| Message ID | 20250204133044.80551-6-bfoster@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | iomap: incremental per-operation iter advance | expand |
Looks good:
Reviewed-by: Christoph Hellwig <hch@lst.de>
On Tue, Feb 04, 2025 at 08:30:39AM -0500, Brian Foster wrote: > The iter termination logic in iomap_iter_advance() is only needed by > iomap_iter() to determine whether to proceed with the next mapping > for an ongoing operation. The old logic sets ret to 1 and then > terminates if the operation is complete (iter->len == 0) or the > previous iteration performed no work and the mapping has not been > marked stale. The stale check exists to allow operations to > retry the current mapping if an inconsistency has been detected. > > To further genericize iomap_iter_advance(), lift the termination > logic into iomap_iter() and update the former to return success (0) > or an error code. iomap_iter() continues on successful advance and > non-zero iter->len or otherwise terminates in the no progress (and > not stale) or error cases. > > Signed-off-by: Brian Foster <bfoster@redhat.com> > --- > fs/iomap/iter.c | 21 +++++++++++++-------- > 1 file changed, 13 insertions(+), 8 deletions(-) > > diff --git a/fs/iomap/iter.c b/fs/iomap/iter.c > index fcc8d75dd22f..04bd39ee5d47 100644 > --- a/fs/iomap/iter.c > +++ b/fs/iomap/iter.c > @@ -27,17 +27,11 @@ static inline void iomap_iter_reset_iomap(struct iomap_iter *iter) > */ > static inline int iomap_iter_advance(struct iomap_iter *iter, s64 count) > { > - bool stale = iter->iomap.flags & IOMAP_F_STALE; > - int ret = 1; > - > if (WARN_ON_ONCE(count > iomap_length(iter))) > return -EIO; > iter->pos += count; > iter->len -= count; > - if (!iter->len || (!count && !stale)) > - ret = 0; > - > - return ret; > + return 0; > } > > static inline void iomap_iter_done(struct iomap_iter *iter) > @@ -69,6 +63,7 @@ static inline void iomap_iter_done(struct iomap_iter *iter) > */ > int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops) > { > + bool stale = iter->iomap.flags & IOMAP_F_STALE; > int ret; > > trace_iomap_iter(iter, ops, _RET_IP_); > @@ -89,8 +84,18 @@ int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops) > return iter->processed; > } > > - /* advance and clear state from the previous iteration */ > + /* > + * Advance the iter and clear state from the previous iteration. Use > + * iter->len to determine whether to continue onto the next mapping. > + * Explicitly terminate in the case where the current iter has not > + * advanced at all (i.e. no work was done for some reason) unless the > + * mapping has been marked stale and needs to be reprocessed. > + */ > ret = iomap_iter_advance(iter, iter->processed); > + if (!ret && iter->len > 0) > + ret = 1; > + if (ret > 0 && !iter->processed && !stale) How can ret be greater than zero here other than the previous line setting it? I /think/ this is the same as: if (!ret && iter->len > 0) { if (iter->processed || stale) ret = 1; } but then I wonder if it's really necessary to reset the iter state on error, or if we've finished the whole thing, or if we've done no work and didn't set STALE? What do you think about: ret = iomap_iter_advance(...); if (ret || !iter->len) return ret; if (!iter->processed && !stale) return 0; iomap_iter_reset_iomap(iter); ret = ops->iomap_begin(...); --D > + ret = 0; > iomap_iter_reset_iomap(iter); > if (ret <= 0) > return ret; > -- > 2.48.1 > >
On Tue, Feb 04, 2025 at 11:52:20AM -0800, Darrick J. Wong wrote: > On Tue, Feb 04, 2025 at 08:30:39AM -0500, Brian Foster wrote: > > The iter termination logic in iomap_iter_advance() is only needed by > > iomap_iter() to determine whether to proceed with the next mapping > > for an ongoing operation. The old logic sets ret to 1 and then > > terminates if the operation is complete (iter->len == 0) or the > > previous iteration performed no work and the mapping has not been > > marked stale. The stale check exists to allow operations to > > retry the current mapping if an inconsistency has been detected. > > > > To further genericize iomap_iter_advance(), lift the termination > > logic into iomap_iter() and update the former to return success (0) > > or an error code. iomap_iter() continues on successful advance and > > non-zero iter->len or otherwise terminates in the no progress (and > > not stale) or error cases. > > > > Signed-off-by: Brian Foster <bfoster@redhat.com> > > --- > > fs/iomap/iter.c | 21 +++++++++++++-------- > > 1 file changed, 13 insertions(+), 8 deletions(-) > > > > diff --git a/fs/iomap/iter.c b/fs/iomap/iter.c > > index