Skip to contentSkip to conversation

drivers/net: support hdlc function for QE-UCC

bug report Dan Carpenter 2 participants 3 messages Jul 12, 2016 – Aug 3, 2016 Kernel janitors

lore
Share & tools
lore
#0 lore raw

Hello Zhao Qiang,

The patch c19b6d246a35: "drivers/net: support hdlc function for QE-UCC" from Jun 6, 2016, leads to the following static checker warning:

	drivers/net/wan/fsl_ucc_hdlc.c:1131 ucc_hdlc_probe()
	error: dereferencing freed memory 'uhdlc_priv'

The error handling has quite a few problems. There is an easy way to write error handling. Let me demonstrate.

drivers/net/wan/fsl_ucc_hdlc.c
1062 ret = of_address_to_resource(np, 0, &res);
1063 if (ret)
1064 return -EINVAL;
1065
1066 ut_info->uf_info.regs = res.start;
1067 ut_info->uf_info.irq = irq_of_parse_and_map(np, 0);
1068
1069 uhdlc_priv = kzalloc(sizeof(*uhdlc_priv), GFP_KERNEL);
1070 if (!uhdlc_priv) {
1071 ret = -ENOMEM;
1072 dev_err(&pdev->dev, "No mem to alloc hdlc private data\n");

No need for this printk(). kzalloc() already prints a warning.

1073 goto err_alloc_priv;

This is a do-nothing goto. Change it to "return -ENOMEM;".

		uhdlc_priv = kzalloc(sizeof(*uhdlc_priv), GFP_KERNEL);
		if (!uhdlc_priv)
			return -ENOMEM;
  1074          }
  1075  
  1076          dev_set_drvdata(&pdev->dev, uhdlc_priv);
  1077          uhdlc_priv->dev = &pdev->dev;
  1078          uhdlc_priv->ut_info = ut_info;
  1079  
  1080          if (of_get_property(np, "fsl,tdm-interface", NULL))
  1081                  uhdlc_priv->tsa = 1;
  1082  
  1083          if (of_get_property(np, "fsl,ucc-internal-loopback", NULL))
  1084                  uhdlc_priv->loopback = 1;
  1085  
  1086          if (uhdlc_priv->tsa = 1) {
  1087                  utdm = kzalloc(sizeof(*utdm), GFP_KERNEL);
  1088                  if (!utdm) {
  1089                          ret = -ENOMEM;
  1090                          dev_err(&pdev->dev, "No mem to alloc ucc tdm data\n");

No need again.

1091 goto err_alloc_utdm;

The goto name should say what the goto does. We need to free uhdlc_priv so "goto free_uhdlc;" is a good name.

  1092                  }
  1093                  uhdlc_priv->utdm = utdm;
  1094                  ret = ucc_of_parse_tdm(np, utdm, ut_info);
  1095                  if (ret)

1096 goto err_miss_tsa_property;

Here we need to free utdm so "goto free_utdm;".

  1097          }
  1098  
  1099          ret = uhdlc_init(uhdlc_priv);
  1100          if (ret) {
  1101                  dev_err(&pdev->dev, "Failed to init uhdlc\n");
  1102                  goto err_hdlc_init;

We're still freeing utdm; So goto free_utdm;

  1103          }
  1104  
  1105          dev = alloc_hdlcdev(uhdlc_priv);
  1106          if (!dev) {
  1107                  ret = -ENOMEM;
  1108                  pr_err("ucc_hdlc: unable to allocate memory\n");
  1109                  goto err_hdlc_init;

Hm... Here we need to undo what uhdlc_init() did. I'm not sure the function to do that. Currently we leak resources.

			goto undo_uhdlc_init;
  1110          }
  1111  
  1112          uhdlc_priv->ndev = dev;
  1113          hdlc = dev_to_hdlc(dev);
  1114          dev->tx_queue_len = 16;
  1115          dev->netdev_ops = &uhdlc_ops;
  1116          hdlc->attach = ucc_hdlc_attach;
  1117          hdlc->xmit = ucc_hdlc_tx;
  1118          netif_napi_add(dev, &uhdlc_priv->napi, ucc_hdlc_poll, 32);
  1119          if (register_hdlc_device(dev)) {
  1120                  ret = -ENOBUFS;

Why are we not preserving the error code?

		ret = register_hdlc_device(dev);
		if (ret) {
  1121                  pr_err("ucc_hdlc: unable to register hdlc device\n");

No need for this printk.

