Do not report failure on zero sized writes, and handle them as no-op. There's issues for example in case of writev() when there's iovec containing zero buffer as a first one. It's expected writev() on below example to successfully perform the write to specified writable cgroup file expecting integer value, and to return 2. For now it's returning value -1, and skipping the write: int writetest(int fd) { const char *buf1 = ""; const char *buf2 = "1\n"; struct iovec iov[2] = { { .iov_base = (void*)buf1, .iov_len = 0 }, { .iov_base = (void*)buf2, .iov_len = 2 } }; return writev(fd, iov, 2); } This patch fixes the issue by checking if there's nothing to write, and handling the write as no-op by just returning 0.
Showing the changes to kernel (1 of 1 file). Show all
Unified · Split
| @@ -3681,6 +3681,9 @@ static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, · 3680 unmodified lines | |||
| 3681 | struct cgroup_subsys_state *css; | 3681 | struct cgroup_subsys_state *css; |
| 3682 | int ret; | 3682 | int ret; |
| 3683 | 3683 | ||
| 3684 | if (!nbytes) | ||
| 3685 | return 0; | ||
| 3686 | |||
| 3684 | /* | 3687 | /* |
| 3685 | * If namespaces are delegation boundaries, disallow writes to | 3688 | * If namespaces are delegation boundaries, disallow writes to |
| 3686 | * files in an non-init namespace root from inside the namespace | 3689 | * files in an non-init namespace root from inside the namespace |