| Message ID | 20190610210924.9514-1-yamada.masahiro@socionext.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] kobject: return -ENOSPC when add_uevent_var() fails | expand |
On Tue, Jun 11, 2019 at 06:09:24AM +0900, Masahiro Yamada wrote: > This function never attempts to allocate memory, so returning -ENOMEM > looks weird to me. The reason of the failure is there is no more space > in the given kobj_uevent_env structure. > > Let's change the error code to -ENOSPC. > > This patch is safe since this function had never failed in reality. > > The callers of this function put a fixed number of small strings into > the buffer. > > The buffer is defined to be large enough: > > #define UEVENT_NUM_ENVP 32 /* number of env pointers */ > #define UEVENT_BUFFER_SIZE 2048 /* buffer for the variables */ > > As you see WARN() in the error paths, any failure of this function is > a software bug. > > If such a case had ever happened before, you would have already seen > a noisy back-trace, then you would have increased UEVENT_NUM_ENVP or > UEVENT_BUFFER_SIZE. > > Nobody has ever increased UEVENT_NUM_ENVP or UEVENT_BUFFER_SIZE since > their addition, that is, this structure is always large enough. That implies that we should just drop the WARN() entirely. Especially given that syzbot runs panic-on-warn, right? How about doing both things at the same time? thanks, greg k-h
diff --git a/lib/kobject_uevent.c b/lib/kobject_uevent.c index 7998affa45d4..5ffd44bf4aad 100644 --- a/lib/kobject_uevent.c +++ b/lib/kobject_uevent.c @@ -647,7 +647,7 @@ EXPORT_SYMBOL_GPL(kobject_uevent); * @env: environment buffer structure * @format: printf format for the key=value pair * - * Returns 0 if environment variable was added successfully or -ENOMEM + * Returns 0 if environment variable was added successfully or -ENOSPC * if no space was available. */ int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...) @@ -657,7 +657,7 @@ int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...) if (env->envp_idx >= ARRAY_SIZE(env->envp)) { WARN(1, KERN_ERR "add_uevent_var: too many keys\n"); - return -ENOMEM; + return -ENOSPC; } va_start(args, format); @@ -668,7 +668,7 @@ int add_uevent_var(struct kobj_uevent_env *env, const char *format, ...) if (len >= (sizeof(env->buf) - env->buflen)) { WARN(1, KERN_ERR "add_uevent_var: buffer size too small\n"); - return -ENOMEM; + return -ENOSPC; } env->envp[env->envp_idx++] = &env->buf[env->buflen];
This function never attempts to allocate memory, so returning -ENOMEM looks weird to me. The reason of the failure is there is no more space in the given kobj_uevent_env structure. Let's change the error code to -ENOSPC. This patch is safe since this function had never failed in reality. The callers of this function put a fixed number of small strings into the buffer. The buffer is defined to be large enough: #define UEVENT_NUM_ENVP 32 /* number of env pointers */ #define UEVENT_BUFFER_SIZE 2048 /* buffer for the variables */ As you see WARN() in the error paths, any failure of this function is a software bug. If such a case had ever happened before, you would have already seen a noisy back-trace, then you would have increased UEVENT_NUM_ENVP or UEVENT_BUFFER_SIZE. Nobody has ever increased UEVENT_NUM_ENVP or UEVENT_BUFFER_SIZE since their addition, that is, this structure is always large enough. Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com> --- Changes in v2: - Rephrase the commit log. No code change. lib/kobject_uevent.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) -- 2.17.1