Re: lseek on /proc/kmsg

Re: lseek on /proc/kmsg

[Robert Hancock]

linux-os wrote:
> On Tue, 22 Mar 2005, Jan Engelhardt wrote:
> 
>> Hi,
>>
>>> how am I supposed to clear the kmsg buffer since it's not a terminal??
>>
>>
>> fd = open("/proc/kmsg", O_RDONLY | O_NONBLOCK);
>> while(read(fd, buf, sizeof(buf)) > 0);
>> if(errno == EAGAIN) { printf("Clear!\n"); }
>>
>> This is language (spoken-wise) neutral :p
>>
> 
> Gawd, you are a hacker. I already have to suck on pipes
> because I can't seek them. Now, I can't even seek a
> file-system???!!

I'm not sure that seek makes any sense on that, since it is more like a 
pipe than a normal file..

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: lseek on /proc/kmsg

[Bodo Eggert]

Jan Engelhardt <jengelh@linux01.gwdg.de> wrote:

> +static loff_t kmsg_seek(struct file *filp, loff_t offset, int origin) {
> +    if(origin != 2 /* SEEK_END */ || offset < 0) { return -ESPIPE; }
                                              ^^^

"Allow" seeking past the end of the buffer?

lseek on /proc/kmsg

[linux-os]

Anybody know what is __supposed__ to happen with lseek()
on /proc/kmsg.  Right now, it does nothing, always returns
0. Given that, how am I supposed to clear the kmsg buffer
since it's not a terminal??

Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 on an i686 machine (5537.79 BogoMips).
  Notice : All mail here is now cached for review by Dictator Bush.
                  98.36% of all statistics are fiction.

Re: lseek on /proc/kmsg

[Jan Engelhardt]

Hi,


> how am I supposed to clear the kmsg buffer since it's not a terminal??

fd = open("/proc/kmsg", O_RDONLY | O_NONBLOCK);
while(read(fd, buf, sizeof(buf)) > 0);
if(errno == EAGAIN) { printf("Clear!\n"); }

This is language (spoken-wise) neutral :p



Jan Engelhardt
-- 

Re: lseek on /proc/kmsg

[linux-os]

On Tue, 22 Mar 2005, Jan Engelhardt wrote:

> Hi,
>
>> how am I supposed to clear the kmsg buffer since it's not a terminal??
>
> fd = open("/proc/kmsg", O_RDONLY | O_NONBLOCK);
> while(read(fd, buf, sizeof(buf)) > 0);
> if(errno == EAGAIN) { printf("Clear!\n"); }
>
> This is language (spoken-wise) neutral :p
>

Gawd, you are a hacker. I already have to suck on pipes
because I can't seek them. Now, I can't even seek a
file-system???!!

>
>
> Jan Engelhardt
> -- 
>

Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 on an i686 machine (5537.79 BogoMips).
  Notice : All mail here is now cached for review by Dictator Bush.
                  98.36% of all statistics are fiction.

Re: lseek on /proc/kmsg

[Jan Engelhardt]

> Gawd, you are a hacker. I already have to suck on pipes
> because I can't seek them. Now, I can't even seek a
> file-system???!!

Here goodie goodie...

diff -pdru linux-2.6.11.4/fs/proc/kmsg.c linux-2.6.11-AS9/fs/proc/kmsg.c
--- linux-2.6.11.4/fs/proc/kmsg.c       2005-03-21 20:14:58.000000000 +0100
+++ linux-2.6.11-AS9/fs/proc/kmsg.c     2005-03-22 21:28:40.000000000 +0100
@@ -46,10 +46,15 @@ static unsigned int kmsg_poll(struct fil
        return 0;
 }

+static loff_t kmsg_seek(struct file *filp, loff_t offset, int origin) {
+    if(origin != 2 /* SEEK_END */ || offset < 0) { return -ESPIPE; }
+    return do_syslog(5, NULL, 0);
+}

 struct file_operations proc_kmsg_operations = {
        .read           = kmsg_read,
        .poll           = kmsg_poll,
        .open           = kmsg_open,
        .release        = kmsg_release,
+        .llseek         = kmsg_seek,
 };


Works so far that do_syslog is called with the correct parameters --
however, that does not work. (Did I discover a bug?)

# rcsyslog stop;  # so that kmsg is not slurped by someone else
# perl -le 'open X,"</proc/kmsg";seek X,0,2;print read X,$b,3'

the perl command should block, because with the seek(), we've just emptied the 
syslog ring buffer. Obviously, it does not, and read() succeeds - prints 3.
Any hints on what's wrong here?


