Messages in this thread | | | Date | Wed, 16 Oct 2013 18:41:05 -0400 | From | Douglas Gilbert <> | Subject | Re: [Bug] 12.864681 BUG: lock held when returning to user space! |
| |
On 13-10-16 09:24 AM, James Bottomley wrote: > On Tue, 2013-10-08 at 09:45 -0400, Douglas Gilbert wrote: >> On 13-10-08 02:44 AM, vaughan wrote: >>> Hi Madper, >>> >>> CC to Douglas to get comments. >>> I use the rw_semaphore o_sem to protect excl open, introduced in commit >>> 15b06f9a02406e5460001db6d5af5c738cd3d4e7 since v3.12-rc1. >>> Is it forbidden to do like that in kernel?... >> >> It appears you can not (allow sg_open() to hold a semaphore >> then return to the user space). So you will need to do some >> rework on that patch or revert it. > > OK, there being no reply on this, I'll do the revert ... that's all four > patches, correct?
That seems to be the case. Vaughan acknowledged the problem and forwarded it to me 8 days ago. Yes, it seems to be a "no-no" to hold a any kernel semaphore when returning to the user space; in this case from sg_open(). I was hoping a revised patch might appear from Vaughan but to date that has not been the case. So with only a few weeks to go before lk 3.12 is released, reverting the whole 4 patches in that series seems to be the safest course.
Also without a new patch from Vaughan in the next few weeks he may also miss the opportunity of getting his improved O_EXCL logic into the lk 3.13 series.
Thinking about how to solve this problem: a field could be added to 'struct sg_device' with one of three states: no_opens, non_excl_opens and excl_open. It could be manipulated by sg_open() and sg_release() like a read-write semaphore. And the faulty 'struct rw_semaphore o_sem' in sg_device could be replaced by a normal semaphore to protect the manipulation of the new three-state field.
Doug Gilbert
| |