RFCv2: fix fatal signal handling in __blkdev_issue_discard

[Christoph Hellwig]

[sorry for the resend, the hotel wifi just broke down]

Hi all,

this tries to address the block for-next oops Chandan reported on XFS.
I can't actually reproduce it unfortunately, but this series should
sort it out by movign the fatal_signal_pending check out of all but
the ioctl path.  The write_zeroes and secure_erase path will need
similar treatment eventually.

Tested with blktests and the xfstests discard group for xfs only.

Changes since v1:
 - open code the fatal signal logic in the ioctl handler
 - better bio-level helpers
 - drop the file system cleanups for now

Diffstat:
 block/bio.c         |   48 +++++++++++++++++++++++----
 block/blk-lib.c     |   90 +++++++++++++++-------------------------------------
 block/blk.h         |    1 
 block/ioctl.c       |   35 +++++++++++++++++---
 include/linux/bio.h |    4 ++
 5 files changed, 102 insertions(+), 76 deletions(-)