[PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

Hello,

I don't know if this patch qualifies for inclusion into the mainline kernel, but
maybe someone else would be interested in such, so I'm posting it here.

The main idea for that device was to have a source for 0xff to clear FLASH based
devices. And to make it more generic, I've added the possibility to change the
default value on a per file descriptor basis.

I don't want to discuss if such a device makes sense or if such should be done
in userspace only. I like it as a device, otherwise I wouldn't have written it.
So besides this message, I will not take part in such a discussion.

Regards,

Alexander Holler

[PATCH 1/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

This device outputs by default 0xff instead 0 which makes more sense
than 0 to clear e.g. FLASH based devices.

To make the device more general usable, the value it outputs is changeable
on a per file descriptor basis through simple writes to it.
Values can be decimal (0 - 255), octal (00 - 0377) or hex (0x0 - 0xff).
For other values (or strings) written to it, the write operation returns an
error and the subsequent output is undefined.

Usage examples:

 # Fill /dev/sdX with 0xff.
dd </dev/byte of=/dev/sdX bs=4M

 # Create a file of size 10GB and filled with 0xaa.
exec 5<>/dev/byte # Open /dev/byte and assign fd 5 to it
echo 0xaa >&5     # Instruct the device to output 0xaa
dd <&5 of=foo.img bs=10M count=1024 # create and fill the file
exec 5>&-         # Close fd 5

Signed-off-by: Alexander Holler <holler@ahsoftware.de>
---
 Documentation/byte.txt    |   24 +++++++++++
 Documentation/devices.txt |    1 +
 drivers/char/Kconfig      |   10 ++++
 drivers/char/mem.c        |  101 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 136 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/byte.txt

diff --git a/Documentation/byte.txt b/Documentation/byte.txt
new file mode 100644
index 0000000..c132e7d
--- /dev/null
+++ b/Documentation/byte.txt
@@ -0,0 +1,24 @@
+/dev/byte is a generic byte source (similiar to /dev/zero) which uses
+0xff as the default value.
+
+The idea was to have a source for 0xff to erase FLASH based devices like
+SSDs and SD-Cards for which 0xff makes more sense than 0.
+
+To make the device more general usable, the value it outputs is changeable
+on a per file descriptor basis through simple writes to it.
+Values can be decimal (0 - 255), octal (00 - 0377) or hex (0x0 - 0xff).
+For other values (or strings) written to it, the write operation returns an
+error and the subsequent output is undefined.
+
+Usage examples:
+
+# Fill /dev/sdX with 0xff.
+dd </dev/byte of=/dev/sdX bs=4M
+
+# Create a file of size 10GB and filled with 0xaa.
+exec 5<>/dev/byte # Open /dev/byte and assign fd 5 to it
+echo 0xaa >&5     # Instruct the device to output 0xaa
+dd <&5 of=foo.img bs=10M count=1024 # create and fill the file
+exec 5>&-         # Close fd 5
+
+Apr 2011, Alexander Holler <holler@ahsoftware.de>
diff --git a/Documentation/devices.txt b/Documentation/devices.txt
index eccffe7..b34439e 100644
--- a/Documentation/devices.txt
+++ b/Documentation/devices.txt
@@ -101,6 +101,7 @@ Your cooperation is appreciated.
 		 11 = /dev/kmsg		Writes to this come out as printk's
 		 12 = /dev/oldmem	Used by crashdump kernels to access
 					the memory of the kernel that crashed.
+		 13 = /dev/byte		Generic byte source (default 0xff).
 
   1 block	RAM disk
 		  0 = /dev/ram0		First RAM disk
diff --git a/drivers/char/Kconfig b/drivers/char/Kconfig
index b7980a83..c745bcf 100644
--- a/drivers/char/Kconfig
+++ b/drivers/char/Kconfig
@@ -97,6 +97,16 @@ config DEVKMEM
 	  kind of kernel debugging operations.
 	  When in doubt, say "N".
 
