| Message ID | 20240827151243.63493-1-pchelkin@ispras.ru (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | btrfs: qgroup: add missing extent changeset release | expand |
On Tue, Aug 27, 2024 at 06:12:43PM +0300, Fedor Pchelkin wrote: > The extent changeset may have some additional memory dynamically allocated > for ulist in result of clear_record_extent_bits() execution. This can happen, as clear_record_extent_bits adds more data to the changeset in some cases. What I don't see yet how it happens. An extent range must be split so that a new entry is added with different bits set. This is usual thing, but why does this happen with the quotas disabled.
On Tue, 27. Aug 18:03, David Sterba wrote: > On Tue, Aug 27, 2024 at 06:12:43PM +0300, Fedor Pchelkin wrote: > > The extent changeset may have some additional memory dynamically allocated > > for ulist in result of clear_record_extent_bits() execution. > > This can happen, as clear_record_extent_bits adds more data to the > changeset in some cases. What I don't see yet how it happens. An extent > range must be split so that a new entry is added with different bits > set. This is usual thing, but why does this happen with the quotas > disabled. In the reproducer case, qgroup_reserve_data() which sets the bits happens just before disabling the quotas via ioctl. Commit af0e2aab3b70 ("btrfs: qgroup: flush reservations during quota disable") added a call to clear_record_extent_bits() inside __btrfs_qgroup_release_data(). The changeset being passed is freshly initialized and empty. So the first call to clear_state_bit() there will definitely create a new entry and add it to the ulist. If for some reason clear_state_bit() shouldn't be eventually called then, to be honest, I don't quite understand why a call to clear_record_extent_bits() was added in the first place without expecting it to do the real work with clear_state_bit().
在 2024/8/28 00:42, Fedor Pchelkin 写道: > The extent changeset may have some additional memory dynamically allocated > for ulist in result of clear_record_extent_bits() execution. > > Release it after the local changeset is no longer needed in > BTRFS_QGROUP_MODE_DISABLED case. > > Found by Linux Verification Center (linuxtesting.org) with Syzkaller. > > Reported-by: syzbot+81670362c283f3dd889c@syzkaller.appspotmail.com > Closes: https://lore.kernel.org/lkml/000000000000aa8c0c060ade165e@google.com > Fixes: af0e2aab3b70 ("btrfs: qgroup: flush reservations during quota disable") > Cc: stable@vger.kernel.org # 6.10+ > Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> Reviewed-by: Qu Wenruo <wqu@suse.com> In this particular case, the changeset is really only locally utilized, thus should always be released. Thanks, Qu > --- > fs/btrfs/qgroup.c | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > > diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c > index 5d57a285d59b..4f1fa5d427e1 100644 > --- a/fs/btrfs/qgroup.c > +++ b/fs/btrfs/qgroup.c > @@ -4345,9 +4345,10 @@ static int __btrfs_qgroup_release_data(struct btrfs_inode *inode, > > if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED) { > extent_changeset_init(&changeset); > - return clear_record_extent_bits(&inode->io_tree, start, > - start + len - 1, > - EXTENT_QGROUP_RESERVED, &changeset); > + ret = clear_record_extent_bits(&inode->io_tree, start, > + start + len - 1, > + EXTENT_QGROUP_RESERVED, &changeset); > + goto out; > } > > /* In release case, we shouldn't have @reserved */
在 2024/8/28 18:54, Qu Wenruo 写道: > > > 在 2024/8/28 00:42, Fedor Pchelkin 写道: >> The extent changeset may have some additional memory dynamically >> allocated >> for ulist in result of clear_record_extent_bits() execution. >> >> Release it after the local changeset is no longer needed in >> BTRFS_QGROUP_MODE_DISABLED case. >> >> Found by Linux Verification Center (linuxtesting.org) with Syzkaller. >> >> Reported-by: syzbot+81670362c283f3dd889c@syzkaller.appspotmail.com >> Closes: >> https://lore.kernel.org/lkml/000000000000aa8c0c060ade165e@google.com >> Fixes: af0e2aab3b70 ("btrfs: qgroup: flush reservations during quota >> disable") >> Cc: stable@vger.kernel.org # 6.10+ >> Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> > > Reviewed-by: Qu Wenruo <wqu@suse.com> > > In this particular case, the changeset is really only locally utilized, > thus should always be released. My bad, after checking your latest reply to David, I think we can go one step further. Just do not pass changeset to clear_record_extent_bits(). A changeset is utilized for two reasons: - To let the caller know how many bytes are changed Just like what we did for the qgroup enabled case. - Allow the caller to revert its change This happens for qgroup_unreserve_range() when we hit an error and needs to free what we just reserved. In this particular case, since qgroup is already disabled, we just want to clear the extent io tree bits, not really bother how many bytes are released nor keep the info for reverting. So just pass NULL and everything should be fine. Thanks, Qu > > Thanks, > Qu >> --- >> fs/btrfs/qgroup.c | 7 ++++--- >> 1 file changed, 4 insertions(+), 3 deletions(-) >> >> diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c >> index 5d57a285d59b..4f1fa5d427e1 100644 >> --- a/fs/btrfs/qgroup.c >> +++ b/fs/btrfs/qgroup.c >> @@ -4345,9 +4345,10 @@ static int __btrfs_qgroup_release_data(struct >> btrfs_inode *inode, >> >> if (btrfs_qgroup_mode(inode->root->fs_info) == >> BTRFS_QGROUP_MODE_DISABLED) { >> extent_changeset_init(&changeset); >> - return clear_record_extent_bits(&inode->io_tree, start, >> - start + len - 1, >> - EXTENT_QGROUP_RESERVED, &changeset); >> + ret = clear_record_extent_bits(&inode->io_tree, start, >> + start + len - 1, >> + EXTENT_QGROUP_RESERVED, &changeset); >> + goto out; >> } >> >> /* In release case, we shouldn't have @reserved */ >
diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c index 5d57a285d59b..4f1fa5d427e1 100644 --- a/fs/btrfs/qgroup.c +++ b/fs/btrfs/qgroup.c @@ -4345,9 +4345,10 @@ static int __btrfs_qgroup_release_data(struct btrfs_inode *inode, if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED) { extent_changeset_init(&changeset); - return clear_record_extent_bits(&inode->io_tree, start, - start + len - 1, - EXTENT_QGROUP_RESERVED, &changeset); + ret = clear_record_extent_bits(&inode->io_tree, start, + start + len - 1, + EXTENT_QGROUP_RESERVED, &changeset); + goto out; } /* In release case, we shouldn't have @reserved */
The extent changeset may have some additional memory dynamically allocated for ulist in result of clear_record_extent_bits() execution. Release it after the local changeset is no longer needed in BTRFS_QGROUP_MODE_DISABLED case. Found by Linux Verification Center (linuxtesting.org) with Syzkaller. Reported-by: syzbot+81670362c283f3dd889c@syzkaller.appspotmail.com Closes: https://lore.kernel.org/lkml/000000000000aa8c0c060ade165e@google.com Fixes: af0e2aab3b70 ("btrfs: qgroup: flush reservations during quota disable") Cc: stable@vger.kernel.org # 6.10+ Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru> --- fs/btrfs/qgroup.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)