* [GIT PULL] Floppy fixes for 7.4
@ 2026-10-05 15:21 Denis Efremov (Oracle)
2026-10-05 16:00 ` Jens Axboe
0 siblings, 1 reply; 2+ messages in thread
From: Denis Efremov (Oracle) @ 2026-10-05 15:21 UTC (permalink / raw)
To: Jens Axboe; +Cc: Karl Mehltretter, linux-block, linux-kernel
Hi Jens,
Floppy fixes for the 7.4 merge window, based on your for-7.4/block
branch.
Please pull
The following changes since commit 9690943353cec1690a715e000ba6a74e5139f530:
block: amiflop: synchronize flush timer before track flush (2026-10-01 08:37:58 -0600)
are available in the Git repository at:
https://github.com/evdenis/linux-floppy.git tags/floppy-for-7.4
for you to fetch changes up to bb38473f1bcb287567f2fa7980e483a778092504:
floppy: return the drive state flags from the 32-bit FDGETDRVSTAT (2026-10-05 18:59:41 +0400)
----------------------------------------------------------------
Floppy fixes for 7.4
Six fixes for long-standing floppy bugs, most of them found by syzbot
or by race detectors:
- Flush pending work before releasing the IRQ and DMA when no FDC is
found, from Karl Mehltretter. The work queued by the probe could run
after the IRQ and DMA were already freed ("work still pending").
- Unregister the platform device when add_disk() fails, from Guangshuo
Li. The failing drive's device stayed registered after the module
was unloaded, and reading its sysfs attributes oopsed on freed module
memory.
- Return early from reset_interrupt() when cont is already NULL, from
Yang Xiuwei. This fixes the syzbot GPF in reset_interrupt(). I decided
to take v1 as a partial solution; the full fix will require
additional rework.
- Replace volatile on command_status with READ_ONCE()/WRITE_ONCE() and
annotate floppy_work_fn and current_type[], from Cen Zhang. No
functional change.
- Don't keep a pointer to a stack buffer in timeout_message.
show_floppy() printed it from the interrupt handler after
request_done() had returned (syzbot KASAN report).
- Copy the drive state flags in the 32-bit FDGETDRVSTAT and
FDPOLLDRVSTAT. Since 4.13 the compat path returned flags == 0, so
32-bit fdutils saw every disk as write-protected and never saw a
disk change.
Tested in QEMU with a floppy test suite (including the compat ioctls
from 32-bit userspace, and fdutils/mtools/dosfstools workflows) on
KASAN and KCSAN kernels, with no new warnings.
Signed-off-by: Denis Efremov (Oracle) <efremov@linux.com>
----------------------------------------------------------------
Cen Zhang (1):
floppy: annotate data-races around command_status and floppy_work_fn
Denis Efremov (Oracle) (2):
floppy: don't store a stack buffer in timeout_message
floppy: return the drive state flags from the 32-bit FDGETDRVSTAT
Guangshuo Li (1):
floppy: unregister platform device on add_disk failure
Karl Mehltretter (1):
floppy: flush pending work before releasing resources on init failure
Yang Xiuwei (1):
floppy: avoid NULL deref in reset_interrupt when cont is cleared
drivers/block/floppy.c | 56 ++++++++++++++++++++++++++++++--------------------
1 file changed, 34 insertions(+), 22 deletions(-)
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-10-05 16:00 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-05 15:21 [GIT PULL] Floppy fixes for 7.4 Denis Efremov (Oracle)
2026-10-05 16:00 ` Jens Axboe
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®