+config DEV_BYTE
+	bool "/dev/byte virtual device support"
+	default y
+	help
+	  Say Y here if you want to support the /dev/byte device. The
+	  /dev/byte device is a generic byte source (similiar to /dev/zero),
+	  but uses 0xff as a default value which is changeable on a per
+	  file descriptor basis.
+	  Read <file:Documentation/byte.txt> for a usage description.
+
 config BFIN_JTAG_COMM
 	tristate "Blackfin JTAG Communication"
 	depends on BLACKFIN
diff --git a/drivers/char/mem.c b/drivers/char/mem.c
index 1256454..7a1d558 100644
--- a/drivers/char/mem.c
+++ b/drivers/char/mem.c
@@ -6,6 +6,7 @@
  *  Added devfs support.
  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
+ *  Added /dev/byte support, Apr 2011, Alexander Holler <holler@ahsoftware.de>
  */
 
 #include <linux/mm.h>
@@ -26,6 +27,7 @@
 #include <linux/bootmem.h>
 #include <linux/splice.h>
 #include <linux/pfn.h>
+#include <linux/radix-tree.h>
 
 #include <asm/uaccess.h>
 #include <asm/io.h>
@@ -736,6 +738,7 @@ static int open_port(struct inode * inode, struct file * filp)
 
 #define zero_lseek	null_lseek
 #define full_lseek      null_lseek
+#define byte_lseek	null_lseek
 #define write_zero	write_null
 #define read_full       read_zero
 #define open_mem	open_port
@@ -835,6 +838,101 @@ static const struct file_operations kmsg_fops = {
 	.llseek = noop_llseek,
 };
 
+#ifdef CONFIG_DEV_BYTE
+static u16 byte_items[1] = {}; /* must be aligned even */
+static RADIX_TREE(byte_tree, GFP_KERNEL);
+static spinlock_t byte_lock;
+
+/*
+ * Read a value the device should output afterwards.
+ * Values can be decimal (0 - 255), octal (00 - 0377) or hex (0x0 - 0xff).
+ */
+static ssize_t write_byte(struct file *filp, const char __user *buffer,
+		   size_t count, loff_t *off)
+{
+	u16 *val;
+	unsigned long num;
+	char buff[5];
+
+	num = (count < sizeof(buff)-1) ? count : sizeof(buff)-1;
+	if (copy_from_user(buff, buffer, num))
+		return -EFAULT;
+	buff[num] = '\0';
+	if (strict_strtoul(buff, 0, &num) || num > 255)
+		return -EFAULT;
+	rcu_read_lock();
+	val = radix_tree_lookup(&byte_tree, (unsigned long)filp);
+	rcu_read_unlock();
+	spin_lock(&byte_lock);
+	if (val != NULL)
+		radix_tree_delete(&byte_tree, (unsigned long)filp);
+	num = radix_tree_insert(&byte_tree, (unsigned long)filp,
+		byte_items+num);
+	spin_unlock(&byte_lock);
+	if (num)
+		return -EFAULT;
+	return count;
+}
+
+static ssize_t read_byte(struct file *file, char __user *buf,
+			 size_t count, loff_t *ppos)
+{
+	size_t written;
+	u16 *val;
+	unsigned char byte = 0xff;
+
+	if (!count)
+		return 0;
+
+	if (!access_ok(VERIFY_WRITE, buf, count))
+		return -EFAULT;
+
+	rcu_read_lock();
+	val = radix_tree_lookup(&byte_tree, (unsigned long)file);
+	rcu_read_unlock();
+	if (val != NULL)
+		byte = (unsigned char)(val-byte_items);
+
+	written = 0;
+	while (count) {
+		size_t chunk = count;
+
+		if (chunk > PAGE_SIZE)
+			chunk = PAGE_SIZE;	/* Just for latency reasons */
+		memset(buf, byte, chunk);
+		written += chunk;
+		if (signal_pending(current))
+			return written;
+		buf += chunk;
+		count -= chunk;
+		cond_resched();
+	}
+	return written;
+}
+
+static int release_byte(struct inode *inode, struct file *filp)
+{
+	u16 *val;
+
+	rcu_read_lock();
+	val = radix_tree_lookup(&byte_tree, (unsigned long)filp);
+	rcu_read_unlock();
+	if (val != NULL) {
+		spin_lock(&byte_lock);
+		radix_tree_delete(&byte_tree, (unsigned long)filp);
+		spin_unlock(&byte_lock);
+	}
+	return 0;
+}
+
+static const struct file_operations byte_fops = {
+	.llseek		= byte_lseek,
+	.read		= read_byte,
+	.write		= write_byte,
+	.release	= release_byte,
+};
+#endif /* CONFIG_DEV_BYTE */
+
 static const struct memdev {
 	const char *name;
 	mode_t mode;
@@ -857,6 +955,9 @@ static const struct memdev {
 #ifdef CONFIG_CRASH_DUMP
 	[12] = { "oldmem", 0, &oldmem_fops, NULL },
 #endif
+#ifdef CONFIG_DEV_BYTE
+	[13] = { "byte", 0666, &byte_fops, NULL },
+#endif
 };
 
 static int memory_open(struct inode *inode, struct file *filp)
-- 
1.7.3.4

Re: [PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Mike Frysinger]

