On Thu, May 16, 2024 at 10:12:25AM +0800, YangYang wrote: > On 2024/5/16 0:00, Benjamin Marzinski wrote: > > On Wed, May 15, 2024 at 11:42:04AM -0400, Benjamin Marzinski wrote: > > > When a target calls dm_get_device(), if it adds a new table device to > > > t->devices, then it's the first target in this table to use that device. > > > If flush_pass_around is set for this target, then it also sets > > > sends_pass_around_flush. In __send_empty_flush() if the table has > > > flush_pass_around set, when you iterate through the devices, you only > > > > Err, "When you iterate through the *targets*, you only ..." In this > > method you don't iterate through the list of devices (which is supposed > > to be protected by t->devices_lock). > > I'm not very familiar with this area, I thought that the device list > of an active table cannot be modified, so it doesn't need to be > protected by t->devices_lock. Actually, looking at this some more you're basically correct. The only place where dm can modify the devices list of active table is in multipath_message(), and that's bug. So you should be safe not locking here. I'll post a patch to fix the multipath target. -Ben > > > > > call __send_empty_flush_bios() for the ones with sends_pass_around_flush > > > set. > > > > > > Or am I overlooking something? > > > > > > -Ben > >
