sg_set_page not usable for .bss?

sg_set_page not usable for .bss?

[Jan Engelhardt]

On Monday 2008-12-01 23:02, John Haxby wrote:
>>>+	sg_init_table(sg, 2);
>>>+	sg_set_buf(&sg[0], data, n);
>>>+	strcpy(digest_password, sysrq_password);
>>>+	i = strlen(digest_password);
>>>+	sg_set_buf(&sg[1], digest_password, i);
>>
>> Could we directly use sysrq_password instead of copying it to
>> digest_password first?
>
> No :-)   Eventually I discovered the reason my code wasn't working
> boils down to the definition of sg_set_buf:
>
>   sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf))
>
> which doesn't work for sysrq_password.   I don't know why I'll
> double check.

Well, sysrq_password is in the .bss section, where as digest_password
is on the heap due to being kmalloc'ed. Maybe that makes a difference?
Someone more versed with the virtual memory layer might know.

>+static char sysrq_password[64];
>[...]
>+	digest_password = kmalloc(sizeof(sysrq_password), GFP_KERNEL);

Re: sg_set_page not usable for .bss?

[David Miller]

From: Jan Engelhardt <jengelh@medozas.de>
Date: Mon, 1 Dec 2008 23:40:18 +0100 (CET)

> 
> On Monday 2008-12-01 23:02, John Haxby wrote:
> >>>+	sg_init_table(sg, 2);
> >>>+	sg_set_buf(&sg[0], data, n);
> >>>+	strcpy(digest_password, sysrq_password);
> >>>+	i = strlen(digest_password);
> >>>+	sg_set_buf(&sg[1], digest_password, i);
> >>
> >> Could we directly use sysrq_password instead of copying it to
> >> digest_password first?
> >
> > No :-)   Eventually I discovered the reason my code wasn't working
> > boils down to the definition of sg_set_buf:
> >
> >   sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf))
> >
> > which doesn't work for sysrq_password.   I don't know why I'll
> > double check.
> 
> Well, sysrq_password is in the .bss section, where as digest_password
> is on the heap due to being kmalloc'ed. Maybe that makes a difference?
> Someone more versed with the virtual memory layer might know.

You can't use these interfaces on kernel image addresses.

Re: sg_set_page not usable for .bss?

[Jan Engelhardt]

On Tuesday 2008-12-02 01:10, David Miller wrote:
>> On Monday 2008-12-01 23:02, John Haxby wrote:
>> >>>+	sg_init_table(sg, 2);
>> >>>+	sg_set_buf(&sg[0], data, n);
>> >>>+	strcpy(digest_password, sysrq_password);
>> >>>+	i = strlen(digest_password);
>> >>>+	sg_set_buf(&sg[1], digest_password, i);
>> >>
>> >> Could we directly use sysrq_password instead of copying it to
>> >> digest_password first?
>> >
>> > No :-)   Eventually I discovered the reason my code wasn't working
>> > boils down to the definition of sg_set_buf:
>> >
>> >   sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf))
>> >
>> > which doesn't work for sysrq_password.   I don't know why I'll
>> > double check.
>> 
>> Well, sysrq_password is in the .bss section, where as digest_password
>> is on the heap due to being kmalloc'ed. Maybe that makes a difference?
>> Someone more versed with the virtual memory layer might know.
>
>You can't use these interfaces on kernel image addresses.
>
Great :-)  So what is the best way to use the SHA1 crypto algo
with in-kernel addresses?


Jan

Re: sg_set_page not usable for .bss?

[David Miller]

From: Jan Engelhardt <jengelh@medozas.de>
Date: Tue, 2 Dec 2008 01:13:34 +0100 (CET)

> 
> On Tuesday 2008-12-02 01:10, David Miller wrote:
> >> On Monday 2008-12-01 23:02, John Haxby wrote:
> >> >>>+	sg_init_table(sg, 2);
> >> >>>+	sg_set_buf(&sg[0], data, n);
> >> >>>+	strcpy(digest_password, sysrq_password);
> >> >>>+	i = strlen(digest_password);
> >> >>>+	sg_set_buf(&sg[1], digest_password, i);
> >> >>
> >> >> Could we directly use sysrq_password instead of copying it to
> >> >> digest_password first?
> >> >
> >> > No :-)   Eventually I discovered the reason my code wasn't working
> >> > boils down to the definition of sg_set_buf:
> >> >
> >> >   sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf))
> >> >
> >> > which doesn't work for sysrq_password.   I don't know why I'll
> >> > double check.
> >> 
> >> Well, sysrq_password is in the .bss section, where as digest_password
> >> is on the heap due to being kmalloc'ed. Maybe that makes a difference?
> >> Someone more versed with the virtual memory layer might know.
> >
> >You can't use these interfaces on kernel image addresses.
> >
> Great :-)  So what is the best way to use the SHA1 crypto algo
> with in-kernel addresses?

kmalloc and copy it there, or something like that, you just
can't use in-kernel addresses, ever.

Re: sg_set_page not usable for .bss?

[Jan Engelhardt]

On Tuesday 2008-12-02 01:14, David Miller wrote:
>> >> 
>> >> Well, sysrq_password is in the .bss section, where as digest_password
>> >> is on the heap due to being kmalloc'ed. Maybe that makes a difference?
>> >> Someone more versed with the virtual memory layer might know.
>> >
>> >You can't use these interfaces on kernel image addresses.
>> >
>> Great :-)  So what is the best way to use the SHA1 crypto algo
>> with in-kernel addresses?
>
>kmalloc and copy it there, or something like that, you just
>can't use in-kernel addresses, ever.
>
Yes, kmalloc is already used. But then, what sort of address
does kmalloc return, if not an address within kernelspace?
(usually >=0xc0000000 on standard i386)

Re: sg_set_page not usable for .bss?

[David Miller]

From: Jan Engelhardt <jengelh@medozas.de>
Date: Tue, 2 Dec 2008 02:41:02 +0100 (CET)

> 
> On Tuesday 2008-12-02 01:14, David Miller wrote:
> >> >> 
> >> >> Well, sysrq_password is in the .bss section, where as digest_password
> >> >> is on the heap due to being kmalloc'ed. Maybe that makes a difference?
> >> >> Someone more versed with the virtual memory layer might know.
> >> >
> >> >You can't use these interfaces on kernel image addresses.
> >> >
> >> Great :-)  So what is the best way to use the SHA1 crypto algo
> >> with in-kernel addresses?
> >
> >kmalloc and copy it there, or something like that, you just
> >can't use in-kernel addresses, ever.
> >
> Yes, kmalloc is already used. But then, what sort of address
> does kmalloc return, if not an address within kernelspace?
> (usually >=0xc0000000 on standard i386)

I said "kernel image" addresses are a problem, not "kernel space."

And by "kernel image" I mean addresses within the confines defined
by the sections of the vmlinux binary.