Re: [OpenWrt-Devel] fstools: factory reset on power failure during startup

2015-11-30 Thread Florian Eckert 
Hello,

i think the patch fixes the issue.
We could reproduce a factory reset with the dc/dc test very often
befor we apply your patch.
But now i think there is an other problem with ubifs! See log

The system fails at Saturday evening with this error

Press the [f] key and hit [enter] to enter failsafe mode
Press the [1], [2], [3] or [4] key and hit [enter] to select the debug level
[7.184000] random: nonblocking pool is initialized
[   10.236000] mount_root: Boot cmd:
[   10.264000] UBIFS: background thread "ubifs_bgt0_1" started, PID 302
[   10.288000] UBIFS: recovery needed
[   10.292000] UBIFS error (pid 300): check_lpt_type: invalid type
(15) in LPT node type 2
[   10.30] CPU: 0 PID: 300 Comm: mount_root Not tainted 3.18.23 #37
[   10.304000] Stack :     804e45f2
0038  
  87bb9350 0001 8041316c 80490b43 012c 804e382c 87bb9350 0001
  87013db8 87013dbc c01bd000 80055e54 0006 8002d3e8  
  804179e8 87013d2c      
         
  ...
[   10.34] Call Trace:
[   10.344000] [<800199a8>] show_stack+0x50/0x84
[   10.348000] [<80161b88>] check_lpt_type.constprop.9+0x4c/0x60
[   10.352000] [<80163bc0>] ubifs_lpt_init+0x148/0x4e8
[   10.36] [<8014adc8>] ubifs_mount+0xfec/0x1794
[   10.364000] [<800cba38>] mount_fs+0x20/0xdc
[   10.368000] [<800e4634>] vfs_kern_mount+0x64/0x138
[   10.372000] [<800e7f74>] do_mount+0x95c/0xb20
[   10.376000] [<800e8478>] SyS_mount+0x90/0xd4
[   10.38] [<8000875c>] handle_sys+0x11c/0x140
[   10.384000]
[   10.388000] UBIFS: background thread "ubifs_bgt0_1" stops
[   10.392000] mount_root: failed to mount -t ubifs /dev/ubi0_1
/tmp/overlay: Invalid argument
[   10.40] mount_root: TDT readlink return value of file
/tmp/overlay/.fs_state is -1
[   10.408000] mount_root: TDT readlink error string is ** No such
file or directory **
[   10.416000] mount_root: TDT actual filystem state is 0
[   10.42] mount_root: TDT readlink return value of file
/tmp/overlay/.fs_state is -1
[   10.428000] mount_root: TDT readlink error string is ** No such
file or directory **
[   10.436000] mount_root: TDT readlink return value of file
/tmp/overlay/.fs_state is 1
[   10.444000] mount_root: overlay filesystem has not been fully initialized yet
[   10.452000] mount_root: switching to jffs2 overlay
[   10.456000] mount_root: TDT olddir /tmp/overlay failed: No such
file or directory
[   10.464000] mount_root: switching to jffs2 failed - fallback to ramoverlay
[   10.472000] mount_root: failed /rom/overlay /overlay: Invalid argument
[   10.488000] procd: - early -
[   10.492000] procd: - watchdog -
[   11.212000] procd: - ubus -
[   12.22] procd: - init -
Please press Enter to activate this console.
#

I dont know if you are the right persons?
I think it has do to with the ubifs layer and not with your fstool.

Flo


2015-11-27 13:17 GMT+01:00 Florian Eckert :
> Hi,
>
> I have tested your patch since last day when I got the patch from you.
> I am using an programmable dc/dc power supply, who switches the power
> between startup randomly.
> I will run the test during the weekend. And then I will see what
> happend on Monday morning.
> But for now i havn't lost the configuration anymore.
>
> Flo
>
>
> 2015-11-26 14:25 GMT+01:00 Florian Eckert :
>> Hi,
>>
>> thank your very much for your fast reply and the fix :-).
>> I will try it and give you an respond.
>>
>> flo
>>
>> 2015-11-26 14:05 GMT+01:00 Felix Fietkau :
>>> On 2015-11-26 09:24, Florian Eckert wrote:
 Hi,

 I am using openwrt CC and an ubifs filesystem on top of ubi.
 If I plug off the power during startup, some times my configuration
 get lost and i am back to factory reset.