On Mon, Apr 18, 2011 at 07:37, Alexander Holler wrote:
> I don't know if this patch qualifies for inclusion into the mainline kernel, but
> maybe someone else would be interested in such, so I'm posting it here.
>
> The main idea for that device was to have a source for 0xff to clear FLASH based
> devices. And to make it more generic, I've added the possibility to change the
> default value on a per file descriptor basis.
>
> I don't want to discuss if such a device makes sense or if such should be done
> in userspace only. I like it as a device, otherwise I wouldn't have written it.
> So besides this message, I will not take part in such a discussion.

cant you do this with cuse ?  that would satisfy the "has to be a
device" requirement while keeping it in userspace.

although, i have a hard time seeing there be a realistic perf diff with:
tr '\000' '\377' < /dev/zero | dd ......
-mike

Re: [PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

Am 18.04.2011 15:57, schrieb Mike Frysinger:
> On Mon, Apr 18, 2011 at 07:37, Alexander Holler wrote:
>> I don't know if this patch qualifies for inclusion into the mainline kernel, but
>> maybe someone else would be interested in such, so I'm posting it here.
>>
>> The main idea for that device was to have a source for 0xff to clear FLASH based
>> devices. And to make it more generic, I've added the possibility to change the
>> default value on a per file descriptor basis.
>>
>> I don't want to discuss if such a device makes sense or if such should be done
>> in userspace only. I like it as a device, otherwise I wouldn't have written it.
>> So besides this message, I will not take part in such a discussion.
>
> cant you do this with cuse ?  that would satisfy the "has to be a
> device" requirement while keeping it in userspace.

I know almost nothing about cuse, but I assume it should be possible to 
build such a device with cuse too.

> although, i have a hard time seeing there be a realistic perf diff with:
> tr '\000' '\377'<  /dev/zero | dd ......

There are many other ways to achieve such in userspace (even without 
using /dev/zero) and I want to avoid such a discussion.

As for /dev/zero there are many other possible reasons to use such a 
device, besides filling something with a value. For me it's as 
reasonable as dev/zero, just that it offers a bit more flexibility and 
provides another, at least for me useful, default value. Maybe 
/dev/nzero would have been a good name too. ;)

But I don't really care about inclusion into the kernel, it's just 
something I had lying around (and needed only marginally work to 
finalize as a proper patch) and I thought someone else could find it 
usefull and I should share that here.

Regards,

Alexander Holler

Re: [PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alan Cox]

> As for /dev/zero there are many other possible reasons to use such a 
> device, besides filling something with a value. For me it's as 
> reasonable as dev/zero, just that it offers a bit more flexibility and 
> provides another, at least for me useful, default value. Maybe 
> /dev/nzero would have been a good name too. ;)

/dev/zero exists not to put \0's into files as such but because it is
very useful to be able to map the zero page (a read only, or
copy-on-write blank page) into programs. The mmap is the reason it is
there.

> But I don't really care about inclusion into the kernel, it's just 
> something I had lying around (and needed only marginally work to 
> finalize as a proper patch) and I thought someone else could find it 
> usefull and I should share that here.

Implementationwise I think I would have gone for allocating a new device
and range of 256 minors - that would avoid the funky stuff setting what
it fills with as you'd just fill with the minor number.

Alan

Re: [PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

Hello,

Am 19.04.2011 10:44, schrieb Alan Cox:
>> As for /dev/zero there are many other possible reasons to use such a
>> device, besides filling something with a value. For me it's as
>> reasonable as dev/zero, just that it offers a bit more flexibility and
>> provides another, at least for me useful, default value. Maybe
>> /dev/nzero would have been a good name too. ;)
>
> /dev/zero exists not to put \0's into files as such but because it is
> very useful to be able to map the zero page (a read only, or
> copy-on-write blank page) into programs. The mmap is the reason it is
> there.

