Skip to contentSkip to conversation

groups: integer underflow in groups_alloc()

PATCH In review Dan Carpenter 4 participants 8 messages Feb 23, 2015 – Feb 26, 2015

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 groups: integer underflow in groups_alloc()

This is called from rsc_parse() with a use controlled value. Say for example that "gidsetsize" is negative, then we could end up allocating less than sizeof(struct group_info) leading to memory corruption.

---
I copied the NGROUPS_MAX limit from the surrounding code, I'm not absolutely that it's the correct limit to use.

kernel/groups.c +3 -0
index 664411f..e9341b3 100644
@@ -18,6 +18,9 @@ struct group_info *groups_alloc(int gidsetsize)
 	int nblocks;
 	int i;
 
+	if ((unsigned)gidsetsize > NGROUPS_MAX)
+		return NULL;
+
 	nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
 	/* Make sure we always allocate at least one indirect block pointer */
 	nblocks = nblocks ? : 1;
#1 lore raw
Quoted 3 lines · Dan Carpenter

This is called from rsc_parse() with a use controlled value. Say for example that "gidsetsize" is negative, then we could end up allocating less than sizeof(struct group_info) leading to memory corruption.

Right now it is the responsibility of the caller of groups_alloc to make certain that gidsetsize is a valid value, and the callers of groups_alloc who know what they are doing already validate this value.

Either the pattern of caller validates the messages needs to continue, or groups_alloc needs to be changed and all of the callers need to be updated.

Changing groups_alloc for one particular caller is just going to cause maintenance problems.

Eric

Quoted 19 lines

---
I copied the NGROUPS_MAX limit from the surrounding code, I'm not absolutely that it's the correct limit to use.

kernel/groups.c +3 -0
index 664411f..e9341b3 100644
@@ -18,6 +18,9 @@ struct group_info *groups_alloc(int gidsetsize)
 	int nblocks;
 	int i;
 
+	if ((unsigned)gidsetsize > NGROUPS_MAX)
+		return NULL;
+
 	nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
 	/* Make sure we always allocate at least one indirect block pointer */
 	nblocks = nblocks ? : 1;
#2 lore raw
Quoted 10 lines · Eric W. Biederman
Quoted 3 lines · Dan Carpenter

This is called from rsc_parse() with a use controlled value. Say for example that "gidsetsize" is negative, then we could end up allocating less than sizeof(struct group_info) leading to memory corruption.

Right now it is the responsibility of the caller of groups_alloc to make certain that gidsetsize is a valid value, and the callers of groups_alloc who know what they are doing already validate this value.

Quoted 7 lines

Either the pattern of caller validates the messages needs to continue, or groups_alloc needs to be changed and all of the callers need to be updated.

Changing groups_alloc for one particular caller is just going to cause maintenance problems.

This only affects NFS so let's hear from them if this limit is correct and decide from there.

regards,
dan carpenter

#3 lore raw
Quoted 15 lines · Dan Carpenter
Quoted 14 lines · Eric W. Biederman
Quoted 3 lines · Dan Carpenter

This is called from rsc_parse() with a use controlled value. Say for example that "gidsetsize" is negative, then we could end up allocating less than sizeof(struct group_info) leading to memory corruption.

Right now it is the responsibility of the caller of groups_alloc to make certain that gidsetsize is a valid value, and the callers of groups_alloc who know what they are doing already validate this value.

Either the pattern of caller validates the messages needs to continue, or groups_alloc needs to be changed and all of the callers need to be updated.

Quoted 6 lines
Quoted 3 lines

Changing groups_alloc for one particular caller is just going to cause maintenance problems.

This only affects NFS so let's hear from them if this limit is correct and decide from there.

The bug may be nfs specific bug changing groups_alloc does not only affect nfs.

NGROUPS_MAX is the maxmimum number of groups the linux kernel supports so NGROUPS_MAX may be high but it is certainly not wrong.

Your patch takes the wrong approach, creates code that is an inconsistent mess and is thus wrong. As setgroups is code that is called every day I don't think only paying attention to NFS when talking how to change this is in any way appropriate, unless you propose an NFS specific fix (which you clearly did not).

Eric

#4 lore raw
In reply to Dan Carpenter
Quoted 15 lines · Dan Carpenter
Quoted 14 lines · Eric W. Biederman
Quoted 3 lines · Dan Carpenter

This is called from rsc_parse() with a use controlled value. Say for example that "gidsetsize" is negative, then we could end up allocating less than sizeof(struct group_info) leading to memory corruption.

Right now it is the responsibility of the caller of groups_alloc to make certain that gidsetsize is a valid value, and the callers of groups_alloc who know what they are doing already validate this value.

Either the pattern of caller validates the messages needs to continue, or groups_alloc needs to be changed and all of the callers need to be updated.

Quoted 6 lines
Quoted 3 lines

Changing groups_alloc for one particular caller is just going to cause maintenance problems.

