Skip to contentSkip to conversation

ipmi procfs bogosity

RFC [email protected] 5 participants 9 messages Sep 1, 2005 – Sep 5, 2005

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

drivers/char/ipmi/ipmi_poweroff.c::proc_write_chassctrl()
a) does sscanf on userland pointer
b) does sscanf on array that is not guaranteed to have NUL in it
c) interprets input in incredibly cretinous way:
if strings doesn't start with a decimal number => as if it was "0". if it starts with decimal number equal to 0 (e.g. is "-0000splat") - as if it was "0".
if it starts with decimal number equal to 2 (e.g. is "00002FOAD") - as if it was "2".
otherwise - -EINVAL.
In any case that doesn't end up with -EINVAL, pretend that entire
buffer had been written.

(a) and (b) are immediate bugs; (c) is a valid reason for immediate severe LARTing of the pervert who had done _that_ in a user-visible API.

Note that API _is_ user-visible, so we can't blindly change it - not without checking WTF do its users actually write to /proc/ipmi/poweroff_control.

Could somebody comment on the actual uses of that FPOS? My preference would be to remove the damn thing completely - it's too ugly to live.

#1 lore raw

Indeed, this function is badly written. In rewriting, I couldn't find a nice function for reading integers from userspace, and the proc_dointvec stuff didn't seem terribly suitable. So I wrote my own general function, which I can move someplace else if someone likes. Patch is attached. It should not affect correct usage of this file.

Thank you for pointing this out.

Andrew, can you include this? Matt, can you test?

-Corey

Quoted 13 lines · On Thu, 2005-09-01 at 07:43 +0100, [email protected] wrote:

drivers/char/ipmi/ipmi_poweroff.c::proc_write_chassctrl()
a) does sscanf on userland pointer
b) does sscanf on array that is not guaranteed to have NUL in it
c) interprets input in incredibly cretinous way:
if strings doesn't start with a decimal number => as if it was "0". if it starts with decimal number equal to 0 (e.g. is "-0000splat") - as if it was "0".
if it starts with decimal number equal to 2 (e.g. is "00002FOAD") - as if it was "2".
otherwise - -EINVAL.
In any case that doesn't end up with -EINVAL, pretend that entire
buffer had been written.

Quoted 8 lines

(a) and (b) are immediate bugs; (c) is a valid reason for immediate severe LARTing of the pervert who had done _that_ in a user-visible API.

Note that API _is_ user-visible, so we can't blindly change it - not without checking WTF do its users actually write to /proc/ipmi/poweroff_control.

Could somebody comment on the actual uses of that FPOS? My preference would be to remove the damn thing completely - it's too ugly to live.

The IPMI power control function proc_write_chassctrl was badly written, it directly used userspace pointers, it assumed that strings were NULL terminated, and it used the evil sscanf function. This adds a new function to read integers from userspace and uses this function to get the integer in question.

drivers/char/ipmi/ipmi_poweroff.c +170 -16
Index: linux-2.6.13/drivers/char/ipmi/ipmi_poweroff.c
===================================================================
@@ -40,6 +40,8 @@
 #include <linux/kdev_t.h>
 #include <linux/ipmi.h>
 #include <linux/ipmi_smi.h>
+#include <linux/ctype.h>
+#include <asm/uaccess.h>
 
 #define PFX "IPMI poweroff: "
 
@@ -545,27 +547,179 @@ static int proc_read_chassctrl(char *pag
 			poweroff_control);
 }
 
