kernfs, cgroup: reimplement "cgroup.procs" reading for v2

On cgroup v1, the pid listings in "cgroup.procs" and "tasks" are sorted which adds a lot of complications and overhead. v2 doesn't have such requirement and has been intentionally using a modified sorting order so that the output doesn't look sorted to users.

This patchset re-implements "cgroup.procs" reading for v2 which simply keeps a css_task_iter open while the file is being read. Keeping the iterator open makes it unnecessary to skip to the right position on each read segment and associated errors - e.g. incorrectly skipping over pids because earlier pids disappeared between the reads.

Using persistent iterator across multiple read calls requires ->release() callback to clean it up. kernfs operations ->open/release() are added and piped through cftype.

This patchset contains the following five patches.

0001-kernfs-make-kernfs_open_file-mmapped-a-bitfield.patch 0002-kernfs-add-kernfs_ops-open-release-callbacks.patch 0003-cgroup-add-cftype-open-release-callbacks.patch 0004-cgroup-reimplement-reading-cgroup.procs-on-cgroup-v2.patch 0005-cgroup-remove-cgroup_pid_fry-and-friends.patch

0001 is a misc kernfs patch and 0002 adds ->open/release() to kernfs. 0003 pipes ->open/release() through cftype. 0004 implements the new cgroup.procs for v2 and 0005 removes the now unused sort order frying logic.

Greg, would it be okay to route the kernfs patches through cgroup/for-4.11?

The patches are also available in the following git branch.

git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git review-cgroup2-procs

diffstat follows. Thanks.

 fs/kernfs/dir.c             |    2 
 fs/kernfs/file.c            |   53 +++++++++++++++--
 fs/kernfs/kernfs-internal.h |    2 
 include/linux/cgroup-defs.h |    3 +
 include/linux/kernfs.h      |   12 +++-
 kernel/cgroup.c             |  130 +++++++++++++++++++++++++++++---------------
 6 files changed, 148 insertions(+), 54 deletions(-)

--
tejun

More kernfs_open_file->mutex synchronized flags are planned to be added. Convert ->mmapped to a bitfield in preparation.

While at it, make kernfs_fop_mmap() use "true" instead of "1" on ->mmapped.

---
 fs/kernfs/file.c       | 2 +-
 include/linux/kernfs.h | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

index 78219d5..d0b6fb1 100644
@@ -516,7 +516,7 @@ static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
 		goto out_put;
 
 	rc = 0;
-	of->mmapped = 1;
+	of->mmapped = true;
 	of->vm_ops = vma->vm_ops;
 	vma->vm_ops = &kernfs_vm_ops;
 out_put:

index 7056238..afd4e5a 100644
@@ -185,7 +185,7 @@ struct kernfs_open_file {
 	char			*prealloc_buf;
 
 	size_t			atomic_write_len;
-	bool			mmapped;
+	bool			mmapped:1;
 	const struct vm_operations_struct *vm_ops;
 };
 

2.9.3

Please ignore this thread. git-send-email failed in the middle and I resent the series.

Thanks.

tejun