PATCH: allow JFFS2 to write to really small disks

PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

[-- Attachment #1: Type: text/plain, Size: 262 bytes --]

The attached patch allows JFFS2 to write to really small JFFS2 disks.

This is e.g. useful when using JFFS2 for in-system one time programming
of embedded device parameters: serial number, MAC, etc.


Comments?

-- 
Øyvind Harboe
http://www.zylin.com

[-- Attachment #2: jffs2rom.txt --]
[-- Type: text/x-patch, Size: 893 bytes --]

Index: build.c
===================================================================
RCS file: /home/cvs/mtd/fs/jffs2/build.c,v
retrieving revision 1.76
diff -u -r1.76 build.c
--- build.c    30 Jul 2005 15:29:27 -0000	1.76
+++ build.c    31 Aug 2005 09:01:34 -0000
@@ -279,6 +279,14 @@
 
	c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 
+	/* If the flash disk is smaller than resv_blocks_write, then we 
+	   allow writing to the disk anyway. The flash disk is then most likely
+	      being used as write once - read many medimum, e.g. in-system configuration of 
+	         static parameters(e.g. MAC address of device in stored in "ROM"). */
+		 if (c->resv_blocks_write * c->sector_size > c->flash_size) {
+		   c->resv_blocks_write = 0; 
+		   }
+
	/* When do we let the GC thread run in the background */
 
	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Wed, 31 August 2005 11:28:12 +0200, Øyvind Harboe wrote:
> 
> The attached patch allows JFFS2 to write to really small JFFS2 disks.
> 
> This is e.g. useful when using JFFS2 for in-system one time programming
> of embedded device parameters: serial number, MAC, etc.
> 
> Comments?

> Index: build.c
> ===================================================================
> RCS file: /home/cvs/mtd/fs/jffs2/build.c,v
> retrieving revision 1.76
> diff -u -r1.76 build.c
> --- build.c    30 Jul 2005 15:29:27 -0000	1.76
> +++ build.c    31 Aug 2005 09:01:34 -0000
> @@ -279,6 +279,14 @@
>  
> 	c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
>  
> +	/* If the flash disk is smaller than resv_blocks_write, then we 
> +	   allow writing to the disk anyway. The flash disk is then most likely
> +	      being used as write once - read many medimum, e.g. in-system configuration of 
> +	         static parameters(e.g. MAC address of device in stored in "ROM"). */
> +		 if (c->resv_blocks_write * c->sector_size > c->flash_size) {
> +		   c->resv_blocks_write = 0; 
> +		   }
> +
> 	/* When do we let the GC thread run in the background */
>  
> 	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;

Minor stuff: You indentation is broken.

If you changed the patch to add a mount option for the WORM case and
check for the mount option instead, this might go in.  Looks like it
could be useful.

Jörn

-- 
Courage is not the absence of fear, but rather the judgement that
something else is more important than fear.
-- Ambrose Redmoon

Re: PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

> > 	/* When do we let the GC thread run in the background */
> >  
> > 	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
> 
> Minor stuff: You indentation is broken.

Will fix of course. What emacs mode should I use for indentation? :-)

> If you changed the patch to add a mount option for the WORM case and
> check for the mount option instead, this might go in.  Looks like it
> could be useful.

I prefer it the way it is now. There are less options to read up on and
document.

W.r.t. documentation it would possibly call for a footnote that the
JFFS2 disk will only support read/write many times if it is large
enough. If it is too small, it only supports Write Once Read Many. 

What do you think?


-- 
Øyvind Harboe
http://www.zylin.com

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Thu, 1 September 2005 13:12:56 +0200, Øyvind Harboe wrote:
> 
> > > 	/* When do we let the GC thread run in the background */
> > >  
> > > 	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
> > 
> > Minor stuff: You indentation is broken.
> 
> Will fix of course. What emacs mode should I use for indentation? :-)

No idea, I use vim.  Try to mimic existing code, use scripts/Lindent
or read Documentation/CodingStyle.

> > If you changed the patch to add a mount option for the WORM case and
> > check for the mount option instead, this might go in.  Looks like it
> > could be useful.
> 
> I prefer it the way it is now. There are less options to read up on and
> document.
> 
> W.r.t. documentation it would possibly call for a footnote that the
> JFFS2 disk will only support read/write many times if it is large
> enough. If it is too small, it only supports Write Once Read Many. 
> 
> What do you think?

I think that most users don't want JFFS2 to run in WORM mode.  And for
those users, failing on every write is actually better.  They notice
the problem right away, not just after a while.

People that want WORM mode are the minority and therefore should be
the ones that are inconvenienced.

Jörn

-- 
The only real mistake is the one from which we learn nothing.
-- John Powell

Re: PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

> > What do you think?
> 
> I think that most users don't want JFFS2 to run in WORM mode.  And for
> those users, failing on every write is actually better.  They notice
> the problem right away, not just after a while.

