Re: [PATCH v6 28/56] scsi: qla2xxx: Enhance purex_entry handling for 29xx series

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 7 potential issue(s) to consider:

New issues:
- [High] The 9-bit virtual port index (`vp_idx`) from 29xx hardware is silently truncated to 8 bits.
- [High] Incorrect response IOCB bitfield corruption for 29xx hardware due to writing via a legacy 24xx struct view.
- [High] Buffer over-read in continuation IOCB parsing on 29xx hardware triggers a FORTIFY_SOURCE panic.

Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but a crafted ELS payload containing `0xDEADDEAD` at offset 16 will permanently hang the response queue.
- [High] This problem wasn't introduced by this patch, but `__qla_consume_iocb` consumes an extra, unrelated IOCB and fails to clear the original packet's signature.
- [High] This problem wasn't introduced by this patch, but there is a data race between `atomic_inc_return` and non-atomic `memset` on the default PUREX item.
- [High] This problem wasn't introduced by this patch, but there is an infinite polling loop inside interrupt context causing potential hard lockup.
--

--- Patch [28]: [PATCH v6 28/56] scsi: qla2xxx: Enhance purex_entry handling for 29xx series ---
commit 44ae6bdcc8dd39ac0f056b9473fb3475613ffb9b
Author: Nilesh Javali <njavali@marvell.com>
Subject: scsi: qla2xxx: Enhance purex_entry handling for 29xx series

This patch updates purex_entry handling to support the 29xx series by
differentiating between standard 64-byte and 128-byte extended structures.
It reuses overlapping byte views to avoid duplicating read paths.

> diff --git a/drivers/scsi/qla2xxx/qla_edif.c b/drivers/scsi/qla2xxx/qla_edif.c
> --- a/drivers/scsi/qla2xxx/qla_edif.c
> +++ b/drivers/scsi/qla2xxx/qla_edif.c
> @@ -2544,26 +2544,55 @@ void qla24xx_auth_els(scsi_qla_host_t *vha, void **pkt, struct rsp_que **rsp)
[ ... ]
> +	uint8_t vp_idx;
> +	struct purex_entry_24xx *p = *pkt;
[ ... ]
> +	if (IS_QLA29XX(ha))
> +		vp_idx = le16_to_cpu(((struct purex_entry_24xx_ext *)
> +				      *pkt)->vp_idx);

[Severity: High]
Does assigning the 16-bit packed vp_idx value to a uint8_t local variable
truncate the 9-bit virtual port index for 29xx hardware? This looks like it
might cause responses for NPIV virtual ports with IDs 256 and above to be
misrouted or dropped.

[ ... ]
> diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
> --- a/drivers/scsi/qla2xxx/qla_isr.c
> +++ b/drivers/scsi/qla2xxx/qla_isr.c
> @@ -202,6 +202,11 @@ void __qla_consume_iocb(struct scsi_qla_host *vha,
[ ... ]
>  	entry_count_remaining = purex->entry_count;
>  	while (entry_count_remaining > 0) {
>  		new_pkt = rsp_q->ring_ptr;
>  		*pkt = new_pkt;

[Severity: High]
This is a pre-existing issue, but does __qla_consume_iocb skip the first
packet and over-consume the ring buffer?

The loop executes entry_count times starting from the next packet:

    while (entry_count_remaining > 0) {
        new_pkt = rsp_q->ring_ptr;
        *pkt = new_pkt;

Could this consume and silently drop the next valid IOCB while leaving the
original IOCB signature uncleared?

[ ... ]
> @@ -262,14 +275,24 @@ int __qla_copy_purex_to_buffer(struct scsi_qla_host *vha,
[ ... ]
> +	if (IS_QLA29XX(ha)) {
> +		struct purex_entry_24xx_ext *purex_ext = *pkt;
>  
> -	memcpy(buf, &purex->els_frame_payload[0], no_bytes);
> +		memcpy(buf, &purex_ext->els_frame_payload[0], no_bytes);
> +	} else {
> +		memcpy(buf, &purex->els_frame_payload[0], no_bytes);
> +	}

[Severity: High]
Is there a risk of a FORTIFY_SOURCE buffer over-read panic during continuation
IOCB parsing on 29xx hardware?

In __qla_copy_purex_to_buffer and similar functions, the data pointer is
extracted using the sts_cont_entry_t structure:

    data = ((sts_cont_entry_t *)*pkt)->data;
    data_sz = qla_sts_cont_data_size(ha);

Since sts_cont_entry_t->data is statically defined as a 60-byte array,
FORTIFY_SOURCE sets __builtin_object_size to 60. On 29xx hardware, data_sz
returns 124 bytes. Could the subsequent memcpy trigger a panic by attempting
to copy up to 124 bytes from a 60-byte object?

