reiser4 problem with pam_mktemp

reiser4 problem with pam_mktemp

[sergey ivanov]

I have a problem with reiser4.
Recently I have installed Altlinux Compact-rc2 on reiserfs (v3.6),
patched kernel 2.6.12 with patch 2.6.12-mm2, created new reiser4
partition hda6 and copied by #cp -ax / /mnt/hda6 there. Fixed /etc/fstab
and /boot/grub for new partition and booted. And could not login, with
messages in virtual consoles :
---

pam_tcb[8386]:  login: Authentication passed for root from (uid=0)
Login incorrect
---
So I rebooted in original on reiserfs 3.6, edited /mnt/hda6/etc/inittab to change one of "/sbin/mingetty tty6" to be "/bin/bash" and after reboot to reiser4 based system saw, that the problem in pam_mktemp, by commenting out string 
# account  required  pam_mktemp.so
in file /etc/pam.d/system_auth the problem get fixed. 
Also it get fixed if I create some non-reiser4 partition to place /tmp on it.

	With best regards,
		Sergey Ivanov

Re: reiser4 problem with pam_mktemp

[Vladimir V. Saveliev]

Hello

sergey ivanov wrote:
> I have a problem with reiser4.
> Recently I have installed Altlinux Compact-rc2 on reiserfs (v3.6),
> patched kernel 2.6.12 with patch 2.6.12-mm2, created new reiser4
> partition hda6 and copied by #cp -ax / /mnt/hda6 there. Fixed /etc/fstab
> and /boot/grub for new partition and booted. And could not login, with
> messages in virtual consoles :
> ---
> 
> pam_tcb[8386]:  login: Authentication passed for root from (uid=0)
> Login incorrect
> ---
> So I rebooted in original on reiserfs 3.6, edited /mnt/hda6/etc/inittab to change one of "/sbin/mingetty tty6" to be "/bin/bash" and after reboot to reiser4 based system saw, that the problem in pam_mktemp, by commenting out string 
> # account  required  pam_mktemp.so
> in file /etc/pam.d/system_auth the problem get fixed. 
> Also it get fixed if I create some non-reiser4 partition to place /tmp on it.
> 

Thanks for the report.
The problem is that reiser4 does not support ext2's file attributes pam_mktemp deals with.

Maybe reiser4 should have support for that like reiserfs, but now it does not.

Re: reiser4 problem with pam_mktemp

[sergey ivanov]

Hi Vladimir,
pam_mktemp can work on filesystem which does not support ext2's file attributes.
For example, it works well on tmpfs.
For pam_mktemp to recognize the case and work without attributes, 
the filesystem should return 'ENOTTY' (Inappropriate ioctl for device) 
as a result of an setting/getting attributes call, as it tmpfs does. 
But reiser4 returns 'ENOSYS' (Function not implemented).
Will you accept patch changing this code to 'ENOTTY'?
	Sergey

Vladimir V. Saveliev wrote:

> Hello
>
> sergey ivanov wrote:
>
>> I have a problem with reiser4.
>> Recently I have installed Altlinux Compact-rc2 on reiserfs (v3.6),
>> patched kernel 2.6.12 with patch 2.6.12-mm2, created new reiser4
>> partition hda6 and copied by #cp -ax / /mnt/hda6 there. Fixed /etc/fstab
>> and /boot/grub for new partition and booted. And could not login, with
>> messages in virtual consoles :
>> ---
>>
>> pam_tcb[8386]:  login: Authentication passed for root from (uid=0)
>> Login incorrect
>> ---
>> So I rebooted in original on reiserfs 3.6, edited
>> /mnt/hda6/etc/inittab to change one of "/sbin/mingetty tty6" to be
>> "/bin/bash" and after reboot to reiser4 based system saw, that the
>> problem in pam_mktemp, by commenting out string # account  required 
>> pam_mktemp.so
>> in file /etc/pam.d/system_auth the problem get fixed. Also it get
>> fixed if I create some non-reiser4 partition to place /tmp on it.
>>
>
> Thanks for the report.
> The problem is that reiser4 does not support ext2's file attributes
> pam_mktemp deals with.
>
> Maybe reiser4 should have support for that like reiserfs, but now it
> does not.

