• Ken Chen's avatar
    [PATCH] aio: fix buggy put_ioctx call in aio_complete - v2 · dee11c23
    Ken Chen authored
    An AIO bug was reported that sleeping function is being called in softirq
    context:
    
    BUG: warning at kernel/mutex.c:132/__mutex_lock_common()
    Call Trace:
         [<a000000100577b00>] __mutex_lock_slowpath+0x640/0x6c0
         [<a000000100577ba0>] mutex_lock+0x20/0x40
         [<a0000001000a25b0>] flush_workqueue+0xb0/0x1a0
         [<a00000010018c0c0>] __put_ioctx+0xc0/0x240
         [<a00000010018d470>] aio_complete+0x2f0/0x420
         [<a00000010019cc80>] finished_one_bio+0x200/0x2a0
         [<a00000010019d1c0>] dio_bio_complete+0x1c0/0x200
         [<a00000010019d260>] dio_bio_end_aio+0x60/0x80
         [<a00000010014acd0>] bio_endio+0x110/0x1c0
         [<a0000001002770e0>] __end_that_request_first+0x180/0xba0
         [<a000000100277b90>] end_that_request_chunk+0x30/0x60
         [<a0000002073c0c70>] scsi_end_request+0x50/0x300 [scsi_mod]
         [<a0000002073c1240>] scsi_io_completion+0x200/0x8a0 [scsi_mod]
         [<a0000002074729b0>] sd_rw_intr+0x330/0x860 [sd_mod]
         [<a0000002073b3ac0>] scsi_finish_command+0x100/0x1c0 [scsi_mod]
         [<a0000002073c2910>] scsi_softirq_done+0x230/0x300 [scsi_mod]
         [<a000000100277d20>] blk_done_softirq+0x160/0x1c0
         [<a000000100083e00>] __do_softirq+0x200/0x240
         [<a000000100083eb0>] do_softirq+0x70/0xc0
    
    See report: http://marc.theaimsgroup.com/?l=linux-kernel&m=116599593200888&w=2
    
    flush_workqueue() is not allowed to be called in the softirq context.
    However, aio_complete() called from I/O interrupt can potentially call
    put_ioctx with last ref count on ioctx and triggers bug.  It is simply
    incorrect to perform ioctx freeing from aio_complete.
    
    The bug is trigger-able from a race between io_destroy() and aio_complete().
    A possible scenario:
    
    cpu0                               cpu1
    io_destroy                         aio_complete
      wait_for_all_aios {                __aio_put_req
         ...                                 ctx->reqs_active--;
         if (!ctx->reqs_active)
            return;
      }
      ...
      put_ioctx(ioctx)
    
                                         put_ioctx(ctx);
                                            __put_ioctx
                                              bam! Bug trigger!
    
    The real problem is that the condition check of ctx->reqs_active in
    wait_for_all_aios() is incorrect that access to reqs_active is not
    being properly protected by spin lock.
    
    This patch adds that protective spin lock, and at the same time removes
    all duplicate ref counting for each kiocb as reqs_active is already used
    as a ref count for each active ioctx.  This also ensures that buggy call
    to flush_workqueue() in softirq context is eliminated.
    Signed-off-by: default avatar"Ken Chen" <kenchen@google.com>
    Cc: Zach Brown <zach.brown@oracle.com>
    Cc: Suparna Bhattacharya <suparna@in.ibm.com>
    Cc: Benjamin LaHaise <bcrl@kvack.org>
    Cc: Badari Pulavarty <pbadari@us.ibm.com>
    Cc: <stable@kernel.org>
    Acked-by: default avatarJeff Moyer <jmoyer@redhat.com>
    Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
    dee11c23
aio.c 45 KB