On 3/8/06, Ali Ahmed Thawerani <aaht14@xxxxxxxxx> wrote: > i have written a wrapper for a block device. my wrapper is also > registered as a block device and i am obtaining the pointer bdev of > the wrapper block device by using the open_by_devnum function with > FMODE_READ | FMODE_WRITE > > i want to sync the memory pages that are in the page cache for this > particular wrapper block device > > what i am doing is using two functions filemap_fdatawrite and > filemap_fdatawait which take bdev->bd_inode->i_mapping as argument and > sync the data associated with a particular block device > > the return status of these two functions is 0 which i think means no success > > i have tried to figure out what can be the reason for this and i > followed the function calling that have been made > > some of the possible reasons were: when > i_mapping->backing_dev_info->memory_backed is 1 then these functions > return 0 i have checked that this value is not greater than 0 moreover > i have also checked that the i_mapping->nr_pages are also greater than > 0 > > i was not able to go any further > > i am calling these two functions in a thread > > there was also another problem that when there is some request coming > to my wrapper block device and wake up the thread which tries to sync > the page cache my system gets stuck may be both the threads are trying > to modify the inodes at the same time and none is succeeding > > what i did to solve this problem that i used kernel timer if there is > no request in the request queue for jiffies + 30 then i wake up the > thread to sync the buffer cache and at the same time set a > flag........ as according to my assumption when buffer cache will sync > data will come to my wrapper block device and i have to requeue it to > avoid any sort of problem but the thing is no such thing happens that > is there is no new request in request queue at all > > can any one help me out in this > i want to add a few more things that i have tried i looked at the function filemap_fdatawrite and followed the function calling at got to do_writepages which was in turn calling either the i_mapping->a_ops->writepages or generic_writepages with the parameters mapping and wbc i copied do_writepages as it is in my code and gave i_mapping in place of mapping and defined a structure of struct writeback_control with syncmode = WB_SYNC_ALL, start = 0, end = 0, nr_to_write = i_mapping->nr_pages * 2 and non_blocking = 1 the values of start and end are given 0 in __filemap_fdatawrite and syncmode = WB_SYNC_ALL is given in filemap_fdatawrite so i just gave the same values i have seen that i_mapping->a_ops->writepages is assigned generic_writepages and generic_writepages calls mpage_writepages -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