  1122                  free_netdev(dev);
  1123                  goto err_hdlc_init;

Instead of freeing and then a goto just do "goto free_dev;".

  1124          }
  1125  
  1126          return 0;
  1127  
  1128  err_hdlc_init:
  1129  err_miss_tsa_property:
  1130          kfree(uhdlc_priv);
  1131          if (uhdlc_priv->tsa)
  1132                  kfree(utdm);
  1133  err_alloc_utdm:
  1134          kfree(uhdlc_priv);
  1135  err_alloc_priv:
  1136          return ret;
  1137  }

You can see the problems. We free uhdlc_priv then we dereference it, then we free it a second time. There is stuff that is missing. The labels are bad. Here is the new version:

free_dev:

	free_netdev(dev);
undo_uhdlc_init:
	/* FIXME: impliment this code */
free_utdm:
	if (uhdlc_priv->tsa)
		kfree(utdm);
free_uhdlc:
	kfree(uhdlc);

	return ret;

Just follow the steps and then error handling is easy to understand.

regards,
dan carpenter

#1 lore raw

Hello Dan Carpenter,

Thank you for your comments, I will send a patch to fix it.

Best Regards
Zhao Qiang

Quoted 159 lines

-----Original Message-----
From: Dan Carpenter [mailto:[email protected]]
Sent: Wednesday, July 13, 2016 7:49 AM
To: Qiang Zhao <[email protected]>
Cc: [email protected]
Subject: [bug report] drivers/net: support hdlc function for QE-UCC

Hello Zhao Qiang,

The patch c19b6d246a35: "drivers/net: support hdlc function for QE-UCC" from Jun 6, 2016, leads to the following static checker
warning:

	drivers/net/wan/fsl_ucc_hdlc.c:1131 ucc_hdlc_probe()
	error: dereferencing freed memory 'uhdlc_priv'

The error handling has quite a few problems. There is an easy way to write error handling. Let me demonstrate.

drivers/net/wan/fsl_ucc_hdlc.c
1062 ret = of_address_to_resource(np, 0, &res);
1063 if (ret)
1064 return -EINVAL;
1065
1066 ut_info->uf_info.regs = res.start;
1067 ut_info->uf_info.irq = irq_of_parse_and_map(np, 0);
1068
1069 uhdlc_priv = kzalloc(sizeof(*uhdlc_priv), GFP_KERNEL);
1070 if (!uhdlc_priv) {
1071 ret = -ENOMEM;
1072 dev_err(&pdev->dev, "No mem to alloc hdlc private data\n");

No need for this printk(). kzalloc() already prints a warning.

1073 goto err_alloc_priv;

This is a do-nothing goto. Change it to "return -ENOMEM;".

		uhdlc_priv = kzalloc(sizeof(*uhdlc_priv), GFP_KERNEL);
		if (!uhdlc_priv)
			return -ENOMEM;
  1074          }
  1075
  1076          dev_set_drvdata(&pdev->dev, uhdlc_priv);
  1077          uhdlc_priv->dev = &pdev->dev;
  1078          uhdlc_priv->ut_info = ut_info;
  1079
  1080          if (of_get_property(np, "fsl,tdm-interface", NULL))
  1081                  uhdlc_priv->tsa = 1;
  1082
  1083          if (of_get_property(np, "fsl,ucc-internal-loopback", NULL))
  1084                  uhdlc_priv->loopback = 1;
  1085
  1086          if (uhdlc_priv->tsa = 1) {
  1087                  utdm = kzalloc(sizeof(*utdm), GFP_KERNEL);
  1088                  if (!utdm) {
  1089                          ret = -ENOMEM;
  1090                          dev_err(&pdev->dev, "No mem to alloc ucc tdm data\n");

No need again.

