Hello Bernard Metzler,
The patch bdf1da5df9da: "RDMA/siw: Fix immediate work request flush
to completion queue" from Nov 7, 2022, leads to the following Smatch
static checker warning:
drivers/infiniband/sw/siw/siw_cq.c:96 siw_reap_cqe()
error: buffer overflow 'map_cqe_status' 10 <= 21
drivers/infiniband/sw/siw/siw_cq.c
48 int siw_reap_cqe(struct siw_cq *cq, struct ib_wc *wc)
49 {
50 struct siw_cqe *cqe;
51 unsigned long flags;
52
53 spin_lock_irqsave(&cq->lock, flags);
54
55 cqe = &cq->queue[cq->cq_get % cq->num_cqe];
56 if (READ_ONCE(cqe->flags) & SIW_WQE_VALID) {
57 memset(wc, 0, sizeof(*wc));
58 wc->wr_id = cqe->id;
59 wc->byte_len = cqe->bytes;
60
61 /*
62 * During CQ flush, also user land CQE's may get
63 * reaped here, which do not hold a QP reference
64 * and do not qualify for memory extension verbs.
65 */
66 if (likely(rdma_is_kernel_res(&cq->base_cq.res))) {
67 if (cqe->flags & SIW_WQE_REM_INVAL) {
68 wc->ex.invalidate_rkey = cqe->inval_stag;
69 wc->wc_flags = IB_WC_WITH_INVALIDATE;
70 }
71 wc->qp = cqe->base_qp;
72 wc->opcode = map_wc_opcode[cqe->opcode];
73 wc->status = map_cqe_status[cqe->status].ib;
74 siw_dbg_cq(cq,
75 "idx %u, type %d, flags %2x, id 0x%pK\n",
76 cq->cq_get % cq->num_cqe, cqe->opcode,
77 cqe->flags, (void *)(uintptr_t)cqe->id);
78 } else {
79 /*
80 * A malicious user may set invalid opcode or
81 * status in the user mmapped CQE array.
82 * Sanity check and correct values in that case
83 * to avoid out-of-bounds access to global arrays
84 * for opcode and status mapping.
85 */
86 u8 opcode = cqe->opcode;
87 u16 status = cqe->status;
88
89 if (opcode >= SIW_NUM_OPCODES) {
90 opcode = 0;
91 status = IB_WC_GENERAL_ERR;
IB_WC_GENERAL_ERR is 21.
92 } else if (status >= SIW_NUM_WC_STATUS) {
93 status = IB_WC_GENERAL_ERR;
Here too.
94 }
95 wc->opcode = map_wc_opcode[opcode];
--> 96 wc->status = map_cqe_status[status].ib;
^^^^^^^^^^^^^^^
Out of bounds.
97
98 }
99 WRITE_ONCE(cqe->flags, 0);
100 cq->cq_get++;
101
102 spin_unlock_irqrestore(&cq->lock, flags);
103
104 return 1;
105 }
106 spin_unlock_irqrestore(&cq->lock, flags);
107
108 return 0;
109 }
regards,
dan carpenter