This only affects NFS so let's hear from them if this limit is correct and decide from there.

I think that's probably the correct check, sure.

Putting it in rsc_parse sounds reasonable.

--b.

#5 lore raw

PATCH v2 1/1 sunrpc: integer underflow in rsc_parse()

If we call groups_alloc() with invalid values then it's might lead to memory corruption. For example, with a negative value then we might not allocate enough for sizeof(struct group_info).

---
v2: In v1, I changed groups_alloc(). The other places which call groups_alloc() check the value before calling. Eric wanted that, either have all the callers check, or all the callers rely on groups_alloc(). In the end, Bruce Fields said adding the check here was probably reasonable.

net/sunrpc/auth_gss/svcauth_gss.c +2 -0
index 224a82f..1095be9 100644
@@ -463,6 +463,8 @@ static int rsc_parse(struct cache_detail *cd,
 		/* number of additional gid's */
 		if (get_int(&mesg, &N))
 			goto out;
+		if (N < 0 || N > NGROUPS_MAX)
+			goto out;
 		status = -ENOMEM;
 		rsci.cred.cr_group_info = groups_alloc(N);
 		if (rsci.cred.cr_group_info = NULL)
#6 lore raw

Re: [patch v2] sunrpc: integer underflow in rsc_parse()

Quoted 26 lines · Dan Carpenter

If we call groups_alloc() with invalid values then it's might lead to memory corruption. For example, with a negative value then we might not allocate enough for sizeof(struct group_info).

---
v2: In v1, I changed groups_alloc(). The other places which call groups_alloc() check the value before calling. Eric wanted that, either have all the callers check, or all the callers rely on groups_alloc(). In the end, Bruce Fields said adding the check here was probably reasonable.

net/sunrpc/auth_gss/svcauth_gss.c +2 -1
index 224a82f..1095be9 100644
@@ -463,6 +463,8 @@ static int rsc_parse(struct cache_detail *cd,
 		/* number of additional gid's */
 		if (get_int(&mesg, &N))
 			goto out;
+		if (N < 0 || N > NGROUPS_MAX)
+			goto out;
 		status = -ENOMEM;
 		rsci.cred.cr_group_info = groups_alloc(N);
 		if (rsci.cred.cr_group_info = NULL)
--
Quoted 3 lines

To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

I touched this code relatively recently, and this check looks correct. Feel free to add Reviewed-by: Simo Sorce <[email protected]>

Simo.

Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Reviewed-by Simo Sorce
#7 lore raw

Re: [patch v2] sunrpc: integer underflow in rsc_parse()

Quoted 25 lines · Simo Sorce
Quoted 24 lines · Dan Carpenter

If we call groups_alloc() with invalid values then it's might lead to memory corruption. For example, with a negative value then we might not allocate enough for sizeof(struct group_info).

---
v2: In v1, I changed groups_alloc(). The other places which call groups_alloc() check the value before calling. Eric wanted that, either have all the callers check, or all the callers rely on groups_alloc(). In the end, Bruce Fields said adding the check here was probably reasonable.

net/sunrpc/auth_gss/svcauth_gss.c +2 -0
index 224a82f..1095be9 100644
@@ -463,6 +463,8 @@ static int rsc_parse(struct cache_detail *cd,
 		/* number of additional gid's */
 		if (get_int(&mesg, &N))
 			goto out;
+		if (N < 0 || N > NGROUPS_MAX)
+			goto out;
 		status = -ENOMEM;
 		rsci.cred.cr_group_info = groups_alloc(N);
Quoted 8 lines
Quoted 5 lines
net/sunrpc/auth_gss/svcauth_gss.c +0 -1
 		if (rsci.cred.cr_group_info = NULL)
--

To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

I touched this code relatively recently, and this check looks correct. Feel free to add Reviewed-by: Simo Sorce <[email protected]>

Thanks! I thought your below-the-line context was useful, so pulled a version of it into the commit.

--b.

commit 76cb4be993c0

Author: Dan Carpenter <[email protected]>
Date:   Tue Feb 24 18:34:01 2015 +0300

sunrpc: integer underflow in rsc_parse()

If we call groups_alloc() with invalid values then it's might lead to memory corruption. For example, with a negative value then we might not allocate enough for sizeof(struct group_info).

(We're doing this in the caller for consistency with other callers of groups_alloc(). The other alternative might be to move the check out of all the callers into groups_alloc().)

net/sunrpc/auth_gss/svcauth_gss.c +2 -0
index 224a82f24d3c..1095be9c80ab 100644
@@ -463,6 +463,8 @@ static int rsc_parse(struct cache_detail *cd,
 		/* number of additional gid's */
 		if (get_int(&mesg, &N))
 			goto out;
+		if (N < 0 || N > NGROUPS_MAX)
+			goto out;
 		status = -ENOMEM;
 		rsci.cred.cr_group_info = groups_alloc(N);
 		if (rsci.cred.cr_group_info = NULL)