+/*
+ * Convert a character to an integer, returning -1 if the character is
+ * not a valid digit for "base".
+ */
+static int convdigit(char ch, int base)
+{
+	int v;
+	if ((ch >= '0') && (ch <= '9'))
+		v = ch - '0';
+	else if ((ch >= 'A') && (ch <= 'Z'))
+		v = ch - 'A' + 10;
+	else if ((ch >= 'a') && (ch <= 'z'))
+		v = ch - 'a' + 10;
+	else
+		return -1;
+	if (v >= base)
+		return -1;
+	return v;
+}
+
+/*
+ * Get a unsigned long from userspace, much like strtoul, but getting
+ * the data from userspace.  This is a little complicated to use, but
+ * is quite flexible.
+ *
+ * @buffer - is a pointer to the start of the user buffer.  Basically,
+ * this will read from buffer+*pos up to buffer+count-1.
+ * @pos - points to the start offset in buffer to start reading at.
+ * pos will be updated to the current location after the scan upon return.
+ * May be NULL, where it is assumed to be zero.
+ * @count - the total buffer length.
+ * @fbase - forces the base.  If this value is zero 0, the base will be
+ * hex if the string starts with 0x, octal if it starts with 0, and
+ * decimal otherwise.
+ * @stop_at_inval - a flag that, if true, causes the function to
+ * return if it already has a valid number and has reached an invalid
+ * character.  "pos" will be set to the offset of the invalid
+ * character upon return.  If false, the entire buffer is expected to
+ * be a single integer value surrounded by optional space characters.
+ *
+ * Unlike strtoul, the value is returned in the "val" parameter.  The
+ * return value of this function is a errno if negative or the number
+ * of characters processed if positive.
+ */
+static ssize_t user_strtoul(const char __user *buffer, loff_t *ppos,
+			    size_t count, int fbase, int stop_at_inval,
+			    unsigned long *val)
+{
+	unsigned long newval = 0;
+	char          buf[20];
+	unsigned int  rc;
+	unsigned int  i;
+	enum { SC_ST, SC_ST2, SC_AFTPRE, SC_DAT, SC_END } state = SC_ST;
+	int           base = fbase;
+	loff_t        start;
+	loff_t        pos;
+
+	if (ppos)
+		pos = *ppos;
+	else
+		pos = 0;
+	start = pos;
+	count -= pos;
+	if (base == 0)
+		base = 10;
+	while (count > 0) {
+		rc = min(count, (size_t) sizeof(buf));
+		if (copy_from_user(buf, buffer + pos, rc))
+			return -EFAULT;
+		for (i = 0; i < rc; i++) {
+			char ch = buf[i];
+			int  chval;
+
+			switch(state) {
+			case SC_ST:
+				if (isspace(ch))
+					break;
+				if (ch == '0') {
+					if (fbase == 0)
+						base = 8;
+					state = SC_ST2;
+					break;
+				}
+				goto aftpre;
+
+			case SC_ST2:
+				if (ch == 'x') {
+					if (fbase == 0)
+						base = 16;
+					else if (fbase != 16)
+						return -EINVAL;
+					state = SC_AFTPRE;
+					break;
+				}
+				goto aftpre;
+
+			case SC_AFTPRE:
+			aftpre:
+				chval = convdigit(ch, base);
+				if (chval == -1)
+					return -EINVAL;
+				newval = (newval * base) + chval;
+				state = SC_DAT;
+				break;
+
+			case SC_DAT:
+				if (isspace(ch)) {
+					if (stop_at_inval) {
+						pos += i;
+						goto out;
+					}
+					state = SC_END;
+				} else {
+					unsigned long oldval = newval;
+					chval = convdigit(ch, base);
+					if (chval == -1) {
+						if (stop_at_inval) {
+							pos += i;
+							goto out;
+						}
+						return -EINVAL;
+					}
+					newval *= base;
+					if (newval < oldval) /* overflow */
+						return -EINVAL;
+					oldval = newval;
+					newval += chval;
+					if (newval < oldval) /* overflow */
+						return -EINVAL;
+				}
+				break;
+
+			case SC_END:
+				if (!isspace(ch))
+					return -EINVAL;
+				break;
+			}
+		}
+
+		count -= rc;
+		pos += rc;
+	}
+
+ out:
+	if (ppos)
+		*ppos = pos;
+	*val = newval;
+	return pos - start;
+}
+
 /* process property writes from proc */
-static int proc_write_chassctrl(struct file *file, const char *buffer,
+static int proc_write_chassctrl(struct file *file, const char __user *buffer,
 			        unsigned long count, void *data)
 {
-	int          rv = count;
-	unsigned int newval = 0;
+	unsigned long newval;
+	ssize_t       rv;
 