[ ... ]
> @@ -867,6 +891,13 @@ qla27xx_copy_multiple_pkt(struct scsi_qla_host *vha, void **pkt,
[ ... ]
> +	} else if (IS_QLA29XX(ha)) {
> +		purex_ext = *pkt;
> +		total_bytes = (le16_to_cpu(purex_ext->frame_size) & 0x0FFF) -
> +			      PURX_ELS_HEADER_SIZE;

[Severity: High]
This is a pre-existing issue, but can the spin loop in
qla27xx_copy_multiple_pkt cause a hard lockup?

If the hardware fails to provide the next packet, the code executes an
unbounded spin loop in interrupt context:

    if (rsp_q->ring_ptr->signature == RESPONSE_PROCESSED) {
        ql_dbg(...);
        cpu_relax();
        continue;
    }

Could a hardware fault or maliciously crafted entry_count trap the CPU
forever?

[ ... ]
> @@ -1075,9 +1110,9 @@ qla24xx_alloc_purex_item(scsi_qla_host_t *vha, uint16_t size)
[ ... ]
>  	} else {
>  		if (atomic_inc_return(&vha->default_item.in_use) == 1) {
>  			item = &vha->default_item;

[Severity: High]
This is a pre-existing issue, but is there a data race when acquiring the
default PUREX item?

In qla24xx_alloc_purex_item, the code uses atomic_inc_return to acquire the
item:

    if (atomic_inc_return(&vha->default_item.in_use) == 1) {

However, in qla24xx_free_purex_item, the memory is cleared with a non-atomic
memset:

    memset(&item->vha->default_item, 0, sizeof(struct purex_item));

Could two threads simultaneously acquire the default_item if the memset
clears the in_use field while another thread is checking it out?

[ ... ]
> @@ -4151,6 +4201,8 @@ void qla24xx_process_response_queue(struct scsi_qla_host *vha,
[ ... ]
>  		case PUREX_IOCB_TYPE:
> -			purex_entry = (void *)pkt;
> -			switch (purex_entry->els_frame_payload[3]) {
> +			if (IS_QLA29XX(ha)) {
> +				purex_entry_ext = (void *)pkt;

[Severity: High]
This is a pre-existing issue, but can the response queue permanently hang if
a crafted ELS payload contains 0xDEADDEAD at offset 16?

In qla24xx_process_response_queue, the signature field is evaluated to
determine if the hardware has written a new IOCB:

    while (rsp->ring_index != rsp_in &&
           rsp->ring_ptr->signature != RESPONSE_PROCESSED) {

Because the signature field overlaps with bytes 16-19 of the
els_frame_payload, could a malicious payload prematurely terminate the polling
loop and prevent further packets from being processed?

[ ... ]
> diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
> --- a/drivers/scsi/qla2xxx/qla_os.c
> +++ b/drivers/scsi/qla2xxx/qla_os.c
> @@ -6209,15 +6211,26 @@ void qla24xx_process_purex_rdp(struct scsi_qla_host *vha,
[ ... ]
> +	if (IS_QLA29XX(ha)) {
> +		vp_idx = le16_to_cpu(
> +		    ((struct purex_entry_24xx_ext *)purex)->vp_idx);
> +		purex_sz = sizeof(struct purex_entry_24xx_ext);

[Severity: High]
Similarly, does extracting the 16-bit vp_idx and storing it in a uint8_t
local variable truncate the value before assigning it to the response IOCB?

[ ... ]
> @@ -6257,7 +6270,7 @@ void qla24xx_process_purex_rdp(struct scsi_qla_host *vha,
>  	rsp_els->handle = 0;
>  	rsp_els->nport_handle = purex->nport_handle;
>  	rsp_els->tx_dsd_count = cpu_to_le16(1);
> -	rsp_els->vp_index = purex->vp_idx;
> +	rsp_els->vp_index = vp_idx;
>  	rsp_els->sof_type = EST_SOFI3;
>  	rsp_els->rx_xchg_address = purex->rx_xchg_addr;

[Severity: High]
Will writing to rsp_els->vp_index and rsp_els->sof_type byte-by-byte corrupt
the packed bitfield on 29xx hardware?

In the 24xx struct view, these are separate bytes, but on 29xx they overlap
with the __le16 vp_idx_sof packed bitfield. Writing EST_SOFI3 to the 8-bit
sof_type could clear the 9th bit of vp_index and set the wrong frame
properties.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260723050413.3897522-1-njavali@marvell.com?part=28