[email protected] wrote:
> From: Serge E. Hallyn <[email protected]>
>
> Signed-off-by: Serge E. Hallyn <[email protected]>
>
> Changelog:
> Nove 4: update restart.c ckpt_err usage
>
> Signed-off-by: Serge E. Hallyn <[email protected]>
> ---
> checkpoint/restart.c | 81
> +++++++++++++++++++++++++++++++++++---------------
> 1 files changed, 57 insertions(+), 24 deletions(-)
Is it your intent to entirely get rid of ckpt_debug() ?
We originally discussed two levels of details: only error status
or a detailed log (and we also thought of a detailed debug, that can
be compiled out to save space). How does that fit with the patch(es) ?
To "define" what's "error status" and what's "log" (and maybe what's
"debug"), I suggest a test like:
1) error status: what conveys the most specific reason of failure,
e.g. "failed to open file to restore fd"; The caller should be able
to assume that the total message(s) length will not exceed a pipe
buffer.
2) log status: that gives status about progress, or what lead to and
what followed an error, e.g. file open failure may have happened
when restoring a file descriptor, or when restore a vma, so a log
like "failed to restore vma" would be helpful.
3) debug status: that we want to be able to compile out without having
to reintroduce it for every bug that it may help us debug.
>
> diff --git a/checkpoint/restart.c b/checkpoint/restart.c
> index 130b4b2..e1bd0ad 100644
> --- a/checkpoint/restart.c
> +++ b/checkpoint/restart.c
> @@ -64,7 +64,7 @@ static int restore_debug_task(struct ckpt_ctx *ctx, int
> flags)
>
> s = kmalloc(sizeof(*s), GFP_KERNEL);
> if (!s) {
> - ckpt_debug("no memory to register ?!\n");
> + ckpt_err(ctx, 0, "no memory to register ?!\n");
> return -ENOMEM;
What is the purpose in passing '0' instead of -ENOMEM to ckpt_err() ?
(a few more instances below).
> }
> s->pid = current->pid;
> @@ -197,13 +197,13 @@ static int _ckpt_read_err(struct ckpt_ctx *ctx, struct
> ckpt_hdr *h)
> len = h->len - sizeof(*h);
> ptr = kzalloc(len + 1, GFP_KERNEL);
> if (!ptr) {
> - ckpt_debug("insufficient memory to report image error\n");
> + ckpt_err(ctx, 0, "insufficient memory to report image error\n");
> return -ENOMEM;
> }
>
> ret = ckpt_kread(ctx, ptr, len);
> if (ret >= 0) {
> - ckpt_debug("%s\n", &ptr[1]);
> + ckpt_err(ctx, 0, "%(S)Error recorded in image\n", &ptr[1]);
> ret = -EIO;
> }
>
> @@ -757,7 +757,8 @@ static int set_task_ctx(struct task_struct *task, struct
> ckpt_ctx *ctx)
> task->checkpoint_ctx = ckpt_ctx_get(ctx);
> ret = 0;
> } else {
> - ckpt_debug("task %d has checkpoint_ctx\n", task_pid_vnr(task));
> + ckpt_err(ctx, 0, "task %d already has checkpoint_ctx\n",
> + task_pid_vnr(task));
> ret = 1;
> }
> task_unlock(task);
> @@ -807,7 +808,7 @@ static int restore_activate_next(struct ckpt_ctx *ctx)
> rcu_read_unlock();
>
> if (!task) {
> - ckpt_debug("could not find task %d\n", pid);
> + ckpt_err(ctx, 0, "could not find task %d\n", pid);
> restore_notify_error(ctx, -ESRCH);
> return -ESRCH;
> }
> @@ -962,29 +963,38 @@ static int do_restore_task(void)
> current->flags |= PF_RESTARTING;
>
> ret = wait_sync_threads();
> - if (ret < 0)
> + if (ret < 0) {
> + ckpt_err(ctx, ret, "wait_sync_threads\n");
(#3) - The only error in wait_sync_threads() can be EINTR, so this
is only necessary for our debugging.
> goto out;
> + }
>
> /* wait for our turn, do the restore, and tell next task in line */
> ret = wait_task_active(ctx);
> - if (ret < 0)
> + if (ret < 0) {
> + ckpt_err(ctx, ret, "wait_task_active\n");
Ditto (it could also be EBUSY - but there are ckpt_debug() statements
there).
> goto out;
> + }
>
> restore_debug_running(ctx);
>
> ret = pre_restore_task();
> - if (ret < 0)
> + if (ret < 0) {
> + ckpt_err(ctx, ret, "pre_restore_task\n");
> goto out;
> + }
>
> zombie = restore_task(ctx);
> if (zombie < 0) {
> ret = zombie;
> + ckpt_err(ctx, ret, "restore_task\n");
I'd assume that restore_task() will already have reported a specific
error that caused the failure. So we can save on code size but not
compiling this in when non-debug.
> goto out;
> }
>
> ret = restore_activate_next(ctx);
> - if (ret < 0)
> + if (ret < 0) {
> + ckpt_err(ctx, ret, "restore_activate_next\n");
> goto out;
> + }
>
> /*
> * zombie: we're done here; do_exit() will notice the @ctx on
> @@ -1025,12 +1035,12 @@ static int __prepare_descendants(struct task_struct
> *task, void *data)
> ckpt_debug("consider task %d\n", task_pid_vnr(task));
>
> if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
> - ckpt_debug("stranger task %d\n", task_pid_vnr(task));
> + ckpt_err(ctx, 0, "stranger task %d\n", task_pid_vnr(task));
This is called with tasklist_lock taken...
> return -EPERM;
> }
>
> if (task_ptrace(task) & PT_PTRACED) {
> - ckpt_debug("ptraced task %d\n", task_pid_vnr(task));
> + ckpt_err(ctx, 0, "ptraced task %d\n", task_pid_vnr(task));
And here too.
> return -EBUSY;
> }
>
[...]
There are similar comments also below.
Are you still concerned about the increase in code size with c/r ?
Oren.
_______________________________________________
Containers mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/containers
_______________________________________________
Devel mailing list
[email protected]
https://openvz.org/mailman/listinfo/devel