Re: reiser4 problem with pam_mktemp

[Vladimir V. Saveliev]

Hello

sergey ivanov wrote:
> Hi Vladimir,
> pam_mktemp can work on filesystem which does not support ext2's file attributes.
> For example, it works well on tmpfs.
> For pam_mktemp to recognize the case and work without attributes, 
> the filesystem should return 'ENOTTY' (Inappropriate ioctl for device) 
> as a result of an setting/getting attributes call, as it tmpfs does. 
> But reiser4 returns 'ENOSYS' (Function not implemented).
> Will you accept patch changing this code to 'ENOTTY'?

yes

> 	Sergey
> 
> Vladimir V. Saveliev wrote:
> 
> 
>>Hello
>>
>>sergey ivanov wrote:
>>
>>
>>>I have a problem with reiser4.
>>>Recently I have installed Altlinux Compact-rc2 on reiserfs (v3.6),
>>>patched kernel 2.6.12 with patch 2.6.12-mm2, created new reiser4
>>>partition hda6 and copied by #cp -ax / /mnt/hda6 there. Fixed /etc/fstab
>>>and /boot/grub for new partition and booted. And could not login, with
>>>messages in virtual consoles :
>>>---
>>>
>>>pam_tcb[8386]:  login: Authentication passed for root from (uid=0)
>>>Login incorrect
>>>---
>>>So I rebooted in original on reiserfs 3.6, edited
>>>/mnt/hda6/etc/inittab to change one of "/sbin/mingetty tty6" to be
>>>"/bin/bash" and after reboot to reiser4 based system saw, that the
>>>problem in pam_mktemp, by commenting out string # account  required 
>>>pam_mktemp.so
>>>in file /etc/pam.d/system_auth the problem get fixed. Also it get
>>>fixed if I create some non-reiser4 partition to place /tmp on it.
>>>
>>
>>Thanks for the report.
>>The problem is that reiser4 does not support ext2's file attributes
>>pam_mktemp deals with.
>>
>>Maybe reiser4 should have support for that like reiserfs, but now it
>>does not.
> 
> 
> 
> 
> 

Re: reiser4 problem with pam_mktemp

[sergey ivanov]

I have not succeeded with integrating reiser4 & pam_mktemp.
From one side I recompiled kernel with reiser4 patches modified to
return ENOTTY for attempt to deal with attributes, as tmpfs does.
Pam_mktemp does not work after this. So I returned to the kernel with
original Namesys' patches.
Then by my apeal pam_mktemp was modified to recognize ENOSYS the same
way it recognizes ENOTTY returned by tmpfs and try to deal without
attributes.
But this also does not help pam_mktemp to work.
So I did the following: rebooted to runlevel 1, and then did
# strace -o su.log -fF su - seriv
Once with /tmp on tmpfs and once with /tmp on reiser4.
And the following catches my attention: while on tmpfs, the lines
dealing with attributes looks like the following:
---
2034 mkdir("/tmp/.private", 0711) = 0
2034 lstat64("/tmp/.private", {st_mode=S_IFDIR|0711, st_size=40, ...}) = 0

2034  open("/tmp/.private", O_RDONLY)   = 3
2034  ioctl(3, EXT2_IOC_GETFLAGS, 0xbfb7aab8) = -1 ENOTTY (Inappropriate
ioctl for device)
2034  close(3)                          = 0
2034  mkdir("/tmp/.private/seriv", 01700) = 0
2034  open("/tmp/.private/seriv", O_RDONLY) = 3
2034  ioctl(3, EXT2_IOC_GETFLAGS, 0xbfb7aa88) = -1 ENOTTY (Inappropriate
ioctl for device)
2034  close(3)                          = 0
2034  chown32("/tmp/.private/seriv", 0, 500) = 0
2034  chmod("/tmp/.private/seriv", 01770) = 0
---
But with /tmp on reiser4, they are:
---
1993  mkdir("/tmp/.private", 0711)      = -1 EEXIST (File exists)
1993  lstat64("/tmp/.private", {st_mode=S_IFDIR|0711, st_size=7, ...}) = 0
1993  open("/tmp/.private", O_RDONLY)   = 3
1993  ioctl(3, EXT2_IOC_GETFLAGS, 0xbf9e37f8) = -1 EISDIR (Is a directory)
1993  close(3)                          = 0
1993  mkdir("/tmp/.private/seriv", 01700) = -1 EEXIST (File exists)
1993  open("/tmp/.private/seriv", O_RDONLY) = 3
1993  ioctl(3, EXT2_IOC_GETFLAGS, 0xbf9e37c8) = -1 EISDIR (Is a directory)
1993  close(3)                          = 0
1993  getuid32()                        = 0
1993  write(2, "Account management:- Cannot make"..., 85) = 85
---
So it looks like reiser4 tries to be good and answer to ioctl
getting/setting attributes "EEXIST" and "EISDIR",
and not return error "ENOTTY".
This is the reason, why pam_mktemp passes the tests if filesystem
support attributes with confidence that reiser4 can do attributes.
I think this was the right decision to return fake successes at the
beginning of reiser4 development, but now I think reiser4 should behave
more strictly and rejects this attribute setting/getting ioctl by ENOTTY
codes.
-- 
	Sergey

Re: reiser4 problem with pam_mktemp

[michael chang]

On 8/11/05, sergey ivanov <seriv@parkheights.dyndns.org> wrote:
> I have not succeeded with integrating reiser4 & pam_mktemp.
> From one side I recompiled kernel with reiser4 patches modified to
> return ENOTTY for attempt to deal with attributes, as tmpfs does.
> Pam_mktemp does not work after this. So I returned to the kernel with
> original Namesys' patches.

> # strace -o su.log -fF su - seriv

> 1993  mkdir("/tmp/.private", 0711)      = -1 EEXIST (File exists)
> 1993  lstat64("/tmp/.private", {st_mode=S_IFDIR|0711, st_size=7, ...}) = 0
> 1993  open("/tmp/.private", O_RDONLY)   = 3
> 1993  ioctl(3, EXT2_IOC_GETFLAGS, 0xbf9e37f8) = -1 EISDIR (Is a directory)

> So it looks like reiser4 tries to be good and answer to ioctl
> getting/setting attributes "EEXIST" and "EISDIR",
> and not return error "ENOTTY".

This seems kind of confusing -- the problem with the idea that all
files are files and directories... do we really want to be returning
that it's a directory here?  My guess is that this is the confusing
bit about this filesystem.  AFAIK, this is probably confusing the PAM
code or whatever.  *shrugs*

-- 
~Mike
 - Just my two cents
 - No man is an island, and no man is unable.

Re: reiser4 problem with pam_mktemp

[sergey ivanov]

Thank you, it works!
I've just recompiled kernel and now / mounted on reiser4 with /tmp in
the same filesystem,
and pam_mktemp works without problems.
-- 
    Sergey.

Vladimir V. Saveliev пишет:

> Hello
>
> sergey ivanov wrote:
>
>> I have not succeeded with integrating reiser4 & pam_mktemp.
>
>
> Would you try the attached patch, please.
>
>>> From one side I recompiled kernel with reiser4 patches modified to
>>
>> return ENOTTY for attempt to deal with attributes, as tmpfs does.
>> Pam_mktemp does not work after this. So I returned to the kernel with
>> original Namesys' patches.
>> Then by my apeal pam_mktemp was modified to recognize ENOSYS the same
>> way it recognizes ENOTTY returned by tmpfs and try to deal without
>> attributes.
>> But this also does not help pam_mktemp to work.
>> So I did the following: rebooted to runlevel 1, and then did
>> # strace -o su.log -fF su - seriv
>> Once with /tmp on tmpfs and once with /tmp on reiser4.
>> And the following catches my attention: while on tmpfs, the lines
>> dealing with attributes looks like the following:
>> ---
>> 2034 mkdir("/tmp/.private", 0711) = 0
>> 2034 lstat64("/tmp/.private", {st_mode=S_IFDIR|0711, st_size=40,
>> ...}) = 0
>>
>> 2034  open("/tmp/.private", O_RDONLY)   = 3
>> 2034  ioctl(3, EXT2_IOC_GETFLAGS, 0xbfb7aab8) = -1 ENOTTY (Inappropriate
>> ioctl for device)
>> 2034  close(3)                          = 0
>> 2034  mkdir("/tmp/.private/seriv", 01700) = 0
>> 2034  open("/tmp/.private/seriv", O_RDONLY) = 3
>> 2034  ioctl(3, EXT2_IOC_GETFLAGS, 0xbfb7aa88) = -1 ENOTTY (Inappropriate
>> ioctl for device)
>> 2034  close(3)                          = 0
>> 2034  chown32("/tmp/.private/seriv", 0, 500) = 0
>> 2034  chmod("/tmp/.private/seriv", 01770) = 0
>> ---
>> But with /tmp on reiser4, they are:
>> ---
>> 1993  mkdir("/tmp/.private", 0711)      = -1 EEXIST (File exists)
>> 1993  lstat64("/tmp/.private", {st_mode=S_IFDIR|0711, st_size=7,
>> ...}) = 0
>> 1993  open("/tmp/.private", O_RDONLY)   = 3
>> 1993  ioctl(3, EXT2_IOC_GETFLAGS, 0xbf9e37f8) = -1 EISDIR (Is a
>> directory)
>> 1993  close(3)                          = 0
>> 1993  mkdir("/tmp/.private/seriv", 01700) = -1 EEXIST (File exists)
>> 1993  open("/tmp/.private/seriv", O_RDONLY) = 3
>> 1993  ioctl(3, EXT2_IOC_GETFLAGS, 0xbf9e37c8) = -1 EISDIR (Is a
>> directory)
