io_uring/net: keep recvmsg buffer select length within MAX_RW_COUNT
PATCH 1/1 io_uring/net: keep recvmsg buffer select length within MAX_RW_COUNT
io_recvmsg() takes the transfer length from the ring provided buffer when sqe->len is 0, and such a buffer can be up to 4G. min_ret is an int, so 2G or more makes it negative and the (ret < min_ret) block is skipped, and the oversized length reaches the socket layer unchanged: an empty socket then completes with 0 instead of arming poll, and a short read is reported as complete. Clamp the length to MAX_RW_COUNT, as io_recv() does.
--- io_uring/net.c | 1 + 1 file changed, 1 insertion(+)
io_uring/net.c +1 -0
@@ -1068,6 +1068,7 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags) sel =io_buffer_select (req, &len, sr->buf_group, issue_flags);if (!sel.addr)return -ENOBUFS ; + len =min_t (size_t, len,MAX_RW_COUNT );if (req->flags &REQ_F_APOLL_MULTISHOT ) { ret =io_recvmsg_prep_multishot (kmsg, sr, &sel.addr, &len);
Signature · 1 line
2.51.0
Jens Axboe
maintainerQuoted 8 lines · Yichong Chen
io_recvmsg()takes the transfer length from the ring provided buffer whensqe->lenis 0, and such a buffer can be up to 4G.min_retis an int, so 2G or more makes it negative and the (ret <min_ret) block is skipped, and the oversized length reaches the socket layer unchanged: an empty socket then completes with 0 instead of arming poll, and a short read is reported as complete. Clamp the length toMAX_RW_COUNT, asio_recv()does.Fixes 78d0d2063bab ("io_uring/net: disable partial retries for recvmsg with cmsg")
Don't think that's correct, it just moved the check. Should be:
c7fb19428d67 ("io_uring: add support for ring mapped supplied buffers")
instead. I will fix that up. Waiting on Linus pulling the previous bits, then I will queue this up for 7.4.
Signature · 1 line
Jens Axboe
Quoted 8 lines · Yichong Chen
io_recvmsg()takes the transfer length from the ring provided buffer whensqe->lenis 0, and such a buffer can be up to 4G.min_retis an int, so 2G or more makes it negative and the (ret <min_ret) block is skipped, and the oversized length reaches the socket layer unchanged: an empty socket then completes with 0 instead of arming poll, and a short read is reported as complete. Clamp the length toMAX_RW_COUNT, asio_recv()does.[...]
Applied, thanks!
[1/1] io_uring/net: keep recvmsg buffer select length within MAX_RW_COUNT
commit: 18d3f56eb6e6500a65d81cf265f0fe74fd9ecab6
Best regards,
Signature · 1 line
Jens Axboe