Skip to contentSkip to conversation

reiserfs patch for linux-2.4.2

PATCH In review Alexander Zarochentcev 5 participants 9 messages Feb 28, 2001 – Mar 2, 2001

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 reiserfs patch for linux-2.4.2

Hello !

This reiserfs patch for linux-2.4.2 was prepared by all members of the reiserfs dev team. It contains the following fixes/cleanups/improvements:

1. Fix for possible preallocated blocks leakage after a system crash

Reiserfs does block preallocation for big files like ext2 does. Preallocation code changes both the bitmap and the free blocks count in the superblock. It is possible to have unattached disk blocks which do not belong to any file (i.e. free space leak) after a crash. Unlike ext2, reiserfs does not check for consistency of a filesystem that was not cleanly unmounted. There is no simple way to recover free space.

For a given transaction, the fix is to, at the transaction end, discard the preallocations for those inodes which have one.

2. Object id sharing problem.

Fix for reiserfs/inode.c:reiserfs_iget () It now makes sure that it really got inode of an object it is looking for.

3. Improvement of error handling in reiserfs_add_entry()

There is a place in reiserfs_add_entry() code where fs corruption (existence of hidden entries--objects temporarily used in renaming), I/O errors, or several other occurences, can produce an error message which has no useful information.

We have had several bug reports about this vs-7032 message in the older 3.5 version. Each time we were unable to reproduce it. We do not understand the reason. This error handling fix will make such bug reports much more informative for us if they should occur in the newer version.

4. In reiserfs_add_entry(), we don't need to scan a whole bit_string for non-zero bits to be able to use a smallest generation counter to avoid researching for the dir entry insertion point.

5. bad block support

New ioctls added, enabling direct manipulation of block allocation bitmap by the user space programs: ioctl to mark block as used, ioctl to mark block as free and ioctl to update counter (stored in the super block) of blocks marked through first two ioctls. This can be used as crude form of bad block support (e.g., such ioctls can be used just after creation of file system on the device with known bad blocks). User level interface to ioctls available.

6. Using integer constants from limits.h instead of self made ones

7. other minor fixes.

Regards,
Alex.

#1 lore raw
Quoted 1 line · Alexander Zarochentcev

6. Using integer constants from limits.h instead of self made ones

Urgg. limits.h is a userlevel header...

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

Christoph

Signature · 15 lines
Of course it doesn't work. We've performed a software upgrade.


--- linux-2.4.0/Makefile	Mon Dec 25 19:21:14 2000
+++ linux/Makefile	Mon Dec 25 23:30:03 2000
@@ -85,7 +85,8 @@
 # standard CFLAGS
 #
 
-CPPFLAGS := -D__KERNEL__ -I$(HPATH)
+GCCINCDIR = $(shell gcc -print-search-dirs | sed -ne 's/install: \(.*\)/\1include/gp')
+CPPFLAGS := -D__KERNEL__ -nostdinc -I$(HPATH) -I$(GCCINCDIR)
 
 CFLAGS := $(CPPFLAGS) -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -fno-strict-aliasing
 AFLAGS := -D__ASSEMBLY__ $(CPPFLAGS)
#2 lore raw

Review

Quoted 9 lines · Christoph Hellwig

Urgg. limits.h is a userlevel header...

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

--- linux-2.4.0/Makefile Mon Dec 25 19:21:14 2000

(quoted diff) +2 -1
-CPPFLAGS := -D__KERNEL__ -I$(HPATH)
+GCCINCDIR = $(shell gcc -print-search-dirs | sed -ne 's/install: \(.*\)/\1include/gp')
+CPPFLAGS := -D__KERNEL__ -nostdinc -I$(HPATH) -I$(GCCINCDIR)
cc trimmed to l-k.

CPPFLAGS apply to the kernel compiler which can be a cross compiler but you are extracting data from the host gcc. What exactly are you trying to do here? Are you trying to prevent the use of user space includes or are you trying to pick up the cross compiler includes?

#3 lore raw
In reply to Christoph Hellwig
Quoted 14 lines · Keith Owens

Review

Quoted 9 lines · Christoph Hellwig

Urgg. limits.h is a userlevel header...

