2021-12-17 11:24:10

by Vasily Averin

[permalink] [raw]
Subject: [PATCH] ksmbd: force "fail immediately" flag on fs with its own ->lock

Like nfsd and lockd, ksmbd can cause deadlock if the exported
file system does not support asynchronous processing of blocking locks:
if all ksmbd worker threads handles such requests,
they can never finish and the server will not be able to handle
any other incoming requests.

Any filesystem that leaves ->lock NULL will use posix_lock_file(), which
does the right thing. Simplest is just to assume that any filesystem
that defines its own ->lock is not safe to request a blocking lock from.

To work around the problem we need to drop fl->fl_flag FL_SLEEP before
vfs_lock_file() execution, it forces affected functions to avoid blocking.

Signed-off-by: Vasily Averin <[email protected]>
---
fs/ksmbd/smb2pdu.c | 6 ++++++
1 file changed, 6 insertions(+)

diff --git a/fs/ksmbd/smb2pdu.c b/fs/ksmbd/smb2pdu.c
index 121f8e8c70ac..f2225491af02 100644
--- a/fs/ksmbd/smb2pdu.c
+++ b/fs/ksmbd/smb2pdu.c
@@ -6634,6 +6634,12 @@ static int smb2_set_flock_flags(struct file_lock *flock, int flags)
{
int cmd = -EINVAL;

+ if ((flock->fl_file->f_op->lock) &&
+ ((flags == SMB2_LOCKFLAG_SHARED) ||
+ (flags == SMB2_LOCKFLAG_EXCLUSIVE))) {
+ ksmbd_debug(SMB, "force fail immediately request\n");
+ flags |= SMB2_LOCKFLAG_FAIL_IMMEDIATELY;
+ }
/* Checking for wrong flag combination during lock request*/
switch (flags) {
case SMB2_LOCKFLAG_SHARED:
--
2.25.1



2021-12-22 09:01:23

by Namjae Jeon

[permalink] [raw]
Subject: Re: [PATCH] ksmbd: force "fail immediately" flag on fs with its own ->lock

2021-12-17 20:23 GMT+09:00, Vasily Averin <[email protected]>:
> Like nfsd and lockd, ksmbd can cause deadlock if the exported
> file system does not support asynchronous processing of blocking locks:
> if all ksmbd worker threads handles such requests,
> they can never finish and the server will not be able to handle
> any other incoming requests.
>
> Any filesystem that leaves ->lock NULL will use posix_lock_file(), which
> does the right thing. Simplest is just to assume that any filesystem
> that defines its own ->lock is not safe to request a blocking lock from.
>
> To work around the problem we need to drop fl->fl_flag FL_SLEEP before
> vfs_lock_file() execution, it forces affected functions to avoid blocking.
>
> Signed-off-by: Vasily Averin <[email protected]>
Acked-by: Namjae Jeon <[email protected]>

Thanks!
> ---
> fs/ksmbd/smb2pdu.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/fs/ksmbd/smb2pdu.c b/fs/ksmbd/smb2pdu.c
> index 121f8e8c70ac..f2225491af02 100644
> --- a/fs/ksmbd/smb2pdu.c
> +++ b/fs/ksmbd/smb2pdu.c
> @@ -6634,6 +6634,12 @@ static int smb2_set_flock_flags(struct file_lock
> *flock, int flags)
> {
> int cmd = -EINVAL;
>
> + if ((flock->fl_file->f_op->lock) &&
> + ((flags == SMB2_LOCKFLAG_SHARED) ||
> + (flags == SMB2_LOCKFLAG_EXCLUSIVE))) {
> + ksmbd_debug(SMB, "force fail immediately request\n");
> + flags |= SMB2_LOCKFLAG_FAIL_IMMEDIATELY;
> + }
> /* Checking for wrong flag combination during lock request*/
> switch (flags) {
> case SMB2_LOCKFLAG_SHARED:
> --
> 2.25.1
>
>