On Sat, Dec 28, 2019 at 02:13:07AM +0000, Eric Wheeler wrote: > On Fri, 27 Dec 2019, Eric Wheeler wrote:
> > Just hit the bug again without mq-scsi (scsi_mod.use_blk_mq=n), all other > > times MQ has been turned on. > > > > I'm trying the -ENOSPC hack which will flag the pool as being out of space > > so I can recover more gracefully than a BUG_ON. Here's a first-draft > > patch, maybe the spinlock will even prevent the issue. > > > > Compile tested, I'll try on a real system tomorrow. > > > > Comments welcome: Both sm_ll_find_free_block() and sm_ll_inc() can trigger synchronous IO. So you absolutely cannot use a spin lock. dm_pool_alloc_data_block() holds a big rw semaphore which should prevent anything else trying to allocate at the same time. - Joe -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
