cgroup: Honor caller's cgroup NS when resolving path
Subsystems (2)
Hello Michal Koutný,
The patch 74e4b956eb1c: "cgroup: Honor caller's cgroup NS when resolving path" from Aug 26, 2022, leads to the following Smatch static checker warning:
./include/linux/kernfs.h:600 kernfs_walk_and_get() warn: sleeping in atomic context
kernel/cgroup/cgroup.c
6639 struct cgroup *cgroup_get_from_path(const char *path) 6640 {
6641 struct kernfs_node *kn;
6642 struct cgroup *cgrp = ERR_PTR(-ENOENT);
6643 struct cgroup *root_cgrp;
6644
6645 spin_lock_irq(&css_set_lock);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 6646 root_cgrp = current_cgns_cgroup_from_root(&cgrp_dfl_root);
6647 kn = kernfs_walk_and_get(root_cgrp->kn, path);
^^^^^^^^^^^^^^^^^^^The kernfs_walk_and_get() function calls down_read() (a semaphore) so it can't be called while holding a spinlock.
6648 spin_unlock_irq(&css_set_lock); 6649 if (!kn) 6650 goto out;
regards,
dan carpenter
Michal Koutný
maintainerHello Dan Carpenter.
Quoted 2 lines · Dan Carpenter
The
kernfs_walk_and_get()function callsdown_read()(a semaphore) so it can't be called while holding a spinlock.
Thanks for the report (I should have looked deeper into kernfs_walk_and_get).
I'd propose a patch like bellow.
Are you OK with adding
Quoted 1 line
Reported-by Dan Carpenter <[email protected]>
as I'll send the proposed patch to public mailing lists.
Thanks,
Michal
----8<----
From e2de5c76969e5ac4c7fe5e00a54229cbe59bd86a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Michal=20Koutn=C3=BD?= <[email protected]>
Date: Wed, 31 Aug 2022 10:36:19 +0200
Subject: [PATCH] cgroup: Reorganize css_set_lock and kernfs path processing
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The commit 74e4b956eb1c incorrectly wrapped kernfs_walk_and_get under css_set_lock. That lock is needed by __cset_cgroup_from_root to ensure stable cset->cgrp_links. The returned cgroup object is pinned by the css_set. Kernfs code that traverses paths to this cgroup does not need css_set_lock (lowest cgroup pins its ancestry and cgroup pins its cgroup->kn).
TODO root(s) of named v1 hierarchies
--- kernel/cgroup/cgroup.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-)
kernel/cgroup/cgroup.c +7 -5
@@ -1391,11 +1391,16 @@ static void cgroup_destroy_root(struct cgroup_root *root)cgroup_free_root (root); } +/* +* Returned cgroup is without refcount but it's valid as long as cset pins it. +*/ static inline struct cgroup *__cset_cgroup_from_root (struct css_set *cset,struct cgroup_root *root) {struct cgroup *res_cgroup =NULL ; +lockdep_assert_held (&css_set_lock); +if (cset == &init_css_set) { res_cgroup = &root->cgrp; }else if (root == &cgrp_dfl_root) {
@@ -1426,8 +1431,6 @@ current_cgns_cgroup_from_root(struct cgroup_root *root)struct cgroup *res =NULL ;struct css_set *cset; -lockdep_assert_held (&css_set_lock); -rcu_read_lock (); cset = current->nsproxy->cgroup_ns->root_cset;
@@ -1446,7 +1449,6 @@ static struct cgroup *cset_cgroup_from_root(struct css_set *cset,struct cgroup *res =NULL ;lockdep_assert_held (&cgroup_mutex); -lockdep_assert_held (&css_set_lock); res =__cset_cgroup_from_root (cset, root);
@@ -1861,8 +1863,8 @@ int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,spin_lock_irq (&css_set_lock); ns_cgroup =current_cgns_cgroup_from_root (kf_cgroot); - len =kernfs_path_from_node (kf_node, ns_cgroup->kn, buf,PATH_MAX );spin_unlock_irq (&css_set_lock); + len =kernfs_path_from_node (kf_node, ns_cgroup->kn, buf,PATH_MAX );if (len >=PATH_MAX ) len = -ERANGE ;
@@ -6649,8 +6651,8 @@ struct cgroup *cgroup_get_from_path(const char *path)spin_lock_irq (&css_set_lock); root_cgrp =current_cgns_cgroup_from_root (&cgrp_dfl_root); - kn =kernfs_walk_and_get (root_cgrp->kn, path);spin_unlock_irq (&css_set_lock); + kn =kernfs_walk_and_get (root_cgrp->kn, path);if (!kn) goto out;
base-commit: c0f2df49cf2471289d5aabf16f50ac26eb268f7d
Signature · 1 line
2.37.0
Quoted 6 lines · Michal Koutný
Hello Dan Carpenter.
Quoted 2 lines · Dan Carpenter
The
kernfs_walk_and_get()function callsdown_read()(a semaphore) so it can't be called while holding a spinlock.
Quoted 9 lines
Thanks for the report (I should have looked deeper into
kernfs_walk_and_get).I'd propose a patch like bellow.
Are you OK with adding
Quoted 1 line
Reported-by Dan Carpenter <[email protected]>as I'll send the proposed patch to public mailing lists.
Yes, please. Thanks!
regards,
dan carpenter
Tejun Heo
maintainerHello,
kernel/cgroup/cgroup.c static void cgroup_destroy_root(struct cgroup_root *root) · line 1398
Quoted 5 lines · Michal Koutný
kernel/cgroup/cgroup.c +3
-0+/* +* Returned cgroup is without refcount but it's valid as long as cset pins it. +*/ static inline struct cgroup *__cset_cgroup_from_root (struct css_set *cset,struct cgroup_root *root)
Yeah but what prevents the task from migrating away unpinning the whole thing? Can you just get the ref on the cgroup before releasing css_set? That's not an expensive operation and a lot more straightforward.
Thanks.
Signature · 1 line
tejun