Skip to contentSkip to conversation

KASAN: slab-out-of-bounds Read in getname_kernel

PATCH In review tomas 3 participants 10 messages Jul 1, 2018 – Jul 3, 2018 Kernel automounter

lore
Share & tools
lore

tomas

author
#0 lore raw

PATCH 1/1 KASAN: slab-out-of-bounds Read in getname_kernel

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425b1163

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_dev-ioctl.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c    2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c    2018-07-01 23:10:16.059728621 +0200 +0 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s

             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */ +    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD) +        return -EINVAL;

     err = 0;
 out:

Tested and solves the issue on Linus' main git tree.

Tomas

maintainer
#1 lore raw
Quoted 5 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425b1163

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 20 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_dev-ioct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200 +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 8 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200 +0 -0
     err = 0;
 out:
 

Tested and solves the issue on Linus' main git tree.

Ian

maintainer
#2 lore raw
Quoted 42 lines · Ian Kent
Quoted 6 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 21 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_dev-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200 +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 15 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 8 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200 +0 -0
     err = 0;
 out:
 

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +7 -8
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct autofs_dev_ioctl *param)
 				cmd);
 			goto out;
 		}
+	} else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+		err = -EINVAL;
+		goto out;
 	}
 
 	err = 0;
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
 	dev_t devid;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	devid = sbi->sb->s_dev;
 
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
 	unsigned int devid, magic;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	name = param->path;
 	type = param->ismountpoint.in.type;
author
#3 lore raw

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

Tomas

Quoted 114 lines · Ian Kent
Quoted 39 lines · Ian Kent
Quoted 6 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 20 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_dev-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 15 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 8 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +0 -0
     err = 0;You should just put the "_IOC_NR" directive around "cmd" in the lines added to "validate_dev_ioctl" to make it work.
 out:
 

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +7 -8
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct autofs_dev_ioctl *param)
 				cmd);
 			goto out;
 		}
+	} else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+		err = -EINVAL;
+		goto out;
 	}
 
 	err = 0;
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
 	dev_t devid;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	devid = sbi->sb->s_dev;
 
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
 	unsigned int devid, magic;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	name = param->path;
 	type = param->ismountpoint.in.type;
maintainer
#4 lore raw
Quoted 6 lines · tomas

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing, with both our Signed-off-by added.

Quoted 130 lines

Tomas

Quoted 123 lines · Ian Kent
Quoted 44 lines · Ian Kent
Quoted 7 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 22 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 16 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 9 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +0 -0
     err = 0;You should just put the "_IOC_NR" directive around "cmd" in

the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
 				cmd);
 			goto out;
 		}
+	} else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+		err = -EINVAL;
+		goto out;
 	}
 
fs/autofs/dev-ioctl.c +2 -8
 	err = 0;
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
 	dev_t devid;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	devid = sbi->sb->s_dev;
 
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file
 *fp,
 	unsigned int devid, magic;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	name = param->path;

type = param->ismountpoint.in.type;

author
#5 lore raw

Yes, thanks. Please use my full name, Tomas Bortoli.

Cheers

Quoted 136 lines · Ian Kent
Quoted 6 lines · tomas

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing, with both our Signed-off-by added.

Quoted 124 lines

Tomas

Quoted 119 lines · Ian Kent
Quoted 41 lines · Ian Kent
Quoted 7 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 21 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 16 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 9 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +0 -0
     err = 0;You should just put the "_IOC_NR" directive around "cmd" in

the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
 				cmd);
 			goto out;
 		}
+	} else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+		   cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+		err = -EINVAL;
+		goto out;
 	}
 
fs/autofs/dev-ioctl.c +2 -8
 	err = 0;
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
 	dev_t devid;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	devid = sbi->sb->s_dev;
 
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file
 *fp,
 	unsigned int devid, magic;
 	int err = -ENOENT;
 
-	if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-		err = -EINVAL;
-		goto out;
-	}
+	/* param->path has already been checked */
 
 	name = param->path;

type = param->ismountpoint.in.type;

#6 lore raw
Quoted 1 line · tomas

Yes, thanks. Please use my full name, Tomas Bortoli.

Please also include:

Reported-by [email protected]

from the original bug report. This this help to keep automatic testing process running.

Thanks

Quoted 143 lines · Ian Kent
Quoted 136 lines · tomas
Quoted 6 lines

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing, with both our Signed-off-by added.

Quoted 124 lines

Tomas

Quoted 119 lines · Ian Kent
Quoted 41 lines · Ian Kent
Quoted 7 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f97208ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 21 lines

The autofs subsystem does not check that the "path" parameter is present within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a path is always provided (though a path is not always present, as per how the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/auto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 16 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 9 lines
fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200 renamed from fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621 +0200around +0 -0
     err = 0;You should just put the "_IOC_NR" directive around "cmd" in

the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
                            cmd);
                    goto out;
            }
+   } else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+           err = -EINVAL;
+           goto out;
    }
 
    err = 0;
fs/autofs/dev-ioctl.c +2 -8
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
    dev_t devid;
    int err = -ENOENT;
 
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    devid = sbi->sb->s_dev;
 
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file
 *fp,
    unsigned int devid, magic;
    int err = -ENOENT;
 
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    name = param->path;
    type = param->ismountpoint.in.type;

--
You received this message because you are subscribed to the Google Groups "syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.

maintainer
#7 lore raw
Quoted 7 lines · Dmitry Vyukov
Quoted 1 line · tomas

Yes, thanks. Please use my full name, Tomas Bortoli.

