On Thu, 23 Jul 2009, Andi Kleen wrote: > Sage Weil <sage@xxxxxxxxxxxx> writes: > > > The ceph address space methods are concerned primarily with managing > > the dirty page accounting in the inode, which (among other things) > > must keep track of which snapshot context each page was dirtied in, > > and ensure that dirty data is written out to the OSDs in snapshort > > order. > > > > A writepage() on a page that is not currently writeable due to > > snapshot writeback ordering constraints is ignored (it was presumably > > called from kswapd). > > Not a detailed review. You would need to get one from someone who > knows the VFS interfaces very well (unfortunately those people are hard > to find). I just read through it. > > One thing I noticed is that you seem to do a lot of memory allocation > in the write out paths (some of it even GFP_KERNEL, not GFP_NOFS) I fixed the bad GFP_KERNEL/uncheck kmalloc. Which was just a struct pagevec, which we added a while ago to reduce stack usage. Maybe the stack is a better place for it though? (cifs puts it on the stack...) > There were some changes to make this problem less severe (e.g. better > dirty pages accounting), but I don't think anyone has really declared > it solved yet. The standard workaround for this is to use mempools > for anything allocated in the writeout path, then you are at least > guaranteed to make forward progress. There are two other memory allocations during writeout: a vector of pages to be written, and the message we're sending to the OSD. If I use a mempool for those to guarantee as least some writeout will occur, how do I safely defer when allocations do fail? Will pdflush (or it's replacement) eventually come back and try ->writepages() again? Thanks- sage -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