1091 goto err_alloc_utdm;

The goto name should say what the goto does. We need to free uhdlc_priv so "goto free_uhdlc;" is a good name.

  1092                  }
  1093                  uhdlc_priv->utdm = utdm;
  1094                  ret = ucc_of_parse_tdm(np, utdm, ut_info);
  1095                  if (ret)

1096 goto err_miss_tsa_property;

Here we need to free utdm so "goto free_utdm;".

  1097          }
  1098
  1099          ret = uhdlc_init(uhdlc_priv);
  1100          if (ret) {
  1101                  dev_err(&pdev->dev, "Failed to init uhdlc\n");
  1102                  goto err_hdlc_init;

We're still freeing utdm; So goto free_utdm;

  1103          }
  1104
  1105          dev = alloc_hdlcdev(uhdlc_priv);
  1106          if (!dev) {
  1107                  ret = -ENOMEM;
  1108                  pr_err("ucc_hdlc: unable to allocate memory\n");
  1109                  goto err_hdlc_init;

Hm... Here we need to undo what uhdlc_init() did. I'm not sure the function to do that. Currently we leak resources.

			goto undo_uhdlc_init;
  1110          }
  1111
  1112          uhdlc_priv->ndev = dev;
  1113          hdlc = dev_to_hdlc(dev);
  1114          dev->tx_queue_len = 16;
  1115          dev->netdev_ops = &uhdlc_ops;
  1116          hdlc->attach = ucc_hdlc_attach;
  1117          hdlc->xmit = ucc_hdlc_tx;
  1118          netif_napi_add(dev, &uhdlc_priv->napi, ucc_hdlc_poll, 32);
  1119          if (register_hdlc_device(dev)) {
  1120                  ret = -ENOBUFS;

Why are we not preserving the error code?

		ret = register_hdlc_device(dev);
		if (ret) {
  1121                  pr_err("ucc_hdlc: unable to register hdlc device\n");

No need for this printk.

  1122                  free_netdev(dev);
  1123                  goto err_hdlc_init;

Instead of freeing and then a goto just do "goto free_dev;".

  1124          }
  1125
  1126          return 0;
  1127
  1128  err_hdlc_init:
  1129  err_miss_tsa_property:
  1130          kfree(uhdlc_priv);
  1131          if (uhdlc_priv->tsa)
  1132                  kfree(utdm);
  1133  err_alloc_utdm:
  1134          kfree(uhdlc_priv);
  1135  err_alloc_priv:
  1136          return ret;
  1137  }

You can see the problems. We free uhdlc_priv then we dereference it, then we free it a second time. There is stuff that is missing. The labels are bad. Here is the new version:

free_dev:

	free_netdev(dev);
undo_uhdlc_init:
	/* FIXME: impliment this code */
free_utdm:
	if (uhdlc_priv->tsa)
		kfree(utdm);
free_uhdlc:
	kfree(uhdlc);

	return ret;

Just follow the steps and then error handling is easy to understand.

regards,
dan carpenter

#2 lore raw
In reply to Dan Carpenter

Hello Zhao Qiang,

The patch c19b6d246a35: "drivers/net: support hdlc function for QE-UCC" from Jun 6, 2016, leads to the following static checker warning:

	drivers/net/wan/fsl_ucc_hdlc.c:164 uhdlc_init()
	warn: unsigned 'priv->ucc_pram_offset' is never less than zero.

drivers/net/wan/fsl_ucc_hdlc.c
159
160 /* Alloc parameter ram for ucc hdlc */
161 priv->ucc_pram_offset = qe_muram_alloc(sizeof(priv->ucc_pram),
162 ALIGNMENT_OF_UCC_HDLC_PRAM);
163
164 if (priv->ucc_pram_offset < 0) {

This is slightly complicated what's happening here. qe_muram_alloc() returns negatives but casted to an unsigned long, then we assign it to priv->ucc_pram_offset which is a u32.

   165                  dev_err(priv->dev, "Can not allocate MURAM for hdlc prameter.\n");
   166                  ret = -ENOMEM;
   167                  goto free_tx_bd;

168 }
169

regards,
dan carpenter