 In the package fstools there is the bin "mount_root" it is called
 twice during startup.
 1. S10boot at line 10
 2. S95done at line 6

 this calles edit the ./fs_state link

 S10boot/S95done:
 ---
 Because ubifs is doing filesystem handling in a background process
 "ubi_bgt0d", could it be possible,
 that (see code) ./.fs_state is not rewritten written anymore?
 So that at next boot the system thinks he is in upgrade mode?

 If I add a sleep(10) between unlink and symlink function call i can
 reproduce the failure if i unplug the power.
 I think the first think what should be done is to call
 "snprintf(valstr, sizeof(valstr), "%d", state);" before "unlink(path).

 Or this this a deeper Problem?
>>> I think you're right that there's a race here. When I wrote this code,
>>> I never actually intended for the file to get re-written after the
>>> first boot, it seems that I simply forgot to add the necessary check
>>> for it.
>>> Please

Re: [OpenWrt-Devel] fstools: factory reset on power failure during startup

2015-11-27 Thread Florian Eckert 
Hi,

I have tested your patch since last day when I got the patch from you.
I am using an programmable dc/dc power supply, who switches the power
between startup randomly.
I will run the test during the weekend. And then I will see what
happend on Monday morning.
But for now i havn't lost the configuration anymore.

Flo


2015-11-26 14:25 GMT+01:00 Florian Eckert :
> Hi,
>
> thank your very much for your fast reply and the fix :-).
> I will try it and give you an respond.
>
> flo
>
> 2015-11-26 14:05 GMT+01:00 Felix Fietkau :
>> On 2015-11-26 09:24, Florian Eckert wrote:
>>> Hi,
>>>
>>> I am using openwrt CC and an ubifs filesystem on top of ubi.
>>> If I plug off the power during startup, some times my configuration
>>> get lost and i am back to factory reset.
>>>
>>> In the package fstools there is the bin "mount_root" it is called
>>> twice during startup.
>>> 1. S10boot at line 10
>>> 2. S95done at line 6
>>>
>>> this calles edit the ./fs_state link
>>>
>>> S10boot/S95done:
>>> ---
>>> Because ubifs is doing filesystem handling in a background process
>>> "ubi_bgt0d", could it be possible,
>>> that (see code) ./.fs_state is not rewritten written anymore?
>>> So that at next boot the system thinks he is in upgrade mode?
>>>
>>> If I add a sleep(10) between unlink and symlink function call i can
>>> reproduce the failure if i unplug the power.
>>> I think the first think what should be done is to call
>>> "snprintf(valstr, sizeof(valstr), "%d", state);" before "unlink(path).
>>>
>>> Or this this a deeper Problem?
>> I think you're right that there's a race here. When I wrote this code,
>> I never actually intended for the file to get re-written after the
>> first boot, it seems that I simply forgot to add the necessary check
>> for it.
>> Please test if this fstools patch fixes the problem:
>> ---
>> --- a/libfstools/overlay.c
>> +++ b/libfstools/overlay.c
>> @@ -311,6 +311,9 @@ int fs_state_set(const char *dir, enum fs_state state)
>> char valstr[16];
>> char *path;
>>
>> +   if (fs_state_get(dir) == state)
>> +   return 0;
>> +
>> path = alloca(strlen(dir) + 1 + sizeof("/.fs_state"));
>> sprintf(path, "%s/.fs_state", dir);
>> unlink(path);
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel

[OpenWrt-Devel] fstools: factory reset on power failure during startup

2015-11-26 Thread Florian Eckert 
Hi,

I am using openwrt CC and an ubifs filesystem on top of ubi.
If I plug off the power during startup, some times my configuration
get lost and i am back to factory reset.

In the package fstools there is the bin "mount_root" it is called
twice during startup.
1. S10boot at line 10
2. S95done at line 6

this calles edit the ./fs_state link

S10boot/S95done:
---
Because ubifs is doing filesystem handling in a background process
"ubi_bgt0d", could it be possible,
that (see code) ./.fs_state is not rewritten written anymore?
So that at next boot the system thinks he is in upgrade mode?

