rebuild kernel add syscall

[嘉谟]

I want to and a new syscall
1  add
.long sys_mysyscall
in        arch/x86/kernel/syscall_table_32.S

2    add
       #define   __NR_mysyscall   341
in    arch/x86/include/asm/unistd_32.

3.    add

asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
{
    int source=sys_open(sourceFile,O_RDONLY,0);
    int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
    char buf[1024];
    mm_segment_t fs;
    fs = get_fs();
    set_fs(get_ds());
    int nread;

if(source>0 && dest>0)
    {
            while((nread=sys_read(source,buf,1024)) > 0)
                       sys_write(dest,buf,read);
    }
    else
    {
             printk("Error!");
    }
    sys_close(source);
    sys_close(dest);
    set_fs(fs);
    return 0;
}
in  kerrnel/sys.c

4. make menuconfig
5. make  all
6  make  modules_install

reboot

Now I want just do it one time
How can i make sure the  new syscall here is mysyscall  work fine  ?

[嘉谟]

I mean   weather  i can use sys_open  in   /kernel/sys.c
Or how can i test it  but i am  in user space ?

2011/3/10 xinyou yan :
> I want to and a new syscall
> 1  add
> .long sys_mysyscall
> in        arch/x86/kernel/syscall_table_32.S
>
> 2    add
>       #define   __NR_mysyscall   341
> in    arch/x86/include/asm/unistd_32.
> add
>
> asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
> {
>    int source=sys_open(sourceFile,O_RDONLY,0);
>    int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
>    char buf[1024];
>    mm_segment_t fs;
>    fs = get_fs();
>    set_fs(get_ds());
>    int nread;
>
> if(source>0 && dest>0)
>    {
>            while((nread=sys_read(source,buf,1024)) > 0)
>                       sys_write(dest,buf,read);
>    }
>    else
>    {
>             printk("Error!");
>    }
>    sys_close(source);
>    sys_close(dest);
>    set_fs(fs);
>    return 0;
> }
> in  kerrnel/sys.c
> make menuconfig
> make  all
>
> 6  make  modules_install
>
> reboot
>
> Now I want just do it one time
> How can i make sure the  new syscall here is mysyscall  work fine  ?

[JB]

xinyou yan <yxy.716 <at> gmail.com> writes:
> ...

$ man strace

JB

[Gilboa Davara]

On Thu, 2011-03-10 at 15:17 +0800, xinyou yan wrote:
> I want to and a new syscall
> 1  add
> .long sys_mysyscall
> in        arch/x86/kernel/syscall_table_32.S
>
> 2    add
>        #define   __NR_mysyscall   341
> in    arch/x86/include/asm/unistd_32.
> add
>
> asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
> {
>     int source=sys_open(sourceFile,O_RDONLY,0);
>     int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
>     char buf[1024];
>     mm_segment_t fs;
>     fs = get_fs();
>     set_fs(get_ds());
>     int nread;
> if(source>0 && dest>0)
> {
>         while((nread=sys_read(source,buf,1024)) > 0)
>                    sys_write(dest,buf,read);
> }
> else
> {
>          printk("Error!");
> }
> sys_close(source);
> sys_close(dest);
> set_fs(fs);
> return 0;
>
> }
> in  kerrnel/sys.c
> make menuconfig
> make  all
>
> 6  make  modules_install
>
> reboot
>
> Now I want just do it one time
> How can i make sure the  new syscall here is mysyscall  work fine  ?

Are you locked on using syscalls?
Unless you really require syscalls, I'd imagine that it's far easier to
use ioctl's instead (doesn't require a custom kernel), and use
filp_open / filp_close / file->read / file->write to access files from
within kernel space.

Two more things:
1. I'd avoid using stack based allocations in kernel mode. (Down to 8KB
in certain situations)
2. Always check error codes.

- Gilboa

[嘉谟]

It is a homework .  Use a kernel spcae as a char decieve
Which can offer  open write and so on.

