed during
next mount after a crash.

Reiserfs 3.5 disk format can only store 16 bit uid/gid inside
stat-data. This patch adds error checking so that EINVAL is returned
on attempt to change uid/gid of an old file to value that doesn't
fit into 16 bit, in stead of silently truncating it into 16 bit.

Patch by Chris Mason for bug found and debugged by Anne Milicia
( don't run preallocated blocks
through journal_mark_freed() and don't corrupt i_prealloc_block during

Patch by Chris Mason to avoid duplicate replay of last flushed

Patch to break infinite loop in journal_read() in the case when the
journal log area is completely filled with transactions.

Fixes a problem with v3.6 fs mounted readonly and then remounted rw.

Fixes erroroneous page access before making sure page is really accessable.
Bug can be triggered only on highmem sysetms.

Symlink-body length check was made against an incorrect value, allowing for
too long nodes to be inserted into tree. This might lead to obscure
warnings in some cases.

Fixes a bug when mmap-write to a file tail and subsequent read cause written
data to be lost due to page-cache interacting mistake in low number of free
buffers situation.

Correct a typo in fs/reiserfs/journal.c:
interruptible_sleep_on_timeout() takes timeout in jiffies, rather
than seconds.

Certain disk corruptions and i/o errors may cause lookup() to panic, which
is wrong.

Suppress compilation warnings on big endian platform.

This patch fixes 2 bugs in reiserfs_rename(). First one being attempt to
access item before verifying it was not moved since last access. Second
is a window, where old filename may be written to disk with 'visible'
flag unset without these changes be journaled.

reiserfs_inode_cache seems to be too long. converting it to

This patch makes sure that indirect pointers for holes are correctly filled
in by zeroes at
hole-creation time. (Author is Chris Mason. fs/buffer.c
(generic_cont_expand) were written by Alexander Viro)

Somebody introduced a bug in reiserfs_release_file() leading to corrupting
journal for ro filesystems.

Do not print unsuccesful superblocks read warnings
(if old or new one cannot be found). Print verbose journal info.
Convert warnings to standard format.

Fixes a case where v3.6 filesystem can get wrong magic after converting
from v3.5 one.

--- linux-2.5.3-pre3/fs/reiserfs/inode.c.orig Wed Jan 23 20:19:07 2002
+++ linux-2.5.3-pre3/fs/reiserfs/inode.c Wed Jan 23 20:22:27 2002
@@ -321,6 +321,16 @@
if (buffer_uptodate(bh_result)) {
goto finished ;
+ } else
+ /*
+ ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
+ ** pages without any buffers. If the page is up to date, we don't want
+ ** read old data off disk. Set the up to date bit on the buffer instead
+ ** and jump to the end
+ */
+ if (Page_Uptodate(bh_result->b_page)) {
+ mark_buffer_uptodate(bh_result, 1);
+ goto finished ;

// read file tail into part of page
@@ -827,7 +837,7 @@
if (retval == POSITION_FOUND) {
reiserfs_warning ("vs-825: reiserfs_get_block: "
- "%k should not be found\n", &key);
+ "%K should not be found\n", &key);
retval = -EEXIST;
if (allocated_block_nr)
reiserfs_free_block (&th, allocated_block_nr);
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:18    [W:0.046 / U:5.816 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site