diff options
| author | Andrea Parri <parri.andrea@gmail.com> | 2026-10-26 19:06:18 +0100 |
|---|---|---|
| committer | Christian Brauner <brauner@kernel.org> | 2026-11-05 18:21:57 +0100 |
| commit | 965e5e451146a19f69523fbcbc73d07e401a3775 (patch) | |
| tree | d6ce9f2fce1045b05bbe44a8bd8f668a64bc57bf /include/linux/timerqueue.h | |
| parent | 1d91141f6722822c8ba6a1c7dda791b006eb129d (diff) | |
iov_iter: take an unsigned long segment count in import_iovec()
readv(), writev() and their positional variants, process_vm_readv(),
process_vm_writev(), vmsplice(), process_madvise(), vectored aio and
the btrfs encoded I/O ioctls all take a full-width segment count, and
all of them hand it to import_iovec() or __import_iovec(). Those take
an unsigned int count, so on 64-bit kernels the upper 32 bits are
dropped before iovec_from_user() checks the count against UIO_MAXIOV.
A count of 0x100000001 is therefore imported as one segment: writev()
on a pipe writes a byte and readv() reads it back, instead of both
failing with EINVAL. A count of 0x100000000 returns 0 as an empty
vector.
import_iovec() was added with an unsigned int count, but the helper it
wrapped, rw_copy_check_uvector(), took an unsigned long, as does
iovec_from_user() today. Until v4.1, readv(), writev(), preadv(),
pwritev(), process_vm_readv(), process_vm_writev() and vectored aio
therefore failed these counts with EINVAL, as did vmsplice() until
v4.18.
Widen nr_segs in both import helpers so the UIO_MAXIOV check sees the
count the caller passed. Callers with 32-bit counts are unaffected.
AI assistance was used to find the bug, develop the fix, and run the
reproducers.
Fixes: bc917be81059 ("saner iov_iter initialization primitives")
Signed-off-by: Andrea Parri <parri.andrea@gmail.com>
Link: https://patch.msgid.link/20261026180619.4147910-1-parri.andrea@gmail.com
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
Diffstat (limited to 'include/linux/timerqueue.h')
0 files changed, 0 insertions, 0 deletions
