On 22/01/06 04:59PM, harshad shirwadkar wrote: > First of all thanks for catching this. Yeah, I think the right thing > to do here is to return the return value up to the caller. Also, I > agree with Lukas, we should only set fc_modified_inodes_size if the > allocation succeeds. Luo, would you be okay updating the patch to > include these changes? > > Thanks, > Harshad > > On Thu, Jan 6, 2022 at 2:58 AM Lukas Czerner <lczerner@xxxxxxxxxx> wrote: > > > > On Wed, Jan 05, 2022 at 11:44:39PM -0500, Theodore Ts'o wrote: > > > On Thu, Dec 30, 2021 at 06:29:05AM +0000, cgel.zte@xxxxxxxxx wrote: > > > > From: luo penghao <luo.penghao@xxxxxxxxxx> > > > > > > > > The assignments in these two places will be overwritten by new > > > > assignments later, so they should be deleted. > > > > > > > > The clang_analyzer complains as follows: > > > > > > > > fs/ext4/fast_commit.c > > > > > > > > Value stored to 'ret' is never read > > > Since I was also suspecting a similar issue in ext4_fc_record_modified_inode() (w.r.t. krealloc()) while doing some code reviews a while ago. And I also happened to stumble upon this discussion which added some more context to it. @Luo, I am preparing some other fixes and might submit this fix also as part of those. I am completely ok, if you would like to push a patch from your end based on this discussion. In that case, I will request to drop my patch or won't even publish it, if you submit it before my fixes gets out. -ritesh > > > I suspect the right answer here is that we *should* be checking the > > > return value, and reflecting the error up to caller, if appropriate. > > > > > > Harshad, what do you think? > > > > Indeed we absolutely *must* be checking the return value and bail out > > otherwise we risk overwriting kernel memory among other possible > > problems. > > > > See ext4_fc_record_modified_inode() where we increment > > fc_modified_inodes_size before the actual reallocation which in case of > > allocation failure will leave us with elevated fc_modified_inodes_size > > and the next call to ext4_fc_record_modified_inode() can modify > > fc_modified_inodes[] out of bounds. > > > > In addition to checking the return value we should probably also move > > incrementing the fc_modified_inodes_size until after the successful > > reallocation in order to avoid such pitfalls. > > > > Thanks! > > -Lukas > > > > > > > > - Ted > > > > > > > > > > > Reported-by: Zeal Robot <zealci@xxxxxxxxxx> > > > > Signed-off-by: luo penghao <luo.penghao@xxxxxxxxxx> > > > > --- > > > > fs/ext4/fast_commit.c | 4 ++-- > > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c > > > > index 8ea5a81..8d5d044 100644 > > > > --- a/fs/ext4/fast_commit.c > > > > +++ b/fs/ext4/fast_commit.c > > > > @@ -1660,7 +1660,7 @@ static int ext4_fc_replay_add_range(struct super_block *sb, > > > > return 0; > > > > } > > > > > > > > - ret = ext4_fc_record_modified_inode(sb, inode->i_ino); > > > > + ext4_fc_record_modified_inode(sb, inode->i_ino); > > > > > > > > start = le32_to_cpu(ex->ee_block); > > > > start_pblk = ext4_ext_pblock(ex); > > > > @@ -1785,7 +1785,7 @@ ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl, > > > > return 0; > > > > } > > > > > > > > - ret = ext4_fc_record_modified_inode(sb, inode->i_ino); > > > > + ext4_fc_record_modified_inode(sb, inode->i_ino); > > > > > > > > jbd_debug(1, "DEL_RANGE, inode %ld, lblk %d, len %d\n", > > > > inode->i_ino, le32_to_cpu(lrange.fc_lblk), > > > > -- > > > > 2.15.2 > > > > > > > > > > > > >