Please also include:

Reported-by [email protected]

Done.

Quoted 3 lines

from the original bug report. This this help to keep automatic testing process running.

Umm ... should I include this email address on the actual cc when submitting the patch?

Quoted 163 lines

Thanks

Quoted 159 lines · Ian Kent
Quoted 142 lines · tomas
Quoted 6 lines

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing, with both our Signed-off-by added.

Quoted 129 lines

Tomas

Quoted 125 lines · Ian Kent
Quoted 51 lines · Ian Kent
Quoted 8 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f9720
8ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 28 lines

The autofs subsystem does not check that the "path" parameter is present
within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a
path is always provided (though a path is not always present, as per how
the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/a
uto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of
the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that
a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

--- b/fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621
+0200around
+++ a/fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200

fs/autofs/dev-ioctl.c static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 17 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 10 lines
(quoted diff) +0 -0
     err = 0;You should just put the "_IOC_NR" directive around

"cmd" in
the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
                            cmd);
                    goto out;
            }
+   } else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+           err = -EINVAL;
+           goto out;
    }
 
    err = 0;
fs/autofs/dev-ioctl.c +1 -4
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file
 *fp,
    dev_t devid;
    int err = -ENOENT;
 
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    devid = sbi->sb->s_dev;
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct

file
*fp,

    unsigned int devid, magic;
    int err = -ENOENT;
Quoted 8 lines
Quoted 8 lines

fs/autofs/dev-ioctl.c static int autofs_dev_ioctl_ismountpoint(struct file *fp, · line 529

Quoted 8 lines
fs/autofs/dev-ioctl.c +1 -4
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    name = param->path;
    type = param->ismountpoint.in.type;

--
You received this message because you are subscribed to the Google Groups "syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.

#8 lore raw
Quoted 9 lines · Ian Kent
Quoted 7 lines · Dmitry Vyukov
Quoted 1 line · tomas

Yes, thanks. Please use my full name, Tomas Bortoli.

Please also include:

Reported-by [email protected]
Quoted 8 lines

Done.

Quoted 3 lines

from the original bug report. This this help to keep automatic testing process running.

Umm ... should I include this email address on the actual cc when submitting the patch?

It does not matter. It's a real email address too, so CCing it is fine. So if git does it automatically (does it?) then just go with it. If you are doing it manually, then it's not necessary to add it to CC.

syzbot will scrape git log later to discover the tag and mark the bug fixed.

Thanks

Quoted 163 lines
Quoted 149 lines
Quoted 149 lines · Ian Kent
Quoted 148 lines · tomas
Quoted 6 lines

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing, with both our Signed-off-by added.

Quoted 135 lines

Tomas

Quoted 131 lines · Ian Kent
Quoted 51 lines · Ian Kent
Quoted 8 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f9720
8ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 28 lines

The autofs subsystem does not check that the "path" parameter is present
within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a
path is always provided (though a path is not always present, as per how
the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/linux/a
uto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of
the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that
a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

--- b/fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621
+0200around
+++ a/fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133 +0200

fs/autofs/dev-ioctl.c static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
fs/autofs/dev-ioctl.c +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 17 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 10 lines
(quoted diff) +0 -0
     err = 0;You should just put the "_IOC_NR" directive around

"cmd" in
the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
                            cmd);
                    goto out;
            }
+   } else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+           err = -EINVAL;
+           goto out;
    }
 
    err = 0;
fs/autofs/dev-ioctl.c +1 -4
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct file
 *fp,
    dev_t devid;
    int err = -ENOENT;
 
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    devid = sbi->sb->s_dev;
@@ -521,10 +523,7 @@ static int autofs_dev_ioctl_ismountpoint(struct

file
*fp,

    unsigned int devid, magic;
    int err = -ENOENT;
fs/autofs/dev-ioctl.c +1 -4
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
    name = param->path;
    type = param->ismountpoint.in.type;

--
You received this message because you are subscribed to the Google Groups "syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.

maintainer
#9 lore raw
Quoted 16 lines · Dmitry Vyukov
Quoted 15 lines · Ian Kent
Quoted 7 lines · Dmitry Vyukov
Quoted 1 line · tomas

Yes, thanks. Please use my full name, Tomas Bortoli.

Please also include:

Reported-by [email protected]

Done.

Quoted 3 lines

from the original bug report. This this help to keep automatic testing process running.

Quoted 6 lines
Quoted 2 lines

Umm ... should I include this email address on the actual cc when submitting the patch?

It does not matter. It's a real email address too, so CCing it is fine. So if git does it automatically (does it?) then just go with it. If you are doing it manually, then it's not necessary to add it to CC.

Right, I thought a mail would probably be sent at merge time (or perhaps when Andrew adds it to mmotm) so I didn't think it sensible to cc when submitting.

