[Gluster-devel] Why to fork so many times?

[Pavan T C]

On Tuesday 07 June 2011 02:08 PM, Daniel wrote:
> Thank you, Pavan. It clarifies most of my doubts.
>
> Only I can't find the branch of 'if sync-mtab == 1'. And I thought it
> was without "YES fork()". Could you explain more about that, please?
>

Oh, the variable mtab_pid is what indicates whether the sync-mtab option 
is set or not. Trace back the caller of fuse_mnt_add_mount(). The call 
stack looks like below;

fuse_bridge.c:init() -> gf_fuse_mount() -> fuse_mount_sys() -> 
fuse_mnt_add_mount().

In gf_fuse_mount(), the last parameter is passed as :
sync_mtab ? &ctx->mtab_pid : NULL
So, if the "sync-mtab" option is set, mtab_pid will be non-NULL. 
Otherwise it will be NULL.

Now you can relate back to the condition in fuse_mnt_add_mount().
If mtab_pid is NULL, the "NO" condition in the flow-chart I provided 
comes into picture. Otherwise the code follows the YES condition.

Pavan

> Thanks a lot.
>
> Dan
>
> On Tue, 2011-06-07 at 12:29 +0530, Pavan T C wrote:
>> On Tuesday 07 June 2011 10:54 AM, Daniel wrote:
>>> Hello hackers:
>>>
>>> I am learning glusterfs and get one question when I reading
>>> contrib/fuse-lib/mount.c fuse_mnt_add_mount(). Why does it fork four
>>> times and the grand-grand-child exec("/bin/mount")? What is the benefit
>>> and the consideration?
>>
>> As I see it, it forks thrice. Here is what I understand from the code:
>>
>> +---------------------+
>> | fuse_mnt_add_mount()|<------------------------------+
>> | fork and wait for   |                               |
>> | child to return     |                               |
>> +-----+---------------+                               |
>>         |                                               |
>>         |                                               |
>>         X                                               |
>>        / \                                              |
>>       /if \                                             |
>>      /sync-\    NO    +---------------------------+     |
>>     X mtab  X-------->| Do async ops by returning |     |
>>      \ == 1/   fork() | from here so that parent  |     |
>>       \   /           | does not wait. But before |     |
>>        \ /            | doing that, create a dummy|     |
>>         X             | parent to wait for child  |     |
>>         | YES         | that does the /bin/mount  |     |
>>         |             +------------+--------------+     |
>>         | fork()                   |                    |
>>         V                          |                    |
>> +-----+----------+               | fork() ->  Parent returns here.
>> | Child: execl   |               v
>> | /bin/mount with|         +---------------+
>> | -f option to   |         |dummy parent:  |
>> | only populate  |  fork() | fork and wait |
>> | mtab but don't |<--------+ for /bin/mount|
>> | actually mount |         | to return     |
>> | anything.      |         +---------------+
>> +----------------+
>>
>> HTH,
>> Pavan
>>
>>>
>>> Thank you
>>>
>>> Daniel
>>>
>>>
>>> _______________________________________________
>>> Gluster-devel mailing list
>>> Gluster-devel at nongnu.org
>>> https://lists.nongnu.org/mailman/listinfo/gluster-devel
>>
>
>