>> 1993  close(3)                          = 0
>> 1993  getuid32()                        = 0
>> 1993  write(2, "Account management:- Cannot make"..., 85) = 85
>> ---
>> So it looks like reiser4 tries to be good and answer to ioctl
>> getting/setting attributes "EEXIST" and "EISDIR", and not return
>> error "ENOTTY". This is the reason, why pam_mktemp passes the tests
>> if filesystem support attributes with confidence that reiser4 can do
>> attributes. I think this was the right decision to return fake
>> successes at the beginning of reiser4 development, but now I think
>> reiser4 should behave more strictly and rejects this attribute
>> setting/getting ioctl by ENOTTY codes.
>
>
>------------------------------------------------------------------------
>
> fs/reiser4/plugin/file/file.c |    2 +-
> fs/reiser4/plugin/object.c    |   17 ++++++++++++-----
> 2 files changed, 13 insertions(+), 6 deletions(-)
>
>diff -puN fs/reiser4/plugin/object.c~reiser4-pam_mktemp-fix fs/reiser4/plugin/object.c
>--- linux-2.6.13-rc5-mm1/fs/reiser4/plugin/object.c~reiser4-pam_mktemp-fix	2005-08-12 13:42:05.904058842 +0400
>+++ linux-2.6.13-rc5-mm1-vs/fs/reiser4/plugin/object.c	2005-08-12 13:46:12.622408259 +0400
>@@ -966,6 +966,13 @@ isdir(void)
> 
> #define eisdir ((void *)isdir)
> 
>+static int
>+notty(void)
>+{
>+	return RETERR(-ENOTTY);
>+}
>+#define enotty ((void *)notty)
>+
> static ssize_t
> perm(void)
> {
>@@ -1243,7 +1250,7 @@ file_plugin file_plugins[LAST_FILE_PLUGI
> 		.read = eisdir,
> 		.write = eisdir,
> 		.release = release_dir,
>-		.ioctl = eisdir,
>+		.ioctl = enotty,
> 		.mmap = eisdir,
> 		.get_block = NULL,
> 		.flow_by_inode = NULL,
>@@ -1299,7 +1306,7 @@ file_plugin file_plugins[LAST_FILE_PLUGI
> 		.read = eperm,
> 		.write = eperm,
> 		.release = NULL,
>-		.ioctl = eperm,
>+		.ioctl = enotty,
> 		.mmap = eperm,
> 		.sync = sync_common,
> 		.get_block = NULL,
>@@ -1358,7 +1365,7 @@ file_plugin file_plugins[LAST_FILE_PLUGI
> 		.read = eperm,
> 		.write = eperm,
> 		.release = NULL,
>-		.ioctl = eperm,
>+		.ioctl = enotty,
> 		.mmap = eperm,
> 		.sync = sync_common,
> 		.get_block = NULL,
>@@ -1413,7 +1420,7 @@ file_plugin file_plugins[LAST_FILE_PLUGI
> 		.read              = read_pseudo,
> 		.write             = write_pseudo,
> 		.release           = release_pseudo,
>-		.ioctl             = eperm,
>+		.ioctl             = enotty,
> 		.mmap              = eperm,
> 		.sync = sync_common,
> 		.get_block         = eperm,
>@@ -1470,7 +1477,7 @@ file_plugin file_plugins[LAST_FILE_PLUGI
> 		.read = read_cryptcompress,
> 		.write = write_cryptcompress,
> 		.release = NULL,
>-		.ioctl = NULL,
>+		.ioctl = enotty,
> 		.mmap = mmap_cryptcompress,
> 		.get_block = get_block_cryptcompress,
> 		.sync = sync_common,
>diff -puN fs/reiser4/plugin/file/file.c~reiser4-pam_mktemp-fix fs/reiser4/plugin/file/file.c
>--- linux-2.6.13-rc5-mm1/fs/reiser4/plugin/file/file.c~reiser4-pam_mktemp-fix	2005-08-12 13:45:48.276202815 +0400
>+++ linux-2.6.13-rc5-mm1-vs/fs/reiser4/plugin/file/file.c	2005-08-12 13:45:58.469126159 +0400
>@@ -2361,7 +2361,7 @@ ioctl_unix_file(struct inode *inode, str
> 		break;
> 
> 	default:
>-		result = RETERR(-ENOSYS);
>+		result = RETERR(-ENOTTY);
> 		break;
> 	}
> 	return result;
>
>_
>  
>