Personally I use StGIT to manage patches and that will only send to addresses specified when patches are emailed.

Quoted 186 lines

syzbot will scrape git log later to discover the tag and mark the bug fixed.

Thanks

Quoted 181 lines
Quoted 163 lines
Quoted 163 lines · Ian Kent
Quoted 162 lines · tomas
Quoted 8 lines

Hi Ian,

you are welcome!

yes your patch is much better. You should just put the "_IOC_NR" macro
around "cmd" in the lines added to "validate_dev_ioctl" to make it work.

LOL, yes, that was a dumb mistake.

I'll send it to Andrew Morton, after some fairly simple sanity testing,
with both our Signed-off-by added.

Quoted 146 lines

Tomas

Quoted 142 lines · Ian Kent
Quoted 57 lines · Ian Kent
Quoted 10 lines · tomas

Hi,

I've looked into this issue found by Syzbot and I made a patch:

https://syzkaller.appspot.com/bug?id=d03abd8b42847f7f69b1d1d7f
9720
8ae425
b116
3

Umm ... oops!

Thanks for looking into this Tomas.

Quoted 32 lines

The autofs subsystem does not check that the "path" parameter is
present
within the "param" struct passed by the userspace in case the AUTOFS_DEV_IOCTL_OPENMOUNT_CMD command is passed. Indeed, it assumes a
path is always provided (though a path is not always present, as
per how
the struct is defined:
https://github.com/torvalds/linux/blob/master/include/uapi/lin
ux/a
uto_de
v-io
ct
l.h#L89).
Skipping the check provokes an oob read in "strlen", called by "getname_kernel", in turn called by the autofs to assess the length of
the non-existing path.

To solve it, modify the "validate_dev_ioctl" function to check also that
a path has been provided if the command is AUTOFS_DEV_IOCTL_OPENMOUNT_CMD.

--- b/fs/autofs/dev-ioctl.c 2018-07-01 23:10:16.059728621
+0200around
+++ a/fs/autofs/dev-ioctl.c 2018-07-01 23:10:24.311792133
+0200

Review static int validate_dev_ioctl(int cmd, s · line 141

Quoted 6 lines
(quoted diff) +3 -0
@@ -136,6 +136,9 @@ static int validate_dev_ioctl(int cmd, s
             goto out;
         }
     }
+    /* AUTOFS_DEV_IOCTL_OPENMOUNT_CMD without path */
+    else if(_IOC_NR(cmd) == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD)
+        return -EINVAL;

My preference is to put the comment inside the else but ...

Quoted 19 lines

There's another question, should the check be done in autofs_dev_ioctl_openmount() in the same way it's checked in other
ioctls that need a path, such as in autofs_dev_ioctl_requester() and autofs_dev_ioctl_ismountpoint()?

For consistency I'd say it should.

Quoted 11 lines
(quoted diff) +0 -0
     err = 0;You should just put the "_IOC_NR" directive

around
"cmd" in
the lines added to "validate_dev_ioctl" to make it work.
out:

Tested and solves the issue on Linus' main git tree.

Or perhaps this (not even compile tested) patch would be better?

autofs - fix slab out of bounds read in getname_kernel()

From: Ian Kent <[email protected]>

The autofs subsystem does not check that the "path" parameter is present for all cases where it is required when it is passed in via the "param" struct.

In particular it isn't checked for the
AUTOFS_DEV_IOCTL_OPENMOUNT_CMD
ioctl command.

To solve it, modify validate_dev_ioctl() function to check that a path has been provided for ioctl commands that require it.

---
 fs/autofs/dev-ioctl.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)
fs/autofs/dev-ioctl.c +5 -0
index ea4ca1445ab7..61c63715c3fb 100644
@@ -135,6 +135,11 @@ static int validate_dev_ioctl(int cmd, struct
 autofs_dev_ioctl *param)
                            cmd);
                    goto out;
            }
+   } else if (cmd == AUTOFS_DEV_IOCTL_OPENMOUNT_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_REQUESTER_CMD ||
+              cmd == AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD) {
+           err = -EINVAL;
+           goto out;
    }
 
    err = 0;
fs/autofs/dev-ioctl.c +0 -0
@@ -433,10 +438,7 @@ static int autofs_dev_ioctl_requester(struct

file
*fp,

    dev_t devid;
    int err = -ENOENT;
fs/autofs/dev-ioctl.c +1 -4
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
 
    devid = sbi->sb->s_dev;
@@ -521,10 +523,7 @@ static int

autofs_dev_ioctl_ismountpoint(struct
file
*fp,

    unsigned int devid, magic;
    int err = -ENOENT;
fs/autofs/dev-ioctl.c +1 -4
-   if (param->size <= AUTOFS_DEV_IOCTL_SIZE) {
-           err = -EINVAL;
-           goto out;
-   }
+   /* param->path has already been checked */
Quoted 11 lines
Quoted 11 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
    name = param->path;
    type = param->ismountpoint.in.type;

--
You received this message because you are subscribed to the Google Groups
"syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to [email protected].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "syzkaller" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.