I just  use a   char buffer[128] in kernel space.

Then i have some ideas.
I can use kernel space as char device. I can also use  kernel spcae as
a file system?  Could it be ?
If  memory  can be a filesystem  . Can  i use a file  to   simulate  。
It just  not be taken as a file . Like something
we store file  in it.

2011/3/13 Gilboa Davara :
> On Thu, 2011-03-10 at 15:17 +0800, xinyou yan wrote:
> > I want to and a new syscall
> > 1  add
> > .long sys_mysyscall
> > in        arch/x86/kernel/syscall_table_32.S
> >
> > 2    add
> >        #define   __NR_mysyscall   341
> > in    arch/x86/include/asm/unistd_32.
> > add
> >
> > asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
> > {
> >     int source=sys_open(sourceFile,O_RDONLY,0);
> >     int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
> >     char buf[1024];
> >     mm_segment_t fs;
> >     fs = get_fs();
> >     set_fs(get_ds());
> >     int nread;
> > if(source>0 && dest>0)
> > {
> >         while((nread=sys_read(source,buf,1024)) > 0)
> >                    sys_write(dest,buf,read);
> > }
> > else
> > {
> >          printk("Error!");
> > }
> > sys_close(source);
> > sys_close(dest);
> > set_fs(fs);
> > return 0;
> >
> > }
> > in  kerrnel/sys.c
> > make menuconfig
> > make  all
> >
> > 6  make  modules_install
> >
> > reboot
> >
> > Now I want just do it one time
> > How can i make sure the  new syscall here is mysyscall  work fine  ?
>
> Are you locked on using syscalls?
> Unless you really require syscalls, I'd imagine that it's far easier to
> use ioctl's instead (doesn't require a custom kernel), and use
> filp_open / filp_close / file->read / file->write to access files from
> within kernel space.
>
> Two more things:
> I'd avoid using stack based allocations in kernel mode. (Down to 8KB
>
> in certain situations)
> 2. Always check error codes.
> Gilboa
>
> --
> users mailing list
> [e-mail]
> To unsubscribe or change subscription options:
> https://admin.fedoraproject.org/mailman/listinfo/users
> Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines

[Gilboa Davara]

On Sun, 2011-03-13 at 21:51 +0800, xinyou yan wrote:
> It is a homework .  Use a kernel spcae as a char decieve
> Which can offer  open write and so on.
>
> I just  use a   char buffer[128] in kernel space.

As a general rule, using static parameters in kernel space is considered
hazardous; nevertheless, a 128 bytes is relatively safe.
> Then i have some ideas.
> I can use kernel space as char device. I can also use  kernel spcae as
> a file system?  Could it be ?

