deadlock encountered for my network file system

deadlock encountered for my network file system

[Rongfeng Tang]

[-- Attachment #1: Type: text/plain, Size: 915 bytes --]

Hi all,

	I'm encountering a deadlock problem within kernel(2.4.19smp):
    
    A lot of data writen into a network file system and the physical memory is almost exhausted, so that the kswapd calls the kswapd_balance -> try_to_free_pages->...->shrink_cache try to free a dirty page(just name it PageA) of the file system. the file system's writepage locks the page and sends the data to a dedicated process (named ProcB)of this file system in the same node which just send the data to remote server. ProcB communicates with server by TCP/IP. when the TCP/IP layer try to alloc a free page for transmission, it also arouse the kernel to free some dirty pages, what's a pity, the PageA has been selected. Since it's locked now , the shrink_cache just calls on wait_on_page, thus a big deadlock is generated.
   Is it a bug? if not, please give me some advice on how to just out of this cycle. 


Thanks

Rongfeng
  

[-- Attachment #2: fox.gif --]
[-- Type: image/gif, Size: 9519 bytes --]