Hi all, Al pointed out multiple times that seq_get_buf is highly dangerous as it opens up the tight seq_file abstractions to buffer overflows. The last such caller now is sysfs. This series allows attributes to implement a seq_show method and switch the block and XFS code as users that I'm most familiar with to use seq_files directly after a few preparatory cleanups. With this series "leaf" users of sysfs_ops can be converted one at at a time, after that we can move the seq_get_buf into the multiplexers (e.g. kobj, device, class attributes) and remove the show method in sysfs_ops and repeat the process until all attributes are converted. This will probably take a fair amount of time. Diffstat: block/bfq-iosched.c | 12 +- block/blk-integrity.c | 44 +++++---- block/blk-mq-sysfs.c | 64 ++++++-------- block/blk-sysfs.c | 209 ++++++++++++++++++++++++++--------------------- block/blk-throttle.c | 5 - block/blk.h | 2 block/elevator.c | 42 +++++---- block/kyber-iosched.c | 7 - block/mq-deadline.c | 5 - fs/sysfs/file.c | 135 +++++++++++++++++------------- fs/sysfs/group.c | 15 +-- fs/sysfs/sysfs.h | 8 + fs/xfs/xfs_error.c | 14 +-- fs/xfs/xfs_stats.c | 24 ++--- fs/xfs/xfs_stats.h | 2 fs/xfs/xfs_sysfs.c | 96 ++++++++++----------- include/linux/elevator.h | 4 include/linux/kernfs.h | 28 ------ include/linux/seq_file.h | 4 include/linux/sysfs.h | 9 +- 20 files changed, 376 insertions(+), 353 deletions(-)