I'm not sure what you mean.
But in general, yes, you could register a character device that has
read/write/seek operations on a memory buffer. (Preferably, dynamically
allocated memory buffer)
A relatively simple (and easy to understand / implement) example is
the /proc file system. (Search for create_proc_entry in the kernel code)
> If  memory  can be a filesystem  . Can  i use a file  to   simulate  。
> It just  not be taken as a file . Like something
> we store file  in it.
>
> 2011/3/13 Gilboa Davara :
> > On Thu, 2011-03-10 at 15:17 +0800, xinyou yan wrote:
> > > I want to and a new syscall
> > > 1  add
> > > .long sys_mysyscall
> > > in        arch/x86/kernel/syscall_table_32.S
> > >
> > > 2    add
> > >        #define   __NR_mysyscall   341
> > > in    arch/x86/include/asm/unistd_32.
> > > add
> > >
> > > asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
> > > {
> > >     int source=sys_open(sourceFile,O_RDONLY,0);
> > >     int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
> > >     char buf[1024];
> > >     mm_segment_t fs;
> > >     fs = get_fs();
> > >     set_fs(get_ds());
> > >     int nread;
> > > if(source>0 && dest>0)
> > > {
> > >         while((nread=sys_read(source,buf,1024)) > 0)
> > >                    sys_write(dest,buf,read);
> > > }
> > > else
> > > {
> > >          printk("Error!");
> > > }
> > > sys_close(source);
> > > sys_close(dest);
> > > set_fs(fs);
> > > return 0;
> > >
> > > }
> > > in  kerrnel/sys.c
> > > make menuconfig
> > > make  all
> > >
> > > 6  make  modules_install
> > >
> > > reboot
> > >
> > > Now I want just do it one time
> > > How can i make sure the  new syscall here is mysyscall  work fine  ?
> >
> > Are you locked on using syscalls?
> > Unless you really require syscalls, I'd imagine that it's far easier to
> > use ioctl's instead (doesn't require a custom kernel), and use
> > filp_open / filp_close / file->read / file->write to access files from
> > within kernel space.
> >
> > Two more things:
> > I'd avoid using stack based allocations in kernel mode. (Down to 8KB
> >
> > in certain situations)
> > 2. Always check error codes.
> > Gilboa
> >
> > --
> > users mailing list
> > [e-mail]
> > To unsubscribe or change subscription options:
> > https://admin.fedoraproject.org/mailman/listinfo/users
> > Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines

[嘉谟]

2011/3/14 Gilboa Davara :
> On Sun, 2011-03-13 at 21:51 +0800, xinyou yan wrote:
> > It is a homework .  Use a kernel spcae as a char decieve
> > Which can offer  open write and so on.
> >
> > I just  use a   char buffer[128] in kernel space.
>
> As a general rule, using static parameters in kernel space is considered
> hazardous; nevertheless, a 128 bytes is relatively safe.
> > Then i have some ideas.
> > I can use kernel space as char device. I can also use  kernel spcae as
> > a file system?  Could it be ?
>
> I'm not sure what you mean.

May be it can't be implement.
I mean . I create  a file  with a   big size .
Then i  write   a super block struct  ( I define it like the  linux)
Then  Inode   table.
Then  Date    Blocks.

I  reimplement  some syscall on it .
I just want to  make it like a filesystem  .
However  I stop at . How to  init  all of this  struct.

So sorry to make you confuse .   English is not my native language , I
will try to   do better than what early i was .
> But in general, yes, you could register a character device that has
> read/write/seek operations on a memory buffer. (Preferably, dynamically
> allocated memory buffer)
> A relatively simple (and easy to understand / implement) example is
> the /proc file system. (Search for create_proc_entry in the kernel code)
> > If  memory  can be a filesystem  . Can  i use a file  to   simulate  。
> > It just  not be taken as a file . Like something
> > we store file  in it.
> >
> > 2011/3/13 Gilboa Davara :
> > > On Thu, 2011-03-10 at 15:17 +0800, xinyou yan wrote:
> > > > I want to and a new syscall
> > > > 1  add
> > > > .long sys_mysyscall
> > > > in        arch/x86/kernel/syscall_table_32.S
> > > >
> > > > 2    add
> > > >        #define   __NR_mysyscall   341
> > > > in    arch/x86/include/asm/unistd_32.
> > > > add
> > > >
> > > > asmlinkage int sys_mysyscall(char* sourceFile,char* destFile)
> > > > {
> > > >     int source=sys_open(sourceFile,O_RDONLY,0);
> > > >     int dest=sys_open(destFile,O_WRONLY|O_CREAT|O_TRUNC,0600);
> > > >     char buf[1024];
> > > >     mm_segment_t fs;
> > > >     fs = get_fs();
> > > >     set_fs(get_ds());
> > > >     int nread;
> > > > if(source>0 && dest>0)
> > > > {
> > > >         while((nread=sys_read(source,buf,1024)) > 0)
> > > >                    sys_write(dest,buf,read);
> > > > }
> > > > else
> > > > {
> > > >          printk("Error!");
> > > > }
> > > > sys_close(source);
> > > > sys_close(dest);
> > > > set_fs(fs);
> > > > return 0;
> > > >
> > > > }
> > > > in  kerrnel/sys.c
> > > > make menuconfig
> > > > make  all
> > > >
> > > > 6  make  modules_install
> > > >
> > > > reboot
> > > >
> > > > Now I want just do it one time
> > > > How can i make sure the  new syscall here is mysyscall  work fine  ?
> > >
> > > Are you locked on using syscalls?
> > > Unless you really require syscalls, I'd imagine that it's far easier to
> > > use ioctl's instead (doesn't require a custom kernel), and use
> > > filp_open / filp_close / file->read / file->write to access files from
> > > within kernel space.
> > >
> > > Two more things:
> > > I'd avoid using stack based allocations in kernel mode. (Down to 8KB
> > >
> > > in certain situations)
> > > 2. Always check error codes.
> > > Gilboa
> > >
> > > --
> > > users mailing list
> > > [e-mail]
> > > To unsubscribe or change subscription options:
> > > https://admin.fedoraproject.org/mailman/listinfo/users
> > > Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
>
> --
> users mailing list
> [e-mail]
> To unsubscribe or change subscription options:
> https://admin.fedoraproject.org/mailman/listinfo/users
> Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines

