2020-03-06 15:21:58

by Corey Minyard

Subject: [PATCH] pid: Fix error return value in some cases

From: Corey Minyard <[email protected]>

Recent changes to alloc_pid() allow the pid number to be specified on
the command line. If set_tid_size is set, then the code scanning the
levels will hard-set retval to -EPERM, overriding it's previous -ENOMEM
value.

After the code scanning the levels, there are error returns that do not
set retval, assuming it is still set to -ENOMEM.

In the first place, pid_ns_prepare_proc() returns its own error, just
use that.

In the second place:

if (!(ns->pid_allocated & PIDNS_ADDING))
goto out_unlock;

a return value of -ENOMEM is probably wrong, since that means that the
namespace is in deletion while this happened. -EINVAL is probably a
better choice.

Fixes: 49cb2fc42ce4 "fork: extend clone3() to support setting a PID"
Signed-off-by: Corey Minyard <[email protected]>
Cc: <[email protected]> # 5.5
Cc: Adrian Reber <[email protected]>
Cc: Christian Brauner <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Dmitry Safonov <[email protected]>
Cc: Andrei Vagin <[email protected]>
Cc: Christian Brauner <[email protected]>
---
kernel/pid.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/kernel/pid.c b/kernel/pid.c
index 0f4ecb57214c..1921f7f4b236 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c
@@ -248,7 +248,8 @@ struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
}

if (unlikely(is_child_reaper(pid))) {
- if (pid_ns_prepare_proc(ns))
+ retval = pid_ns_prepare_proc(ns);
+ if (retval)
goto out_free;
}

@@ -261,8 +262,10 @@ struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,

upid = pid->numbers + ns->level;
spin_lock_irq(&pidmap_lock);
- if (!(ns->pid_allocated & PIDNS_ADDING))
+ if (!(ns->pid_allocated & PIDNS_ADDING)) {
+ retval = -EINVAL;
goto out_unlock;
+ }
for ( ; upid >= pid->numbers; --upid) {
/* Make the PID visible to find_pid_ns. */
idr_replace(&upid->ns->idr, pid, upid->nr);
--
2.17.1

2020-03-06 16:01:51

by Christian Brauner

Subject: Re: [PATCH] pid: Fix error return value in some cases

On Fri, Mar 06, 2020 at 09:20:01AM -0600, [email protected] wrote:
> From: Corey Minyard <[email protected]>
>
> Recent changes to alloc_pid() allow the pid number to be specified on
> the command line. If set_tid_size is set, then the code scanning the
> levels will hard-set retval to -EPERM, overriding it's previous -ENOMEM
> value.
>
> After the code scanning the levels, there are error returns that do not
> set retval, assuming it is still set to -ENOMEM.
>
> In the first place, pid_ns_prepare_proc() returns its own error, just
> use that.
>
> In the second place:
>
> if (!(ns->pid_allocated & PIDNS_ADDING))
> goto out_unlock;
>
> a return value of -ENOMEM is probably wrong, since that means that the
> namespace is in deletion while this happened. -EINVAL is probably a
> better choice.
>
> Fixes: 49cb2fc42ce4 "fork: extend clone3() to support setting a PID"
> Signed-off-by: Corey Minyard <[email protected]>
> Cc: <[email protected]> # 5.5
> Cc: Adrian Reber <[email protected]>
> Cc: Christian Brauner <[email protected]>
> Cc: Oleg Nesterov <[email protected]>
> Cc: Dmitry Safonov <[email protected]>
> Cc: Andrei Vagin <[email protected]>
> Cc: Christian Brauner <[email protected]>
> ---
> kernel/pid.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/pid.c b/kernel/pid.c
> index 0f4ecb57214c..1921f7f4b236 100644
> --- a/kernel/pid.c
> +++ b/kernel/pid.c
> @@ -248,7 +248,8 @@ struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
> }
>
> if (unlikely(is_child_reaper(pid))) {
> - if (pid_ns_prepare_proc(ns))
> + retval = pid_ns_prepare_proc(ns);
> + if (retval)
> goto out_free;
> }
>
> @@ -261,8 +262,10 @@ struct pid *alloc_pid(struct pid_namespace *ns, pid_t *set_tid,
>
> upid = pid->numbers + ns->level;
> spin_lock_irq(&pidmap_lock);
> - if (!(ns->pid_allocated & PIDNS_ADDING))
> + if (!(ns->pid_allocated & PIDNS_ADDING)) {
> + retval = -EINVAL;
> goto out_unlock;
> + }
> for ( ; upid >= pid->numbers; --upid) {
> /* Make the PID visible to find_pid_ns. */
> idr_replace(&upid->ns->idr, pid, upid->nr);

Thanks for the patch.

We definitely regressed the ENOMEM return value case. But both of the
changes here are userspace visible and break assumptions. So I'd rather
just ensure that ENOMEM is returned in both cases like it was before.
In fact, ENOMEM is documented on the pid_namespace man page so we can't
really change it.
Btw, this is the same problem as we had with:

commit 35f71bc0a09a45924bed268d8ccd0d3407bc476f
Author: Michal Hocko <[email protected]>
Date: Thu Apr 16 12:47:38 2015 -0700

fork: report pid reservation failure properly

So please restore just restore the old behavior. Can you add --base to
your next commit, please? Makes it easier for me to pick up.

Thanks!
Christian