But who mounts a *tiny* JFFS2 volume except those that will be using it
as a WORM drive?

This group of users could well be even tinier than those that want WORM
drives. 

My conclusion:

Make it a WORM drive if it is too small to support read/write many
times.

> People that want WORM mode are the minority and therefore should be
> the ones that are inconvenienced.

I agree with this principle, but my assumptions are different than
yours, hence I'm currently leaning towards a different conclusion.


-- 
Øyvind Harboe
http://www.zylin.com

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Thu, 1 September 2005 13:52:35 +0200, Øyvind Harboe wrote:
> 
> > > What do you think?
> > 
> > I think that most users don't want JFFS2 to run in WORM mode.  And for
> > those users, failing on every write is actually better.  They notice
> > the problem right away, not just after a while.
> 
> But who mounts a *tiny* JFFS2 volume except those that will be using it
> as a WORM drive?

People making mistakes.  It is a very important usability issue to
mark mistakes as such.  Making it appear as if everything is ok and
having it fail later, possibly days or weeks later, is bad usability.

I would agree with you if the user-perceived behaviour was
*immediatly* and *noticably* different.  But unless people write
enough data to fill their fs, the difference may go unnoticed.

Jörn

-- 
The wise man seeks everything in himself; the ignorant man tries to get
everything from somebody else.
-- unknown

Re: PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

On Thu, 2005-09-01 at 14:00 +0200, Jörn Engel wrote:

> > But who mounts a *tiny* JFFS2 volume except those that will be using it
> > as a WORM drive?
> 
> People making mistakes.  It is a very important usability issue to
> mark mistakes as such.  Making it appear as if everything is ok and
> having it fail later, possibly days or weeks later, is bad usability.
> 
> I would agree with you if the user-perceived behaviour was
> *immediatly* and *noticably* different.  But unless people write
> enough data to fill their fs, the difference may go unnoticed.

Sounds good.

If I get some indication that there is a consensus from
maintainers/committers to add this mount option, I'll go ahead and write
it up.


-- 
Øyvind Harboe
http://www.zylin.com

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Thu, 1 September 2005 14:45:56 +0200, Øyvind Harboe wrote:
> On Thu, 2005-09-01 at 14:00 +0200, Jörn Engel wrote:
> 
> > > But who mounts a *tiny* JFFS2 volume except those that will be using it
> > > as a WORM drive?
> > 
> > People making mistakes.  It is a very important usability issue to
> > mark mistakes as such.  Making it appear as if everything is ok and
> > having it fail later, possibly days or weeks later, is bad usability.
> > 
> > I would agree with you if the user-perceived behaviour was
> > *immediatly* and *noticably* different.  But unless people write
> > enough data to fill their fs, the difference may go unnoticed.
> 
> Sounds good.
> 
> If I get some indication that there is a consensus from
> maintainers/committers to add this mount option, I'll go ahead and write
> it up.

You're asking for too much. :)

Most of the time, most people just don't care.  If they actually do
care and are opposed to your patch, they will speak up.  But you won't
ever find out before actively trying to merge the patch.

Jörn

-- 
...one more straw can't possibly matter...
-- Kirby Bakken

Re: PATCH: allow JFFS2 to write to really small disks

[Josh Boyer]

On Thu, 2005-09-01 at 14:57 +0200, Jörn Engel wrote:
> On Thu, 1 September 2005 14:45:56 +0200, Øyvind Harboe wrote:
> > On Thu, 2005-09-01 at 14:00 +0200, Jörn Engel wrote:
> > 
> > > > But who mounts a *tiny* JFFS2 volume except those that will be using it
> > > > as a WORM drive?
> > > 
> > > People making mistakes.  It is a very important usability issue to
> > > mark mistakes as such.  Making it appear as if everything is ok and
> > > having it fail later, possibly days or weeks later, is bad usability.
> > > 
> > > I would agree with you if the user-perceived behaviour was
> > > *immediatly* and *noticably* different.  But unless people write
> > > enough data to fill their fs, the difference may go unnoticed.
> > 
> > Sounds good.
> > 
> > If I get some indication that there is a consensus from
> > maintainers/committers to add this mount option, I'll go ahead and write
> > it up.
> 
> You're asking for too much. :)
> 
> Most of the time, most people just don't care.  If they actually do
> care and are opposed to your patch, they will speak up.  But you won't
> ever find out before actively trying to merge the patch.

I agree with Jörn on this point.  It should be a mount option.

Not that my opinion amounts to much... :)

josh

Re: PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

> > If I get some indication that there is a consensus from
> > maintainers/committers to add this mount option, I'll go ahead and write
> > it up.
> 
> You're asking for too much. :)
> 
> Most of the time, most people just don't care.  If they actually do
> care and are opposed to your patch, they will speak up.  But you won't
> ever find out before actively trying to merge the patch.