OT: Re: rebuild kernel add syscall

[Gilboa Davara]

On Mon, 2011-03-14 at 21:24 +0800, xinyou yan wrote:
> May be it can't be implement.
> I mean . I create  a file  with a   big size .
> Then i  write   a super block struct  ( I define it like the  linux)
> Then  Inode   table.
> Then  Date    Blocks.
>
> I  reimplement  some syscall on it .
> I just want to  make it like a filesystem  .
> However  I stop at . How to  init  all of this  struct.
>
> So sorry to make you confuse .   English is not my native language , I
> will try to   do better than what early i was .

First, don't worry about your English - you're not alone. (My English
usually takes a dive when I wake up)
Second, if I understand you correctly, you plan to:
1. Implement your own syscall.
2. Call the syscall from a user process.
3. Open a file (from within kernel mode).
4. Write data.
5. Close the file.
6. Exit from the syscall.

... And you have to use syscall because it's the assignment. (Making it
impossible to use far less intrusive methods such as procfs,
character-device, ioctl, etc)

Am I correct?

- Gilboa

[Joe Zeff]

On 03/14/2011 03:21 AM, Gilboa Davara wrote:
> As a general rule, using static parameters in kernel space is considered
> hazardous; nevertheless, a 128 bytes is relatively safe.

Just make sure that you never write more than 128 bytes (including 
termination) into it.

[嘉谟]

If I want to  sleep  until a  event happened ?
How can i do it in user mode.

2011/3/15 Joe Zeff :
> On 03/14/2011 03:21 AM, Gilboa Davara wrote:
> > As a general rule, using static parameters in kernel space is considered
> > hazardous; nevertheless, a 128 bytes is relatively safe.
>
> Just make sure that you never write more than 128 bytes (including
> termination) into it.
> --
> users mailing list
> [e-mail]
> To unsubscribe or change subscription options:
> https://admin.fedoraproject.org/mailman/listinfo/users
> Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines

[JB]

xinyou yan <yxy.716 <at> gmail.com> writes:
> If I want to  sleep  until a  event happened ?
> How can i do it in user mode.
> ...

Well, it depends what event and under what circumstances.

The are single-threading and multi-threading programs.

An event can be triggered by a counter, timer, signal, inotify, mutex lock/
unlock, barrier, polling, etc.

Some of these logical constructs are generic, others are composites of existing
ones.

Often you put these logical constructs in a loop, perhaps with your own added
timing or break-out logic.

$ apropos timer
$ apropos signal
$ apropos pthread
$ propos notify
$ apropos event
$ apropos sleep
$ apropos poll

JB