-	sscanf(buffer, "%d", &newval);
+	rv = user_strtoul(buffer, NULL, count, 0, 0, &newval);
+	if (rv < 0)
+		return rv;
 	switch (newval) {
-		case IPMI_CHASSIS_POWER_CYCLE:
-			printk(KERN_INFO PFX "power cycle is now enabled\n");
-			poweroff_control = newval;
-			break;
-
-		case IPMI_CHASSIS_POWER_DOWN:
-			poweroff_control = IPMI_CHASSIS_POWER_DOWN;
-			break;
-
-		default:
-			rv = -EINVAL;
-			break;
+	case IPMI_CHASSIS_POWER_CYCLE:
+		printk(KERN_INFO PFX "power cycle is now enabled\n");
+		poweroff_control = newval;
+		break;
+
+	case IPMI_CHASSIS_POWER_DOWN:
+		poweroff_control = IPMI_CHASSIS_POWER_DOWN;
+		break;
+
+	default:
+		rv = -EINVAL;
+		break;
 	}
 
 	return rv;
#2 lore raw
Quoted 5 lines · Corey Minyard

Indeed, this function is badly written. In rewriting, I couldn't find a nice function for reading integers from userspace, and the proc_dointvec stuff didn't seem terribly suitable. So I wrote my own general function, which I can move someplace else if someone likes. Patch is attached. It should not affect correct usage of this file.

Eeeek... Much, _much_ simpler approach would be to have

	char buf[10];
	if (count > 9)
		return -EINVAL;
	if (copy_from_user(buf, buffer, count))
		return -EFAULT;
	buf[count] = '\0';
	/* use sscanf() or anything normal */

Would that change behaviour in any cases you care about?

#7 lore raw
In reply to Corey Minyard
Quoted 4 lines · Corey Minyard

Indeed, this function is badly written. In rewriting, I couldn't find a
nice function for reading integers from userspace, and the proc_dointvec stuff didn't seem terribly suitable.

We write numbers into profs files all the time. Is there something different about the IPMI requirement which makes the approach used by, say, dirty_writeback_centisecs_handler() inappropriate?

#3 lore raw
In reply to [email protected]
Quoted 21 lines · [email protected] wrote:
Quoted 7 lines · Corey Minyard

Indeed, this function is badly written. In rewriting, I couldn't find a nice function for reading integers from userspace, and the proc_dointvec stuff didn't seem terribly suitable. So I wrote my own general function, which I can move someplace else if someone likes. Patch is attached. It should not affect correct usage of this file.

Eeeek... Much, _much_ simpler approach would be to have

	char buf[10];
	if (count > 9)
		return -EINVAL;
	if (copy_from_user(buf, buffer, count))
		return -EFAULT;
	buf[count] = '\0';
	/* use sscanf() or anything normal */
Quoted 3 lines

Would that change behaviour in any cases you care about?

Because then, for a general solution that avoids integer perversion, you need something like:

   char         buf[10];
   char         *end;

   if (count > (sizeof(buf) - 1))
       return -EINVAL;
   if (copy_from_user(buf, buffer, count))
       return -EFAULT;
   buf[count] = '\0';
   newval = simple_strtoul(buf, &end, 0);
   if (buf == end)
       /* Empty string or first char not a number */
       return -EINVAL;
   if (*end && ! isspace(*end))
       /* Bogus number. */
       return -EINVAL;

To me, It's a lot nicer to do:

   rv = user_strtoul(....);
   if (rv < 0)
       return rv;

Plus the scanning function I wrote handles arbitrary leading and trailing space, etc. Not a big deal, but a little nicer.

-Corey

#4 lore raw
Quoted 8 lines · Corey Minyard

To me, It's a lot nicer to do:

   rv = user_strtoul(....);
   if (rv < 0)
       return rv;

Plus the scanning function I wrote handles arbitrary leading and trailing space, etc. Not a big deal, but a little nicer.