I guess. I'll set aside some time to read up on the mount options for
JFFS2(incidentally I'm using JFFS2 for eCos) and put together a properly
indented & commented patch.

Thanks for the feedback btw, hopefully this will work out.




-- 
Øyvind Harboe
http://www.zylin.com

Re: PATCH: allow JFFS2 to write to really small disks

[Øyvind Harboe]

[-- Attachment #1: Type: text/plain, Size: 263 bytes --]

How about this patch?

It adds a mount option to allow Write Once Read Many if the disk is too
small to support read/write many times.

W.r.t. eCos, I expect the WORM capability to be enabled via a CDL
option.


-- 
Øyvind Harboe
http://www.zylin.com

[-- Attachment #2: jffs2worm.txt --]
[-- Type: text/x-patch, Size: 1879 bytes --]

Index: fs/jffs2/build.c
===================================================================
RCS file: /home/cvs/mtd/fs/jffs2/build.c,v
retrieving revision 1.77
diff -u -r1.77 build.c
--- fs/jffs2/build.c	31 Aug 2005 13:51:00 -0000	1.77
+++ fs/jffs2/build.c	2 Sep 2005 12:42:49 -0000
@@ -279,6 +279,18 @@
 
 	c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 
+	/* It is possible to mount a JFFS2 in a mode where it allows
+	   Write Once Read Many operation if the flash disk is too small
+	   to support read/write many times.
+	   
+	   If the flash disk is smaller than resv_blocks_write, then we 
+	   allow writing to the disk anyway. The flash disk is then most likely
+       being used as write once - read many medimum, e.g. in-system configuration of 
+	   static parameters(e.g. MAC address of device in stored in "ROM"). */
+	if ((c->flags & JFFS2_SB_FLAG_ALLOW_WORM)!=0)&& (c->resv_blocks_write * c->sector_size > c->flash_size)) {
+		c->resv_blocks_write = 0; 
+	}
+
 	/* When do we let the GC thread run in the background */
 
 	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
Index: include/linux/jffs2_fs_sb.h
===================================================================
RCS file: /home/cvs/mtd/include/linux/jffs2_fs_sb.h,v
retrieving revision 1.52
diff -u -r1.52 jffs2_fs_sb.h
--- include/linux/jffs2_fs_sb.h	19 May 2005 16:12:17 -0000	1.52
+++ include/linux/jffs2_fs_sb.h	2 Sep 2005 12:42:49 -0000
@@ -16,6 +16,9 @@
 #define JFFS2_SB_FLAG_RO 1
 #define JFFS2_SB_FLAG_SCANNING 2 /* Flash scanning is in progress */
 #define JFFS2_SB_FLAG_BUILDING 4 /* File system building is in progress */
+#define JFFS2_SB_FLAG_ALLOW_WORM 8 /* allow write once read many if the
+                                      disk is too small to support read/write
+                                      many times*/
 
 struct jffs2_inodirty;
 

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Fri, 2 September 2005 14:44:47 +0200, Øyvind Harboe wrote:
>
> To: linux-mtd@lists.infradead.org

Please keep everyone in the thread on Cc:.  Being subscribed to a
dozen mailing lists causes me to miss some things otherwise.

> How about this patch?
> 
> It adds a mount option to allow Write Once Read Many if the disk is too
> small to support read/write many times.
> 
> W.r.t. eCos, I expect the WORM capability to be enabled via a CDL
> option.

I have to admit being clueless about eCos.  For Linux, the parser bits
are still missing.  How about this patch?  The parser stuff is quite
large, as this appears to be the first JFFS2 mount option ever.

Jörn

-- 
The cheapest, fastest and most reliable components of a computer
system are those that aren't there.
-- Gordon Bell, DEC labratories


Signed-off-by: Jörn Engel <joern@wohnheim.fh-wedel.de>
---

 fs/jffs2/build.c            |    6 ++++++
 fs/jffs2/fs.c               |   41 +++++++++++++++++++++++++++++++++++++++++
 include/linux/jffs2_fs_sb.h |    7 ++++---
 3 files changed, 51 insertions(+), 3 deletions(-)

--- linux-2.6.13-rc6mtd/fs/jffs2/build.c~jffs2_worm	2005-08-26 17:18:04.000000000 +0200
+++ linux-2.6.13-rc6mtd/fs/jffs2/build.c	2005-09-04 10:29:39.000000000 +0200
@@ -279,6 +279,12 @@ static void jffs2_calc_trigger_levels(st
 
 	c->resv_blocks_write = c->resv_blocks_deletion + (size / c->sector_size);
 
