Skip to contentSkip to conversation

nvme: Ensure char members of struct nvme_id_ctrl are null-terminated

PATCH In review Erwan Velu 4 participants 7 messages Jun 14, 2022 – Jun 15, 2022 1 message not in the archive NVMe

lore
Share & tools
lore

Erwan Velu

author
#0 lore raw

PATCH 1/2 nvme: Ensure char members of struct nvme_id_ctrl are null-terminated

When issueing an identify command, the resulting data structure stored as a nvme_id_ctrl struct, contains a couple of char members.

These char members features:
- the serial number (sn)
- the model number (mn)
- the firmware revision (fr)

As per the current code, nothing enforce these members to be null-terminated which causes issues when manipulating them with regular string manupulation functions.

This commit ensure the three of them are always null-terminated.

---
 drivers/nvme/host/core.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/nvme/host/core.c +7 -0
index 24165daee3c8..0f7e625e8bd0 100644
@@ -1274,6 +1274,13 @@ static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 			sizeof(struct nvme_id_ctrl));
 	if (error)
 		kfree(*id);
+	else {
+		/* Ensure that model, serial and firmware fields are always null-terminated */
+		(*id)->mn[sizeof((*id)->mn)-1] = 0;
+		(*id)->sn[sizeof((*id)->sn)-1] = 0;
+		(*id)->fr[sizeof((*id)->fr)-1] = 0;
+	}
+
 	return error;
 }
 
Signature · 1 line
2.35.3
author
#1 lore raw

PATCH 2/2 nvme: Report model,sn,fw,pci device information during init

SCSI-based device get their identify properties being printed when initialized like :

[    1.245357] scsi 0:0:0:0: Direct-Access     ATA      HGST HTE721010A9 A3M0 PQ: 0 ANSI: 5

When initializing nvme devices, no identification message is reported making difficult to identify them during the boot process. If the system crashes during boot process or if the init phase fail, it could be very diffcult to identify the faulty disk.

This patch reports model, serial, firmware version and pci information as soon as possible making this early identifying task possible.

A typical output looks like:

[    0.383353] nvme nvme0: pci function 0000:00:03.0
[    0.418184] nvme nvme0: MODEL:QEMU NVMe Ctrl                          SN:deadbeef            FW:1.0     PCI_ID:1b36:1af4
[    0.422020] nvme nvme0: 1/0/0 default/read/poll queues
---
 drivers/nvme/host/core.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/nvme/host/core.c +8 -0
index 0f7e625e8bd0..f73ca92412a9 100644
@@ -2982,6 +2982,14 @@ static int nvme_init_identify(struct nvme_ctrl *ctrl)
 		return -EIO;
 	}
 
+	/* Reporting model, serial, firmware and pci info */
+	dev_info(ctrl->device, "MODEL:%s SN:%s FW:%s PCI_ID:%04x:%04x\n",
+			id->mn,
+			id->sn,
+			id->fr,
+			le16_to_cpu(id->vid),
+			le16_to_cpu(id->ssvid));
+
 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
 		if (ret < 0)
Signature · 1 line
2.35.3
maintainer
#2 lore raw

Re: [PATCH 2/2] nvme: Report model,sn,fw,pci device information during init

drivers/nvme/host/core.c static int nvme_init_identify(struct nvme_ctrl *ctrl) · line 2991

Quoted 11 lines · Erwan Velu
drivers/nvme/host/core.c +7 -0
@@ -2982,6 +2982,14 @@ static int nvme_init_identify(struct nvme_ctrl *ctrl)
 		return -EIO;
 	}
 
+	/* Reporting model, serial, firmware and pci info */
+	dev_info(ctrl->device, "MODEL:%s SN:%s FW:%s PCI_ID:%04x:%04x\n",
+			id->mn,
+			id->sn,
+			id->fr,
+			le16_to_cpu(id->vid),
+			le16_to_cpu(id->ssvid));

We don't need to print this on every controller reset, so I think this needs to be within the "if (!ctrl->identified)" block.

And since you can't just null terminate the these strings, you should use the "%*s" format.

I'm unsure if the serial number should be logged. Firmware and model are probably fine.

#3 lore raw

Re: [PATCH 2/2] nvme: Report model,sn,fw,pci device information during init

Quoted 2 lines · Keith Busch

I'm unsure if the serial number should be logged. Firmware and model are probably fine.

I would much prefer to not print it all. The Linux boot is way to spammy to start with and this doesn't add any real value.

A message that is not in the archive (sent off-list or not yet received): CAL2JzuwnZhMgaVQ0=LAYfe6pWnxGLgR_b4xdF_==F3vL_Hh=gw@mail.gmail.com

#5 lore raw

Re: [PATCH 2/2] nvme: Report model,sn,fw,pci device information during init

Quoted 8 lines · Erwan Velu
Quoted 3 lines

I would much prefer to not print it all. The Linux boot is way to spammy to start with and this doesn't add any real value.

I know the boot is a bit spammy but when systems crash because of nvme drives, that's very handy to get a trace showing who was the culprit and what drive is installed (including the fw version which is usually a major source of troubles).

Well, usually they are all the same or a very small set of different SKUs compared to the total number of drives.

maintainer
#6 lore raw

Re: [PATCH 2/2] nvme: Report model,sn,fw,pci device information during init

Quoted 5 lines · Christoph Hellwig
Quoted 4 lines · Erwan Velu
Quoted 3 lines

I would much prefer to not print it all. The Linux boot is way to spammy to start with and this doesn't add any real value.

Quoted 7 lines
Quoted 4 lines

I know the boot is a bit spammy but when systems crash because of nvme drives, that's very handy to get a trace showing who was the culprit and what drive is installed (including the fw version which is usually a major source of troubles).

Well, usually they are all the same or a very small set of different SKUs compared to the total number of drives.

Agree that we don't want to add unnecessary spam to the boot messages, and we don't need tons of lines like eg SCSI does. But it does seem handy to have a single line of the basic model/fw/sn for each drive along with the enumeration we already print.

Signature · 1 line
Jens Axboe
maintainer
#7 lore raw
In reply to Erwan Velu

drivers/nvme/host/core.c static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) · line 1282

Quoted 6 lines · Erwan Velu
drivers/nvme/host/core.c +6 -0
+	else {
+		/* Ensure that model, serial and firmware fields are always null-terminated */
+		(*id)->mn[sizeof((*id)->mn)-1] = 0;
+		(*id)->sn[sizeof((*id)->sn)-1] = 0;
+		(*id)->fr[sizeof((*id)->fr)-1] = 0;
+	}

But the last byte here is owned by the controller. You may be truncating information.