You can say from the beggining that

echo -n " 2 " >/proc/FUBAR

is illegal and don't add bloat to kernel.

#8 lore raw
In reply to Andrew Morton
Quoted 9 lines · Andrew Morton
Quoted 5 lines · Corey Minyard

Indeed, this function is badly written. In rewriting, I couldn't find a nice function for reading integers from userspace, and the proc_dointvec stuff didn't seem terribly suitable.

Quoted 5 lines

We write numbers into profs files all the time. Is there something different about the IPMI requirement which makes the approach used by, say, dirty_writeback_centisecs_handler() inappropriate?

Ok, that's probably better, and this probably belongs in /proc/sys/dev/ipmi. This is new enough that it doesn't matter, I don't think any one is using it yet.

Patch is attached.

-Corey

The IPMI power control function proc_write_chassctrl was badly written, it directly used userspace pointers, it assumed that strings were NULL terminated, and it used the evil sscanf function. This converts over to using the sysctl interface for this data and changes the semantics to be a little more logical.

 Documentation/IPMI.txt            |   13 +--
 drivers/char/ipmi/ipmi_poweroff.c |  136 ++++++++++++++++----------------------
 include/linux/sysctl.h            |    6 +
 3 files changed, 73 insertions(+), 82 deletions(-)
drivers/char/ipmi/ipmi_poweroff.c +60 -76
Index: linux-2.6.13/drivers/char/ipmi/ipmi_poweroff.c
===================================================================
@@ -52,11 +52,11 @@ extern void (*pm_power_off)(void);
 #define IPMI_CHASSIS_POWER_CYCLE	0x02	/* power cycle */
 
 /* the IPMI data command */
-static int poweroff_control = IPMI_CHASSIS_POWER_DOWN;
+static int poweroff_powercycle = 0;
 
 /* parameter definition to allow user to flag power cycle */
-module_param(poweroff_control, int, IPMI_CHASSIS_POWER_DOWN);
-MODULE_PARM_DESC(poweroff_control, " Set to 2 to enable power cycle instead of power down. Power cycle is contingent on hardware support, otherwise it defaults back to power down.");
+module_param(poweroff_powercycle, int, 0);
+MODULE_PARM_DESC(poweroff_powercycles, " Set to non-zero to enable power cycle instead of power down. Power cycle is contingent on hardware support, otherwise it defaults back to power down.");
 
 /* Stuff from the get device id command. */
 static unsigned int mfg_id;
@@ -365,37 +365,34 @@ static void ipmi_poweroff_chassis (ipmi_
 
  powercyclefailed:
 	printk(KERN_INFO PFX "Powering %s via IPMI chassis control command\n",
-		((poweroff_control != IPMI_CHASSIS_POWER_CYCLE) ? "down" : "cycle"));
+		(poweroff_powercycle ? "cycle" : "down"));
 
 	/*
 	 * Power down
 	 */
 	send_msg.netfn = IPMI_NETFN_CHASSIS_REQUEST;
 	send_msg.cmd = IPMI_CHASSIS_CONTROL_CMD;
-	data[0] = poweroff_control;
+	if (poweroff_powercycle)
+		data[0] = IPMI_CHASSIS_POWER_CYCLE;
+	else
+		data[0] = IPMI_CHASSIS_POWER_DOWN;
 	send_msg.data = data;
 	send_msg.data_len = sizeof(data);
 	rv = ipmi_request_in_rc_mode(user,
 				     (struct ipmi_addr *) &smi_addr,
 				     &send_msg);
 	if (rv) {
-		switch (poweroff_control) {
-			case IPMI_CHASSIS_POWER_CYCLE:
-				/* power cycle failed, default to power down */
-				printk(KERN_ERR PFX "Unable to send chassis power " \
-					"cycle message, IPMI error 0x%x\n", rv);
-				poweroff_control = IPMI_CHASSIS_POWER_DOWN;
-				goto powercyclefailed;
-
-			case IPMI_CHASSIS_POWER_DOWN:
-			default:
-				printk(KERN_ERR PFX "Unable to send chassis power " \
-					"down message, IPMI error 0x%x\n", rv);
-				break;
+		if (poweroff_powercycle) {
+			/* power cycle failed, default to power down */
+			printk(KERN_ERR PFX "Unable to send chassis power " \
+			       "cycle message, IPMI error 0x%x\n", rv);
+			poweroff_powercycle = 0;
+			goto powercyclefailed;
 		}
-	}
 
-	return;
+		printk(KERN_ERR PFX "Unable to send chassis power " \
+		       "down message, IPMI error 0x%x\n", rv);
+	}
 }
 
 
@@ -537,39 +534,35 @@ static struct ipmi_smi_watcher smi_watch
 
 
 #ifdef CONFIG_PROC_FS
-/* displays properties to proc */
-static int proc_read_chassctrl(char *page, char **start, off_t off, int count,
-			       int *eof, void *data)
-{
-	return sprintf(page, "%d\t[ 0=powerdown 2=powercycle ]\n",
-			poweroff_control);
-}
-
-/* process property writes from proc */
-static int proc_write_chassctrl(struct file *file, const char *buffer,
-			        unsigned long count, void *data)
-{
-	int          rv = count;
-	unsigned int newval = 0;
-
-	sscanf(buffer, "%d", &newval);
-	switch (newval) {
-		case IPMI_CHASSIS_POWER_CYCLE:
-			printk(KERN_INFO PFX "power cycle is now enabled\n");
-			poweroff_control = newval;
-			break;
-
-		case IPMI_CHASSIS_POWER_DOWN:
-			poweroff_control = IPMI_CHASSIS_POWER_DOWN;
-			break;
-
-		default:
-			rv = -EINVAL;
-			break;
-	}
+#include <linux/sysctl.h>
 
-	return rv;
-}
+static ctl_table ipmi_table[] = {
+	{ .ctl_name	= DEV_IPMI_POWEROFF_POWERCYCLE,
+	  .procname	= "poweroff_powercycle",
+	  .data		= &poweroff_powercycle,
+	  .maxlen	= sizeof(poweroff_powercycle),
+	  .mode		= 0644,
+	  .proc_handler	= &proc_dointvec },
+	{ }
+};
+
+static ctl_table ipmi_dir_table[] = {
+	{ .ctl_name	= DEV_IPMI,
+	  .procname	= "ipmi",
+	  .mode		= 0555,
+	  .child	= ipmi_table },
+	{ }
+};
+
+static ctl_table ipmi_root_table[] = {
+	{ .ctl_name	= CTL_DEV,
+	  .procname	= "dev",
+	  .mode		= 0555,
+	  .child	= ipmi_dir_table },
+	{ }
+};
+
+static struct ctl_table_header *ipmi_table_header;
 #endif /* CONFIG_PROC_FS */
 
 /*
@@ -577,41 +570,32 @@ static int proc_write_chassctrl(struct f
  */
 static int ipmi_poweroff_init (void)
 {
-	int                   rv;
-	struct proc_dir_entry *file;
+	int rv;
 
 	printk ("Copyright (C) 2004 MontaVista Software -"
 		" IPMI Powerdown via sys_reboot.\n");
 
-	switch (poweroff_control) {
-		case IPMI_CHASSIS_POWER_CYCLE:
-			printk(KERN_INFO PFX "Power cycle is enabled.\n");
-			break;
-
-		case IPMI_CHASSIS_POWER_DOWN:
-		default:
-			poweroff_control = IPMI_CHASSIS_POWER_DOWN;
-			break;
+	if (poweroff_powercycle)
+		printk(KERN_INFO PFX "Power cycle is enabled.\n");
+
+#ifdef CONFIG_PROC_FS
+	ipmi_table_header = register_sysctl_table(ipmi_root_table, 1);
+	if (!ipmi_table_header) {
+		printk(KERN_ERR PFX "Unable to register powercycle sysctl\n");
+		rv = -ENOMEM;
+		goto out_err;
 	}
+#endif
 
+#ifdef CONFIG_PROC_FS
 	rv = ipmi_smi_watcher_register(&smi_watcher);
+#endif
 	if (rv) {
+		unregister_sysctl_table(ipmi_table_header);
 		printk(KERN_ERR PFX "Unable to register SMI watcher: %d\n", rv);
 		goto out_err;
 	}
 
-#ifdef CONFIG_PROC_FS
-	file = create_proc_entry("poweroff_control", 0, proc_ipmi_root);
-	if (!file) {
-		printk(KERN_ERR PFX "Unable to create proc power control\n");
-	} else {
-		file->nlink = 1;
-		file->read_proc = proc_read_chassctrl;
-		file->write_proc = proc_write_chassctrl;
-		file->owner = THIS_MODULE;
-	}
-#endif
-
  out_err:
 	return rv;
 }
@@ -622,7 +606,7 @@ static __exit void ipmi_poweroff_cleanup
 	int rv;
 
 #ifdef CONFIG_PROC_FS
-	remove_proc_entry("poweroff_control", proc_ipmi_root);
+	unregister_sysctl_table(ipmi_table_header);
 #endif
 
 	ipmi_smi_watcher_unregister(&smi_watcher);
Documentation/IPMI.txt +7 -6
Index: linux-2.6.13/Documentation/IPMI.txt
===================================================================
@@ -605,12 +605,13 @@ is in the ipmi_poweroff module.  When th
 it will send the proper IPMI commands to do this.  This is supported on
 several platforms.
 
-There is a module parameter named "poweroff_control" that may either be zero
-(do a power down) or 2 (do a power cycle, power the system off, then power
-it on in a few seconds).  Setting ipmi_poweroff.poweroff_control=x will do
-the same thing on the kernel command line.  The parameter is also available
-via the proc filesystem in /proc/ipmi/poweroff_control.  Note that if the
-system does not support power cycling, it will always to the power off.
+There is a module parameter named "poweroff_powercycle" that may
+either be zero (do a power down) or non-zero (do a power cycle, power
+the system off, then power it on in a few seconds).  Setting
+ipmi_poweroff.poweroff_control=x will do the same thing on the kernel
+command line.  The parameter is also available via the proc filesystem
+in /proc/sys/dev/ipmi/poweroff_powercycle.  Note that if the system
+does not support power cycling, it will always do the power off.
 
 Note that if you have ACPI enabled, the system will prefer using ACPI to
 power off.
include/linux/sysctl.h +6 -0
Index: linux-2.6.13/include/linux/sysctl.h
===================================================================
@@ -711,6 +711,7 @@ enum {
 	DEV_RAID=4,
 	DEV_MAC_HID=5,
 	DEV_SCSI=6,
+	DEV_IPMI=7,
 };
 
 /* /proc/sys/dev/cdrom */
@@ -776,6 +777,11 @@ enum {
 	DEV_SCSI_LOGGING_LEVEL=1,
 };
 
+/* /proc/sys/dev/ipmi */
+enum {
+	DEV_IPMI_POWEROFF_POWERCYCLE=1,
+};
+
 /* /proc/sys/abi */
 enum
 {
#5 lore raw
In reply to Alexey Dobriyan

Hi,

Quoted 9 lines · Alexey Dobriyan
Quoted 2 lines · Corey Minyard

Plus the scanning function I wrote handles arbitrary leading and trailing space, etc. Not a big deal, but a little nicer.

You can say from the beggining that

echo -n " 2 " >/proc/FUBAR

is illegal and don't add bloat to kernel.

No, user interfaces should be robust.
Just remember the mantra "Be liberal in what you accept and conservative in what you send."

I would suggest adding sth. like Coreys user_strtoul() to lib/string.c which would reduce bloat and security threats for the kernel.

Regards

Ingo Oeser

#6 lore raw
Quoted 1 line · Ingo Oeser

I would suggest adding sth. like Coreys user_strtoul()

strtoul_from_user

Quoted 1 line

to lib/string.c

lib/vsprintf.c

Quoted 1 line

which would reduce bloat and security threats for the kernel.