| Message ID | 20220527155036.524743-1-willy@infradead.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | Begin removing PageError | expand |
On Fri, May 27, 2022 at 04:50:12PM +0100, Matthew Wilcox (Oracle) wrote: > After this series, there are some places which still use it. Lots of > the converted places are trivial -- they should have been checking the > uptodate flag. The trickiest are the ones which have multiple steps > and signal "hey, something went wrong with one step in the read of this > page/folio" by setting the error flag. How many of those do we have left, and how many of those can't be changed to just return them through errors or other on-stack means? > I'm thinking about (ab)using PageChecked == PG_owner_priv_1 for this > purpose. It'd be nice to be able to use page->private for this, but > that's not always available. I know some filesystems already ascribe > a meaning to PG_owner_priv_1, but those can be distinguished by whether > the uptodate flag is set. I suspect what to do will be very ad-hoc depending on the actual user. The first priority is probably to isolate PageError so that it is only used for that kind of communication insid a single file system so that all the generic uses can go away.
The approach I'm taking is to remove the checks for PageError (and folio_test_error). Once nobody is checking the error flag, we can safely remove all places that set/clear it. After this series, there are some places which still use it. Lots of the converted places are trivial -- they should have been checking the uptodate flag. The trickiest are the ones which have multiple steps and signal "hey, something went wrong with one step in the read of this page/folio" by setting the error flag. I'm thinking about (ab)using PageChecked == PG_owner_priv_1 for this purpose. It'd be nice to be able to use page->private for this, but that's not always available. I know some filesystems already ascribe a meaning to PG_owner_priv_1, but those can be distinguished by whether the uptodate flag is set. Thoughts? Most of these look like good cleanups that we want to have anyway. Matthew Wilcox (Oracle) (24): block: Remove check of PageError afs: Remove check of PageError freevxfs: Remove check of PageError gfs: Check PageUptodate instead of PageError hfs: Remove check for PageError hfsplus: Remove check for PageError ntfs: Remove check for PageError ext2: Remove check for PageError nilfs2: Remove check for PageError ntfs: Remove check for PageError ntfs3: Remove check for PageError reiserfs: Remove check for PageError ufs: Remove checks for PageError remap_range: Remove check of uptodate flag jfs: Remove check for PageUptodate iomap: Remove test for folio error orangefs: Remove test for folio error buffer: Remove check for PageError nfs: Leave pages in the pagecache if readpage failed btrfs: Use a folio in wait_dev_supers() buffer: Don't test folio error in block_read_full_folio() squashfs: Return the actual error from squashfs_read_folio() hostfs: Handle page write errors correctly ocfs2: Use filemap_write_and_wait_range() in ocfs2_cow_sync_writeback() block/partitions/core.c | 4 ---- fs/afs/mntpt.c | 6 ------ fs/btrfs/disk-io.c | 19 ++++++++----------- fs/buffer.c | 13 ++++++++----- fs/ext2/dir.c | 3 +-- fs/freevxfs/vxfs_subr.c | 6 ------ fs/gfs2/lops.c | 2 +- fs/hfs/bnode.c | 4 ---- fs/hfsplus/bnode.c | 4 ---- fs/hostfs/hostfs_kern.c | 6 +++--- fs/iomap/buffered-io.c | 3 --- fs/jfs/jfs_metapage.c | 2 +- fs/nfs/read.c | 4 ---- fs/nilfs2/dir.c | 2 +- fs/ntfs/aops.h | 7 +------ fs/ntfs/file.c | 5 ----- fs/ntfs3/ntfs_fs.h | 7 +------ fs/ocfs2/refcounttree.c | 42 ++++++----------------------------------- fs/orangefs/inode.c | 4 +--- fs/reiserfs/xattr.c | 9 +-------- fs/remap_range.c | 11 +---------- fs/squashfs/file.c | 15 ++++++++------- fs/ufs/dir.c | 2 +- fs/ufs/util.c | 11 ----------- 24 files changed, 43 insertions(+), 148 deletions(-)