+	/* Allow JFFS2 to run in WORM (write once read many) mode.
+	   This mode makes sense for small flashes with very rare
+	   updates of small amounts of data. */
+	if ((c->flags & JFFS2_SB_FLAG_WORM) && (c->nr_blocks < c->resv_blocks_write))
+		c->resv_blocks_write = 0; 
+
 	/* When do we let the GC thread run in the background */
 
 	c->resv_blocks_gctrigger = c->resv_blocks_write + 1;
--- linux-2.6.13-rc6mtd/fs/jffs2/fs.c~jffs2_worm	2005-09-04 10:15:36.000000000 +0200
+++ linux-2.6.13-rc6mtd/fs/jffs2/fs.c	2005-09-04 10:57:33.000000000 +0200
@@ -22,6 +22,7 @@
 #include <linux/vmalloc.h>
 #include <linux/vfs.h>
 #include <linux/crc32.h>
+#include <linux/parser.h>
 #include "nodelist.h"
 
 static int jffs2_flash_setup(struct jffs2_sb_info *c);
@@ -440,6 +441,44 @@ struct inode *jffs2_new_inode (struct in
 }
 
 
+enum {
+	opt_worm,
+};
+
+
+static match_table_t tokens = {
+	{opt_worm,	"worm"},
+};
+
+
+static int jffs2_parse_options(struct jffs2_sb_info *c, void *_data)
+{
+	char *data = _data;
+	char *p;
+	substring_t args[MAX_OPT_ARGS];
+
+	if (!data)
+		return 0;
+
+	while ((p = strsep(&data, ","))) {
+		int token;
+
+		if (!*p)
+			continue;
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case opt_worm:
+			c->flags |= JFFS2_SB_FLAG_WORM;
+			break;
+		default:
+			printk(KERN_ERR"jffs2: unknown mount flag \"%s\"\n", p);
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+
 int jffs2_do_fill_super(struct super_block *sb, void *data, int silent)
 {
 	struct jffs2_sb_info *c;
@@ -459,6 +498,8 @@ int jffs2_do_fill_super(struct super_blo
 		return -EINVAL;
 	}
 #endif
+	if (jffs2_parse_options(c, data))
+		return -EINVAL;
 
 	c->flash_size = c->mtd->size;
 
--- linux-2.6.13-rc6mtd/include/linux/jffs2_fs_sb.h~jffs2_worm	2005-08-17 13:37:29.000000000 +0200
+++ linux-2.6.13-rc6mtd/include/linux/jffs2_fs_sb.h	2005-09-04 10:55:07.000000000 +0200
@@ -13,9 +13,10 @@
 #include <linux/list.h>
 #include <linux/rwsem.h>
 
-#define JFFS2_SB_FLAG_RO 1
-#define JFFS2_SB_FLAG_SCANNING 2 /* Flash scanning is in progress */
-#define JFFS2_SB_FLAG_BUILDING 4 /* File system building is in progress */
+#define JFFS2_SB_FLAG_RO	1
+#define JFFS2_SB_FLAG_SCANNING	2 /* Flash scanning is in progress */
+#define JFFS2_SB_FLAG_BUILDING	4 /* File system building is in progress */
+#define JFFS2_SB_FLAG_WORM	8 /* Write once read many - writes but no GC */
 
 struct jffs2_inodirty;
 

Re: PATCH: allow JFFS2 to write to really small disks

[oyvind.harboe]

>> W.r.t. eCos, I expect the WORM capability to be enabled via a CDL
>> option.
>
> I have to admit being clueless about eCos.  For Linux, the parser bits
> are still missing.  How about this patch?  The parser stuff is quite
> large, as this appears to be the first JFFS2 mount option ever.

Do you think this is useful for Linux?

I mostly foresaw it as being useful for deeply embedded systems, like eCos.

Øyvind

Re: PATCH: allow JFFS2 to write to really small disks

[Jörn Engel]

On Mon, 5 September 2005 12:05:11 +0200, oyvind.harboe@zylin.com wrote:
> 
> >>W.r.t. eCos, I expect the WORM capability to be enabled via a CDL
> >>option.
> >
> >I have to admit being clueless about eCos.  For Linux, the parser bits
> >are still missing.  How about this patch?  The parser stuff is quite
> >large, as this appears to be the first JFFS2 mount option ever.
> 
> Do you think this is useful for Linux?
> 
> I mostly foresaw it as being useful for deeply embedded systems, like eCos.

I don't know.  So far you've been the only person to ask for this.  If
that doesn't change, maybe it would be best if you maintained a
minimal patch and we would just forget about this.  But if the patch
is useful for others as well, we could merge it.

Anyone?

Jörn

-- 
A quarrel is quickly settled when deserted by one party; there is
no battle unless there be two.
-- Seneca