On Thu, 8 Nov 2012, Lukáš Czerner wrote: > Date: Thu, 8 Nov 2012 14:43:19 +0100 (CET) > From: Lukáš Czerner <lczerner@xxxxxxxxxx> > To: Dmitry Monakhov <dmonakhov@xxxxxxxxxx> > Cc: Lukas Czerner <lczerner@xxxxxxxxxx>, linux-ext4@xxxxxxxxxxxxxxx, > tytso@xxxxxxx > Subject: Re: [PATCH] ext4: Prevent race while waling extent tree > > On Thu, 8 Nov 2012, Dmitry Monakhov wrote: > > > Date: Thu, 08 Nov 2012 16:01:17 +0400 > > From: Dmitry Monakhov <dmonakhov@xxxxxxxxxx> > > To: Lukas Czerner <lczerner@xxxxxxxxxx>, linux-ext4@xxxxxxxxxxxxxxx > > Cc: tytso@xxxxxxx, Lukas Czerner <lczerner@xxxxxxxxxx> > > Subject: Re: [PATCH] ext4: Prevent race while waling extent tree > > > > On Thu, 8 Nov 2012 12:08:49 +0100, Lukas Czerner <lczerner@xxxxxxxxxx> wrote: > > > Currently ext4_ext_walk_space() only takes i_data_sem for read when > > > searching for the extent at given block with ext4_ext_find_extent(). > > > Then it drops the lock and the extent tree can be changed at will. > > > However later on we're searching for the 'next' extent, but the extent > > > tree might already have changed, so the information might not be > > > accurate. > > > > > > In fact we can hit BUG_ON(end <= start) if the extent got inserted into > > > the tree after the one we found and before the block we were searching > > > for. This has been reproduced by running xfstests 225 in loop on s390x > > > architecture, but theoretically we could hit this on any other > > > architecture as well, but probably not as often. > > > > > > ext4_ext_walk_space() is currently only used from ext4_fiemap() and even > > > if we do not hit the BUG_ON() fiemap might return scrambled information > > > to the user. > > > > > > Fix this by requiring ext4_ext_walk_space() to be called with i_data_sem > > > held. By calling it from ext4_fiemap() we can only take the i_data_sem > > > for read, but possibly other users might want to modify the extents so > > > they will be able to take write lock. > > Agree as a short term fix for BUGON case, but Theodore suggested to use > > seqlock approach http://lists.openwall.net/linux-ext4/2011/10/26/25 > > Yeah, it make sense to protect us from fiemap abuse, however using > seqlock for walking the extent tree seems like an overkill > especially considering how much work will that require. We would > have to make sure that everything we do in the ext4_ext_walk_space() > and other function we're calling there is safe even if the extent > tree change under our hands. I do not think this is the right way. > > I was thinking about checking for contentions on the semaphore from > within the ext4_ext_walk_space() - possibly enabling/disabling it > with a function parameter ? > > Sadly kernel does not provide a helper to check for that so what > about something like this in the beginning of the while loop in > ext4_ext_walk_space ? > > if (check_contention) { > int contends = 0; > unsigned int flags; > > raw_spin_lock_irqsave(&EXT4_I(inode)->i_data_sem->wait_lock, flags); > if (!list_empty(&EXT4_I(inode)->i_data_sem->wait_list) > contends = 1 > raw_spin_unlock_irqrestore(&EXT4_I(inode)->i_data_sem->wait_lock, flags); > > if (contends) > break > } > > or we can add the helper to the rwsem code and use that. > > > What do you think ? Nevermind, trhere is no generic way to tell how many waiters for the semaphore there is... -Lukas > > Thanks! > -Lukas > > > > > > > > > Signed-off-by: Lukas Czerner <lczerner@xxxxxxxxxx> > > > --- > > > fs/ext4/extents.c | 9 +++++++-- > > > 1 files changed, 7 insertions(+), 2 deletions(-) > > > > > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c > > > index 7011ac9..f1aca06 100644 > > > --- a/fs/ext4/extents.c > > > +++ b/fs/ext4/extents.c > > > @@ -1959,6 +1959,11 @@ cleanup: > > > return err; > > > } > > > > > > +/* > > > + * ext4_ext_walk_space() should be called with i_data_sem locked. If we're > > > + * not modifying found extents, or extent tree in callback function, then > > > + * read lock is ok. > > > + */ > > > static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, > > > ext4_lblk_t num, ext_prepare_callback func, > > > void *cbdata) > > > @@ -1976,9 +1981,7 @@ static int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, > > > while (block < last && block != EXT_MAX_BLOCKS) { > > > num = last - block; > > > /* find extent for this block */ > > > - down_read(&EXT4_I(inode)->i_data_sem); > > > path = ext4_ext_find_extent(inode, block, path); > > > - up_read(&EXT4_I(inode)->i_data_sem); > > > if (IS_ERR(path)) { > > > err = PTR_ERR(path); > > > path = NULL; > > > @@ -5021,8 +5024,10 @@ int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, > > > * Walk the extent tree gathering extent information. > > > * ext4_ext_fiemap_cb will push extents back to user. > > > */ > > > + down_read(&EXT4_I(inode)->i_data_sem); > > > error = ext4_ext_walk_space(inode, start_blk, len_blks, > > > ext4_ext_fiemap_cb, fieinfo); > > > + up_read(&EXT4_I(inode)->i_data_sem); > > > } > > > > > > return error; > > > -- > > > 1.7.7.6 > > > > > > -- > > > 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 > > >