If I add a sleep(10) between unlink and symlink function call i can
reproduce the failure if i unplug the power.
I think the first think what should be done is to call
"snprintf(valstr, sizeof(valstr), "%d", state);" before "unlink(path).

Or this this a deeper Problem?

int fs_state_set(const char *dir, enum fs_state state)
{
char valstr[16];
char *path;

->path = alloca(strlen(dir) + 1 + sizeof("/.fs_state"));
->sprintf(path, "%s/.fs_state", dir);
->unlink(path);
->snprintf(valstr, sizeof(valstr), "%d", state);

return symlink(valstr, path);
}

Regards
Flo
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel

Re: [OpenWrt-Devel] fstools: factory reset on power failure during startup

2015-11-26 Thread John Crispin 


On 26/11/2015 09:24, Florian Eckert wrote:
> Hi,
> 
> I am using openwrt CC and an ubifs filesystem on top of ubi.
> If I plug off the power during startup, some times my configuration
> get lost and i am back to factory reset.
> 
> In the package fstools there is the bin "mount_root" it is called
> twice during startup.
> 1. S10boot at line 10
> 2. S95done at line 6
> 
> this calles edit the ./fs_state link
> 
> S10boot/S95done:
> ---
> Because ubifs is doing filesystem handling in a background process
> "ubi_bgt0d", could it be possible,
> that (see code) ./.fs_state is not rewritten written anymore?
> So that at next boot the system thinks he is in upgrade mode?
> 
> If I add a sleep(10) between unlink and symlink function call i can
> reproduce the failure if i unplug the power.
> I think the first think what should be done is to call
> "snprintf(valstr, sizeof(valstr), "%d", state);" before "unlink(path).
> 
> Or this this a deeper Problem?

interesting bug. felix recently added this "i crashed during firstboot,
lets retry" feature recently but it seesm that this works well on
classical (spi-)nor flash but fails on nand/ubi.

i've made a note and will look at this, however i am porting a new
target today so might not happen till start of the week. (unless felix
has time to look at it beforehand)

John

> 
> int fs_state_set(const char *dir, enum fs_state state)
> {
> char valstr[16];
> char *path;
> 
> ->path = alloca(strlen(dir) + 1 + sizeof("/.fs_state"));
> ->sprintf(path, "%s/.fs_state", dir);
> ->unlink(path);
> ->snprintf(valstr, sizeof(valstr), "%d", state);
> 
> return symlink(valstr, path);
> }
> 
> Regards
> Flo
> ___
> openwrt-devel mailing list
> openwrt-devel@lists.openwrt.org
> https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
> 
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel

Re: [OpenWrt-Devel] fstools: factory reset on power failure during startup

2015-11-26 Thread Felix Fietkau 
On 2015-11-26 09:24, Florian Eckert wrote:
> Hi,
> 
> I am using openwrt CC and an ubifs filesystem on top of ubi.
> If I plug off the power during startup, some times my configuration
> get lost and i am back to factory reset.
> 
> In the package fstools there is the bin "mount_root" it is called
> twice during startup.
> 1. S10boot at line 10
> 2. S95done at line 6
> 
> this calles edit the ./fs_state link
> 
> S10boot/S95done:
> ---
> Because ubifs is doing filesystem handling in a background process
> "ubi_bgt0d", could it be possible,
> that (see code) ./.fs_state is not rewritten written anymore?
> So that at next boot the system thinks he is in upgrade mode?
> 
> If I add a sleep(10) between unlink and symlink function call i can
> reproduce the failure if i unplug the power.
> I think the first think what should be done is to call
> "snprintf(valstr, sizeof(valstr), "%d", state);" before "unlink(path).
> 
> Or this this a deeper Problem?
I think you're right that there's a race here. When I wrote this code,
I never actually intended for the file to get re-written after the
first boot, it seems that I simply forgot to add the necessary check
for it.
Please test if this fstools patch fixes the problem:
---
--- a/libfstools/overlay.c
+++ b/libfstools/overlay.c
@@ -311,6 +311,9 @@ int fs_state_set(const char *dir, enum fs_state state)
char valstr[16];
char *path;
 
+   if (fs_state_get(dir) == state)
+   return 0;
+
path = alloca(strlen(dir) + 1 + sizeof("/.fs_state"));
sprintf(path, "%s/.fs_state", dir);
unlink(path);
___
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel