On Tue, Sep 04, 2012 at 11:37:13AM -0400, Lukáš Czerner wrote: > On Tue, 4 Sep 2012, J. Bruce Fields wrote: > > > Date: Tue, 4 Sep 2012 10:52:13 -0400 > > From: J. Bruce Fields <bfields@xxxxxxxxxxxx> > > To: Lukas Czerner <lczerner@xxxxxxxxxx> > > Cc: linux-fsdevel@xxxxxxxxxxxxxxx, linux-ext4@xxxxxxxxxxxxxxx, tytso@xxxxxxx, > > hughd@xxxxxxxxxx, linux-mm@xxxxxxxxx > > Subject: Re: [PATCH 02/15 v2] jbd2: implement > > jbd2_journal_invalidatepage_range > > > > On Fri, Aug 31, 2012 at 06:21:38PM -0400, Lukas Czerner wrote: > > > mm now supports invalidatepage_range address space operation and there > > > are two file system using jbd2 also implementing punch hole feature > > > which can benefit from this. We need to implement the same thing for > > > jbd2 layer in order to allow those file system take benefit of this > > > functionality. > > > > > > With new function jbd2_journal_invalidatepage_range() we can now specify > > > length to invalidate, rather than assuming invalidate to the end of the > > > page. > > > > > > Signed-off-by: Lukas Czerner <lczerner@xxxxxxxxxx> > > > --- > > > fs/jbd2/journal.c | 1 + > > > fs/jbd2/transaction.c | 19 +++++++++++++++++-- > > > include/linux/jbd2.h | 2 ++ > > > 3 files changed, 20 insertions(+), 2 deletions(-) > > > > > > diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c > > > index e149b99..e4618e9 100644 > > > --- a/fs/jbd2/journal.c > > > +++ b/fs/jbd2/journal.c > > > @@ -86,6 +86,7 @@ EXPORT_SYMBOL(jbd2_journal_force_commit_nested); > > > EXPORT_SYMBOL(jbd2_journal_wipe); > > > EXPORT_SYMBOL(jbd2_journal_blocks_per_page); > > > EXPORT_SYMBOL(jbd2_journal_invalidatepage); > > > +EXPORT_SYMBOL(jbd2_journal_invalidatepage_range); > > > EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); > > > EXPORT_SYMBOL(jbd2_journal_force_commit); > > > EXPORT_SYMBOL(jbd2_journal_file_inode); > > > diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c > > > index fb1ab953..65c1374 100644 > > > --- a/fs/jbd2/transaction.c > > > +++ b/fs/jbd2/transaction.c > > > @@ -1993,10 +1993,20 @@ zap_buffer_unlocked: > > > * > > > */ > > > void jbd2_journal_invalidatepage(journal_t *journal, > > > - struct page *page, > > > - unsigned long offset) > > > + struct page *page, > > > + unsigned long offset) > > > +{ > > > + jbd2_journal_invalidatepage_range(journal, page, offset, > > > + PAGE_CACHE_SIZE - offset); > > > +} > > > + > > > +void jbd2_journal_invalidatepage_range(journal_t *journal, > > > + struct page *page, > > > + unsigned int offset, > > > + unsigned int length) > > > { > > > struct buffer_head *head, *bh, *next; > > > + unsigned int stop = offset + length; > > > unsigned int curr_off = 0; > > > int may_free = 1; > > > > > > @@ -2005,6 +2015,8 @@ void jbd2_journal_invalidatepage(journal_t *journal, > > > if (!page_has_buffers(page)) > > > return; > > > > > > + BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); > > > > This misses e.g. length == (unsigned int)(-1), offset = 1. Could make > > it obvious with: > > Hmm.. So if length = -1 (e.g. UINT_MAX) and offset = 1 then: > > offset + length = 0 > > so > > length is bigger than (offset + length) right ? Speaking in numbers: > > length = 4294967295 > offset = 1 > stop = length + offset = 0 > > so (0 < 4294967295) is true and we'll BUG() on this, right ? > > Am I missing something ? Gah, no, I just wasn't thinking straight: the only way offset or length could individually be greater than PAGE_CACHE_SIZE while their sum is less would be if their sum overflows, in which case the second condition (stop < length) would trigger. So the two conditions are enough. --b. -- 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