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

--- linux-2.4.0/Makefile Mon Dec 25 19:21:14 2000

(quoted diff) +2 -1
-CPPFLAGS := -D__KERNEL__ -I$(HPATH)
+GCCINCDIR = $(shell gcc -print-search-dirs | sed -ne 's/install: \(.*\)/\1include/gp')
+CPPFLAGS := -D__KERNEL__ -nostdinc -I$(HPATH) -I$(GCCINCDIR)
cc trimmed to l-k.
Quoted 3 lines

CPPFLAGS apply to the kernel compiler which can be a cross compiler but you are extracting data from the host gcc. What exactly are you trying to do here?

Prevent inclusion of the usual (/usr/include) headers.

Quoted 2 lines

Are you trying to prevent the use of user space includes or are you trying to pick up the cross compiler includes?

Ok, gcc should be replaced by $(CC).

Christoph

Signature · 1 line
Of course it doesn't work. We've performed a software upgrade.
#4 lore raw
In reply to Christoph Hellwig
Quoted 4 lines · Christoph Hellwig

Urgg. limits.h is a userlevel header...

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

There are very few files needed from gcc's include dir. Linux ought to be able to survive without them. Linux is already gcc-specific anyway.

#5 lore raw
In reply to Christoph Hellwig
Quoted 9 lines · Albert D. Cahalan
Quoted 4 lines · Christoph Hellwig

Urgg. limits.h is a userlevel header...

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

There are very few files needed from gcc's include dir. Linux ought to be able to survive without them. Linux is already gcc-specific anyway.

I think we want stdarg.h from gcc...

Christoph

Signature · 1 line
Of course it doesn't work. We've performed a software upgrade.
#6 lore raw
Quoted 5 lines · Christoph Hellwig
Quoted 4 lines · Albert D. Cahalan
Quoted 2 lines · Christoph Hellwig

Urgg. limits.h is a userlevel header...

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

The attached patch will make similar atempts fail (but not this one as there is also a limits.h in gcc's include dir).

There are very few files needed from gcc's include dir. Linux ought to be able to survive without them. Linux is already gcc-specific anyway.

I think we want stdarg.h from gcc...

Yes, just as apps want <linux/*.h> files.

The kernel can have a copy. If the stack frame layout changes enough to cause trouble with stdarg.h, then most likely there will be huge trouble in 42 other places.

If you insist on using whatever random stdarg.h might be on the system, then just copy it into the build area. The compile might even run a bit faster without the extra directory to search.

#7 lore raw
In reply to Alexander Zarochentcev
Quoted 1 line · Alexander Zarochentcev

6. Using integer constants from limits.h instead of self made ones

That's a userland header file. Don't use it in the kernel.

Quoted 1 line

7. other minor fixes.

Does this patch contain Chris Mason's "tail conversion" fix that he made after my bug report?

Erik

Signature · 5 lines
J.A.K. (Erik) Mouw, Information and Communication Theory Group, Department
of Electrical Engineering, Faculty of Information Technology and Systems,
Delft University of Technology, PO BOX 5031,  2600 GA Delft, The Netherlands
Phone: +31-15-2783635  Fax: +31-15-2781843  Email: [email protected]
WWW: http://www-ict.its.tudelft.nl/~erik/
#8 lore raw
In reply to Alexander Zarochentcev
Quoted 9 lines · Erik Mouw
Quoted 1 line · Alexander Zarochentcev

6. Using integer constants from limits.h instead of self made ones

That's a userland header file. Don't use it in the kernel.

Quoted 1 line

7. other minor fixes.

Does this patch contain Chris Mason's "tail conversion" fix that he made after my bug report?

No.
The "tail conversion" fix was sent to Alan.
It is already included into 2.4.2-ac6.

All fixes are in one combined patch:
ftp://ftp.namesys.com/pub/reiserfs-for-2.4/linux-2.4.2-reiserfs-20010301-full.patch.gz

Tail conversion fix is available as separate patch at namesys.com ftp site:
ftp://ftp.namesys.com/pub/reiserfs-for-2.4/another-null.diff .

Quoted 8 lines

Erik

Signature · 5 lines
J.A.K. (Erik) Mouw, Information and Communication Theory Group, Department
of Electrical Engineering, Faculty of Information Technology and Systems,
Delft University of Technology, PO BOX 5031,  2600 GA Delft, The Netherlands
Phone: +31-15-2783635  Fax: +31-15-2781843  Email: [email protected]
WWW: http://www-ict.its.tudelft.nl/~erik/

Thanks,
Alex.