Jan Engelhardt
-- 

Re: lseek on /proc/kmsg

[linux-os]

On Tue, 22 Mar 2005, Jan Engelhardt wrote:

>> Gawd, you are a hacker. I already have to suck on pipes
>> because I can't seek them. Now, I can't even seek a
>> file-system???!!
>
> Here goodie goodie...
>
> diff -pdru linux-2.6.11.4/fs/proc/kmsg.c linux-2.6.11-AS9/fs/proc/kmsg.c
> --- linux-2.6.11.4/fs/proc/kmsg.c       2005-03-21 20:14:58.000000000 +0100
> +++ linux-2.6.11-AS9/fs/proc/kmsg.c     2005-03-22 21:28:40.000000000 +0100
> @@ -46,10 +46,15 @@ static unsigned int kmsg_poll(struct fil
>        return 0;
> }
>
> +static loff_t kmsg_seek(struct file *filp, loff_t offset, int origin) {
> +    if(origin != 2 /* SEEK_END */ || offset < 0) { return -ESPIPE; }
> +    return do_syslog(5, NULL, 0);
> +}
>
> struct file_operations proc_kmsg_operations = {
>        .read           = kmsg_read,
>        .poll           = kmsg_poll,
>        .open           = kmsg_open,
>        .release        = kmsg_release,
> +        .llseek         = kmsg_seek,
> };
>
>
> Works so far that do_syslog is called with the correct parameters --
> however, that does not work. (Did I discover a bug?)
>
> # rcsyslog stop;  # so that kmsg is not slurped by someone else
> # perl -le 'open X,"</proc/kmsg";seek X,0,2;print read X,$b,3'
>
> the perl command should block, because with the seek(), we've just emptied the
> syslog ring buffer. Obviously, it does not, and read() succeeds - prints 3.
> Any hints on what's wrong here?
>

Sure, read() needs to be modified to respect the file-position
set by kmsg_seek(). I don't think you can get away with the
call back into do_syslog.

In other words we shouldn't move a user-mode hack into the
kernel.

>
> Jan Engelhardt
> -- 
>

Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 on an i686 machine (5537.79 BogoMips).
  Notice : All mail here is now cached for review by Dictator Bush.
                  98.36% of all statistics are fiction.

Re: lseek on /proc/kmsg

[Jan Engelhardt]

1> Sure, read() needs to be modified to respect the file-position
1> set by kmsg_seek(). I don't think you can get away with the
1> call back into do_syslog.

2>I'm not sure that seek makes any sense on that, since it is more like a 
2>pipe than a normal file..

Well, seek(fd, 0, SEEK_END) could be used to flush a pipe's buffers.

0>> +static loff_t kmsg_seek(struct file *filp, loff_t offset, int origin) {
0>> +    if(origin != 2 /* SEEK_END */ || offset < 0) { return -ESPIPE; }
3> "Allow" seeking past the end of the buffer?

Well, what does lseek(fd, >0, SEEK_END) do on normal files?



Jan Engelhardt
-- 

Re: lseek on /proc/kmsg

[H. Peter Anvin]

Followup to:  <Pine.LNX.4.61.0503230811020.21578@yvahk01.tjqt.qr>
By author:    Jan Engelhardt <jengelh@linux01.gwdg.de>
In newsgroup: linux.dev.kernel
> 
> Well, what does lseek(fd, >0, SEEK_END) do on normal files?
> 

Sets the file pointer beyond the end of the file (a write() there will
extend the file.)

	-hpa

Re: lseek on /proc/kmsg

[linux-os]

On Wed, 23 Mar 2005, Jan Engelhardt wrote:

>
> 1> Sure, read() needs to be modified to respect the file-position
> 1> set by kmsg_seek(). I don't think you can get away with the
> 1> call back into do_syslog.
>
> 2>I'm not sure that seek makes any sense on that, since it is more like a
> 2>pipe than a normal file..
>
> Well, seek(fd, 0, SEEK_END) could be used to flush a pipe's buffers.
>

Yep. That's what I tried to do. Just returns 0 and continues to
contain all the cached data.


> 0>> +static loff_t kmsg_seek(struct file *filp, loff_t offset, int origin) {
> 0>> +    if(origin != 2 /* SEEK_END */ || offset < 0) { return -ESPIPE; }
> 3> "Allow" seeking past the end of the buffer?
>
> Well, what does lseek(fd, >0, SEEK_END) do on normal files?
>

Goes to the end plus offset. A subsequent read returns EOF
or 0 depending upon your read mechanism. This is what I wanted
to do but with no offset.

Currently, I do this crap:

     for(;;)
     {
         pfd.fd      = ifd;
         pfd.events  = POLLIN;
         pfd.revents = 0;
         if(poll(&pfd, 1, 0) <= 0)
             break;
         (void)read(ifd, ibuf, BUF_LEN);
     }

I should be able to just lseek(ifd, 0 SEEK_END);

> -- 
>

Cheers,
Dick Johnson
Penguin : Linux version 2.6.11 on an i686 machine (5537.79 BogoMips).
  Notice : All mail here is now cached for review by Dictator Bush.
                  98.36% of all statistics are fiction.