fibmap: Reject negative addresses
Subsystems (2)
PATCH 1/1 fibmap: Reject negative addresses
FIBMAP receives an integer from userspace which is then implicitly converted into sector_t on ->bmap(). No check is made to ensure userspace didn't send a negative address, which can confuse the ->bmap interface, and return fuzzy addresses.
---
I am not 100% sure if is there any reason for ->bmap interface to accept negative values, from the top of my mind, I really can't see a reason, if there is a reason why we don't check for negative addresses, I'd be happy to know, otherwise we should reject it.
Cheers
fs/ioctl.c | 5 +++++ 1 file changed, 5 insertions(+)
fs/ioctl.c +5 -0
@@ -64,6 +64,11 @@ static int ioctl_fibmap(struct file *filp, int __user *p) res =get_user (block, p);if (res)return res; + +/* No reason to query a negative block addr */ +if (block <0 ) +return -EINVAL ; + res = mapping->a_ops->bmap (mapping, block);return put_user (res, p); }
Signature · 1 line
2.20.1
Quoted 4 lines · Carlos Maiolino
FIBMAP receives an integer from userspace which is then implicitly converted into
sector_ton ->bmap(). No check is made to ensure userspace didn't send a negative address, which can confuse the ->bmap interface, and return fuzzy addresses.
Nitpick on summary & comments,
I think this should be "negative logical block number?" because "address" often means something completely different. ;)
what is a "fuzzy address?" What actually happens if a negative number gets through?
Quoted 19 lines
Signed-off-by Carlos Maiolino <[email protected]>---
I am not 100% sure if is there any reason for ->bmap interface to accept negative values, from the top of my mind, I really can't see a reason, if there is a reason why we don't check for negative addresses, I'd be happy to know, otherwise we should reject it.
Cheers
fs/ioctl.c | 5 +++++ 1 file changed, 5 insertions(+)fs/ioctl.c +0
-0
fs/ioctl.c static int ioctl_fibmap(struct file *filp, int __user *p) · line 68
Quoted 5 lines
fs/ioctl.c +2
-0@@ -64,6 +64,11 @@ static int ioctl_fibmap(struct file *filp, int __user *p) res =get_user (block, p);if (res)return res; + +/* No reason to query a negative block addr */
/* Invalid to query a negative block number */
Actually, the next 2 lines of code are not hard to understand, so probably doesn't really even need a comment.
As for the change, it seems fine, I guess we don't document what the user interface is supposed to be anywhere, sooo... go with the current signed int implementation, and disallow anything that doesn't fit, makes sense to me.
-Eric
Quoted 7 lines
fs/ioctl.c +3
-0+if (block <0 ) +return -EINVAL ; + res = mapping->a_ops->bmap (mapping, block);return put_user (res, p); }
Quoted 11 lines · Eric Sandeen
Quoted 4 lines · Carlos Maiolino
FIBMAP receives an integer from userspace which is then implicitly converted into
sector_ton ->bmap(). No check is made to ensure userspace didn't send a negative address, which can confuse the ->bmap interface, and return fuzzy addresses.Nitpick on summary & comments,
I think this should be "negative logical block number?" because "address" often means something completely different. ;)
Thanks, yeah, I meant block address, but I omitted the 'block'
Quoted 2 lines
what is a "fuzzy address?" What actually happens if a negative number gets through?
The type internally is a sector_t, which is a u64, so it will underflow the internal type, making the request look for block (2<<63)-1 (IIRC), which is most likely what the userspace doesn't want.
Also, this affects iomap bmap interface, which has a WARN() in iomap_bmap_actor() triggered by any block address bigger than INT_MAX, which, makes sense giving the userspace limitation.
I should have written a better description for the patch =/ I'll send a V2 with a better description, if people are ok with the code change, so, I'll just wait for a few more reviews before writing a V2.
Quoted 32 lines
Quoted 20 lines
Signed-off-by Carlos Maiolino <[email protected]>---
I am not 100% sure if is there any reason for ->bmap interface to accept negative values, from the top of my mind, I really can't see a reason, if there is a reason why we don't check for negative addresses, I'd be happy to know, otherwise we should reject it.
Cheers
fs/ioctl.c | 5 +++++ 1 file changed, 5 insertions(+)fs/ioctl.c +0
-0fs/ioctl.c static int ioctl_fibmap(struct file *filp, int __user *p) · line 68
Quoted 5 lines
fs/ioctl.c +2
-0@@ -64,6 +64,11 @@ static int ioctl_fibmap(struct file *filp, int __user *p) res =get_user (block, p);if (res)return res; + +/* No reason to query a negative block addr */ /* Invalid to query a negative block number */
Actually, the next 2 lines of code are not hard to understand, so probably doesn't really even need a comment.
Quoted 5 lines
As for the change, it seems fine, I guess we don't document what the user interface is supposed to be anywhere, sooo... go with the current signed int implementation, and disallow anything that doesn't fit, makes sense to me.
-Eric
I much appreciate your review
Quoted 8 lines
Quoted 8 lines
fs/ioctl.c +3
-0+if (block <0 ) +return -EINVAL ; + res = mapping->a_ops->bmap (mapping, block);return put_user (res, p); }
Signature · 1 line
Carlos