fcc8d75dd22f..04bd39ee5d47 100644 > > --- a/fs/iomap/iter.c > > +++ b/fs/iomap/iter.c ... > > @@ -89,8 +84,18 @@ int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops) > > return iter->processed; > > } > > > > - /* advance and clear state from the previous iteration */ > > + /* > > + * Advance the iter and clear state from the previous iteration. Use > > + * iter->len to determine whether to continue onto the next mapping. > > + * Explicitly terminate in the case where the current iter has not > > + * advanced at all (i.e. no work was done for some reason) unless the > > + * mapping has been marked stale and needs to be reprocessed. > > + */ > > ret = iomap_iter_advance(iter, iter->processed); > > + if (!ret && iter->len > 0) > > + ret = 1; > > + if (ret > 0 && !iter->processed && !stale) > > How can ret be greater than zero here other than the previous line > setting it? I /think/ this is the same as: > > if (!ret && iter->len > 0) { > if (iter->processed || stale) > ret = 1; > } > > but then I wonder if it's really necessary to reset the iter state on > error, or if we've finished the whole thing, or if we've done no work > and didn't set STALE? What do you think about: > > ret = iomap_iter_advance(...); > if (ret || !iter->len) > return ret; > if (!iter->processed && !stale) > return 0; > > iomap_iter_reset_iomap(iter); > ret = ops->iomap_begin(...); > Two thoughts come to mind... The first is that I'm really trying to be as incremental as possible here to minimize possibility of subtle issues that might come along with subtle changes (i.e. like when we do or don't reset, etc.). If you wanted to tack this on as a followon cleanup, I'd have no issue with it in general. But that leads to... The second is that I have a followup series to convert the rest of the iomap ops to use iter_advance(), after which the iomap_iter() advance in the logic above goes away. So it might be best to defer this sort of reset and continuation logic cleanup to after that.. Brian > --D > > > + ret = 0; > > iomap_iter_reset_iomap(iter); > > if (ret <= 0) > > return ret; > > -- > > 2.48.1 > > > > >
diff --git a/fs/iomap/iter.c b/fs/iomap/iter.c index fcc8d75dd22f..04bd39ee5d47 100644 --- a/fs/iomap/iter.c +++ b/fs/iomap/iter.c @@ -27,17 +27,11 @@ static inline void iomap_iter_reset_iomap(struct iomap_iter *iter) */ static inline int iomap_iter_advance(struct iomap_iter *iter, s64 count) { - bool stale = iter->iomap.flags & IOMAP_F_STALE; - int ret = 1; - if (WARN_ON_ONCE(count > iomap_length(iter))) return -EIO; iter->pos += count; iter->len -= count; - if (!iter->len || (!count && !stale)) - ret = 0; - - return ret; + return 0; } static inline void iomap_iter_done(struct iomap_iter *iter) @@ -69,6 +63,7 @@ static inline void iomap_iter_done(struct iomap_iter *iter) */ int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops) { + bool stale = iter->iomap.flags & IOMAP_F_STALE; int ret; trace_iomap_iter(iter, ops, _RET_IP_); @@ -89,8 +84,18 @@ int iomap_iter(struct iomap_iter *iter, const struct iomap_ops *ops) return iter->processed; } - /* advance and clear state from the previous iteration */ + /* + * Advance the iter and clear state from the previous iteration. Use + * iter->len to determine whether to continue onto the next mapping. + * Explicitly terminate in the case where the current iter has not + * advanced at all (i.e. no work was done for some reason) unless the + * mapping has been marked stale and needs to be reprocessed. + */ ret = iomap_iter_advance(iter, iter->processed); + if (!ret && iter->len > 0) + ret = 1; + if (ret > 0 && !iter->processed && !stale) + ret = 0; iomap_iter_reset_iomap(iter); if (ret <= 0) return ret;
The iter termination logic in iomap_iter_advance() is only needed by iomap_iter() to determine whether to proceed with the next mapping for an ongoing operation. The old logic sets ret to 1 and then terminates if the operation is complete (iter->len == 0) or the previous iteration performed no work and the mapping has not been marked stale. The stale check exists to allow operations to retry the current mapping if an inconsistency has been detected. To further genericize iomap_iter_advance(), lift the termination logic into iomap_iter() and update the former to return success (0) or an error code. iomap_iter() continues on successful advance and non-zero iter->len or otherwise terminates in the no progress (and not stale) or error cases. Signed-off-by: Brian Foster <bfoster@redhat.com> --- fs/iomap/iter.c | 21 +++++++++++++-------- 1 file changed, 13 insertions(+), 8 deletions(-)