Thanks for the explanation.

>> But I don't really care about inclusion into the kernel, it's just
>> something I had lying around (and needed only marginally work to
>> finalize as a proper patch) and I thought someone else could find it
>> usefull and I should share that here.
>
> Implementationwise I think I would have gone for allocating a new device
> and range of 256 minors - that would avoid the funky stuff setting what
> it fills with as you'd just fill with the minor number.

I thought about that too, but that would have been to easy (and static). ;)

And the usage of file descriptors was the only idea I've come up with, 
which is multitasking and multiuser aware.

Regards,

Alexander

Re: [PATCH 0/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

Hello,

Am 19.04.2011 11:05, schrieb Alexander Holler:

>> Implementationwise I think I would have gone for allocating a new device
>> and range of 256 minors - that would avoid the funky stuff setting what
>> it fills with as you'd just fill with the minor number.
>
> I thought about that too, but that would have been to easy (and static). ;)
>
> And the usage of file descriptors was the only idea I've come up with,
> which is multitasking and multiuser aware.

If I add the possibility to use minors to set the default value (but 
still leave the possibility to change the default dynamically), would 
that raise the chance to become included? ;)

Regards,

Alexander

Re: [PATCH 1/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Pavel Machek]

On Mon 2011-04-18 13:37:56, Alexander Holler wrote:
> This device outputs by default 0xff instead 0 which makes more sense
> than 0 to clear e.g. FLASH based devices.

Well, now you should provide example where you mmap /dev/byte, then
write() the flash directly from the mapping.

... hmm, that brings good question: what happens on existing mappings
when the byte is changed?

> To make the device more general usable, the value it outputs is changeable
> on a per file descriptor basis through simple writes to it.
> Values can be decimal (0 - 255), octal (00 - 0377) or hex (0x0 - 0xff).
> For other values (or strings) written to it, the write operation returns an
> error and the subsequent output is undefined.
...
>  # Create a file of size 10GB and filled with 0xaa.
> exec 5<>/dev/byte # Open /dev/byte and assign fd 5 to it
> echo 0xaa >&5     # Instruct the device to output 0xaa

That's seriously strange. /dev/byte should be changeable... by writing
bytes.
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [PATCH 1/1] Implement /dev/byte (a generic byte source similiar to /dev/zero)

[Alexander Holler]

Am 20.04.2011 12:57, schrieb Pavel Machek:
> On Mon 2011-04-18 13:37:56, Alexander Holler wrote:
>> This device outputs by default 0xff instead 0 which makes more sense
>> than 0 to clear e.g. FLASH based devices.
>
> Well, now you should provide example where you mmap /dev/byte, then
> write() the flash directly from the mapping.
>
> ... hmm, that brings good question: what happens on existing mappings
> when the byte is changed?

I never used mmap (never had a need to use it) explicit and barely know 
what it does. And I don't see a reason to use mmap on /dev/byte. 
Otherwise I would have known before the reason why /dev/zero exists. ;)

So I can't answer what happens when someone uses mmap on /dev/byte and 
changes the value while the map exists (without testing it by myself).

>> To make the device more general usable, the value it outputs is changeable
>> on a per file descriptor basis through simple writes to it.
>> Values can be decimal (0 - 255), octal (00 - 0377) or hex (0x0 - 0xff).
>> For other values (or strings) written to it, the write operation returns an
>> error and the subsequent output is undefined.
> ...
>>   # Create a file of size 10GB and filled with 0xaa.
>> exec 5<>/dev/byte # Open /dev/byte and assign fd 5 to it
>> echo 0xaa>&5     # Instruct the device to output 0xaa
>
> That's seriously strange. /dev/byte should be changeable... by writing
> bytes.

As I've written before, that was the only solution I could come up with 
which makes it possible to change the output of /dev/byte on the fly 
without introducing race conditions (except ioctl). I know, it's ugly, 
but works, at least for common (read) file operations.

So it seems the only proper solution would be to use minors which would 
make such a device static.
So I better stick to something in userland, even when I would like to 
have at least /dev/byteFF. Chances seem to be minimal to get such 
included in the kernel if no one else sees a usage pattern for such.

Regards,

Alexander