| Message ID | c2ef5a47-85b2-3a98-020f-766f153a65d0@virtuozzo.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] nfs4: handle async processing of F_SETLK with FL_SLEEP | expand |
Patch is wrong, I missed that FL_FLOCK is handled here too, moreover locks_lock_inode_wait() is used in nfs_proc_lock and nfs3_proc_locks, and FL_POSIX request for nfs v2 and v3 can be blocked too. On 27.12.2021 18:51, Vasily Averin wrote: > nfsd and lockd use F_SETLK cmd with the FL_SLEEP flag set to request > asynchronous processing of blocking locks. > > Currently nfs4 use locks_lock_inode_wait() function blocked on such requests. > To handle such requests properly, non-blocking posix_file_lock() > function should be used instead. > > https://bugzilla.kernel.org/show_bug.cgi?id=215383 > Signed-off-by: Vasily Averin <vvs@virtuozzo.com> > --- > VvS: I'm not sure that request->fl_file points to the same state->inode > used in locks_lock_inode_wait(). If it is not, posix_lock_inode() can be > used here, but this function is static currently and should be exported first. > > v2: fixed 'fl_flags && FL_SLEEP' => 'fl_flags & FL_SLEEP' > --- > fs/nfs/nfs4proc.c | 7 +++++-- > 1 file changed, 5 insertions(+), 2 deletions(-) > > diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c > index ee3bc79f6ca3..f899f4bcdae5 100644 > --- a/fs/nfs/nfs4proc.c > +++ b/fs/nfs/nfs4proc.c > @@ -7200,8 +7200,11 @@ static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock > int status; > > request->fl_flags |= FL_ACCESS; > - status = locks_lock_inode_wait(state->inode, request); > - if (status < 0) > + if ((request->fl_flags & FL_SLEEP) && IS_SETLK(cmd)) > + status = posix_lock_file(request->fl_file, request, NULL); > + else > + status = locks_lock_inode_wait(state->inode, request); > + if (status) > goto out; > mutex_lock(&sp->so_delegreturn_mutex); > down_read(&nfsi->rwsem); >
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index ee3bc79f6ca3..f899f4bcdae5 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -7200,8 +7200,11 @@ static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock int status; request->fl_flags |= FL_ACCESS; - status = locks_lock_inode_wait(state->inode, request); - if (status < 0) + if ((request->fl_flags & FL_SLEEP) && IS_SETLK(cmd)) + status = posix_lock_file(request->fl_file, request, NULL); + else + status = locks_lock_inode_wait(state->inode, request); + if (status) goto out; mutex_lock(&sp->so_delegreturn_mutex); down_read(&nfsi->rwsem);
nfsd and lockd use F_SETLK cmd with the FL_SLEEP flag set to request asynchronous processing of blocking locks. Currently nfs4 use locks_lock_inode_wait() function blocked on such requests. To handle such requests properly, non-blocking posix_file_lock() function should be used instead. https://bugzilla.kernel.org/show_bug.cgi?id=215383 Signed-off-by: Vasily Averin <vvs@virtuozzo.com> --- VvS: I'm not sure that request->fl_file points to the same state->inode used in locks_lock_inode_wait(). If it is not, posix_lock_inode() can be used here, but this function is static currently and should be exported first. v2: fixed 'fl_flags && FL_SLEEP' => 'fl_flags & FL_SLEEP' --- fs/nfs/nfs4proc.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)