On Wed, May 11, 2011 at 12:43 PM, Jan Kara <jack@xxxxxxx> wrote: > Hi Amir, > > On Wed 11-05-11 10:33:17, Amir G. wrote: >> Can you provide a bit of a wider scope review of how this related to the >> work on stable pages. > It is related in the sense that Darrick had to return locked page from > ext4_page_mkwrite() to avoid races and the result (although working) is > really like scratching your left ear with your right hand (not sure if > English has this idiom ;). So with my patches, what Darrick needs to do is > simply achieved by updating block_page_mkwrite() (__block_page_mkwrite() in > fact). > >> For example, when to the pages get unlocked? > That's handled by mm code in mm/memory.c. You can read do_wp_page() but > it's not a light reading ;) This is why I asked for the cheat sheet ;-) > >> If the pages supposed to be stable during writeback, how is this related >> to returning locked pages from page_mkwrite? > Returning locked page is needed to avoid races with page writeback - for > stable pages we want to do wait_on_page_writeback() to be sure that there's > no IO happening while we make the page writeable for user. But once we > release page lock, writepage can come and start the writeback. The basic > scheme of the race we want to avoid is: > do_wp_page() > ext4_page_mkwrite() > wait_on_page_writeback() > unlock_page() > clear_page_dirty_for_io() > page_mkclean() > writepage() > maybe_mkwrite() > > Because this results in writeable page under writeback... We have to make > sure page_mkclean() happens *after* maybe_mkwrite() from do_wp_page() in > this scenario. > >> Is the page going to stay locked until writeback? > No, until the page fault is finished. > >> Do I understand correctly that a page will be marked read-only after >> writeback completes, so page_mkwrite will be called again on next write? > Page is marked read-only before writeback is started in > clear_page_dirty_for_io(). OK. that makes sense. I wasn't sure my move-on-write hook, which is called from ext4_page_mkwrite() is going to be called on every writeback. The use case is: data = mmap(file); data[0] = 1; take snapshot 1; data[0] = 2; take snapshot 2; data[0] = 3; ... That use case doesn't specify when writeback of first page of file happens and I don't care which version of data[0] is seen by the snapshots, as long as the snapshot view is internally consistent (it doesn't change over time or after reboot). So if ext4_page_mkwrite() is called after every writeback and the page is stable during writeback, snapshots should be OK. > > Honza > -- > Jan Kara <jack@xxxxxxx> > SUSE Labs, CR > -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
