by Matt Helsley

[permalink] [raw]

Subject: Re: [RFC v2][PATCH 8/9] Remove some BUG_ON()s that need some proper error handling instead.

On Wed, 2008-08-20 at 12:26 -0700, Dave Hansen wrote:
>
>
> ---
>
> oren-cr.git-dave/checkpoint/checkpoint.c | 12 ++++++++++--
> oren-cr.git-dave/checkpoint/restart.c | 15 +++++++++++++++
> 2 files changed, 25 insertions(+), 2 deletions(-)
>
> diff -puN checkpoint/checkpoint.c~0002-Remove-some-BUG_ON-s-that-need-some-proper-error-ha checkpoint/checkpoint.c
> --- oren-cr.git/checkpoint/checkpoint.c~0002-Remove-some-BUG_ON-s-that-need-some-proper-error-ha 2008-08-20 12:12:51.000000000 -0700
> +++ oren-cr.git-dave/checkpoint/checkpoint.c 2008-08-20 12:12:51.000000000 -0700
> @@ -125,7 +125,8 @@ static int cr_write_tail(struct cr_ctx *
> h.id = 0;
>
> hh->magic = CR_HEADER_MAGIC;
> - hh->cksum[0] = hh->cksum[1] = 1; /* TBD ... */
> + hh->cksum[0] = 1;
> + hh->cksum[1] = 1; /* TBD ... */
>
> ret = cr_write_obj(ctx, &h, hh);
> kfree(hh);
> @@ -183,7 +184,14 @@ static int cr_write_task(struct cr_ctx *
> {
> int ret ;
>
> - BUG_ON(t->state == TASK_DEAD);
> + /*
> + * This was a BUG_ON(), which kinda makes sense if you
> + * are only allowing checkpointing of 'current'. But,
> + * it is still pretty silly in that case. Make it
> + * something a bit more sensible.
> + */
> + if (t->state == TASK_DEAD)
> + return -EAGAIN;

This is OK for a debug patch but I think in the end we'd want to do
something similar early in sys_checkpoint -- before we start writing
anything. We should check to ensure the task is properly frozen there
and return -EBUSY if not frozen. Something like:

/* Only current can self-checkpoint. Everything else must be frozen first. */
for_each_task_being_checkpointed(t) {
if ((t != current) && !is_task_frozen_enough(t))
return -EBUSY;
}

is_task_frozen_enough() current lives in kernel/cgroup_freezer.c. I
think it can easily be pulled out of the cgroup_freezer.c code so that
we don't introduce a CONFIG dependency...

> ret = cr_write_task_struct(ctx, t);
> pr_debug("ret (task_struct) %d\n", ret);
> diff -puN checkpoint/restart.c~0002-Remove-some-BUG_ON-s-that-need-some-proper-error-ha checkpoint/restart.c
> --- oren-cr.git/checkpoint/restart.c~0002-Remove-some-BUG_ON-s-that-need-some-proper-error-ha 2008-08-20 12:12:51.000000000 -0700
> +++ oren-cr.git-dave/checkpoint/restart.c 2008-08-20 12:12:51.000000000 -0700
> @@ -74,6 +74,11 @@ static int cr_read_hdr(struct cr_ctx *ct
> struct cr_hdr_head *hh = kmalloc(sizeof(*hh), GFP_KERNEL);
> int ret;
>
> + if (!hh) {
> + pr_debug("unable to get %d bytes from ctx buf for header\n",
> + sizeof(*hh));
> + return -ENOMEM;
> + }
> ret = cr_read_obj_type(ctx, hh, sizeof(*hh), CR_HDR_HEAD);
> if (ret < 0)
> return ret;
> @@ -99,6 +104,11 @@ static int cr_read_tail(struct cr_ctx *c
> struct cr_hdr_tail *hh = kmalloc(sizeof(*hh), GFP_KERNEL);
> int ret;
>
> + if (!hh) {
> + pr_debug("unable to get %d bytes from ctx buf for tail\n",
> + sizeof(*hh));
> + return -ENOMEM;
> + }
> ret = cr_read_obj_type(ctx, hh, sizeof(*hh), CR_HDR_TAIL);
> if (ret < 0)
> return ret;
> @@ -118,6 +128,11 @@ static int cr_read_task_struct(struct cr
> struct task_struct *t = current;
> int ret;
>
> + if (!hh) {
> + pr_debug("unable to get %d bytes from ctx buf for task\n",
> + sizeof(*hh));
> + return -ENOMEM;
> + }
> ret = cr_read_obj_type(ctx, hh, sizeof(*hh), CR_HDR_TASK);
> if (ret < 0)
> return ret;
> _
> _______________________________________________
> Containers mailing list
> [email protected]
> https://lists.linux-foundation.org/mailman/listinfo/containers