On Thu, 2022-02-17 at 12:45 -0700, Jens Axboe wrote: > On 2/17/22 8:58 AM, Dylan Yudaken wrote: > > Currently submitting multiple read/write for one file with > > IOSQE_IO_LINK > > and offset = -1 will not behave as if calling read(2)/write(2) > > multiple > > times. The offset may be pinned to the same value for each > > submission (for > > example if they are punted to the async worker) and so each > > read/write will > > have the same offset. > > > > This patchset fixes this by grabbing the file position at execution > > time, > > rather than when the job is queued to be run. > > > > A test for this will be submitted to liburing separately. > > > > Worth noting that this does not purposefully change the result of > > submitting multiple read/write without IOSQE_IO_LINK (for example > > as in > > [1]). But then I do not know what the correct approach should be > > when > > submitting multiple r/w without any explicit ordering. > > > > [1]: > > https://lore.kernel.org/io-uring/8a9e55bf-3195-5282-2907-41b2f2b23cc8@xxxxxxxxx/ > > I think this series looks great, clean and to the point. My only real > question is one you reference here already, which is the fpos locking > that we really should get done. Care to respin the referenced patch > on > top of this series? Would hate to make that part harder... > Sure, I will try and figure that out and add it to the series.
