Skip to contentSkip to conversation

CRED: Fixup credentials build breakage

PATCH In review Alex Chiang 2 participants 3 messages Aug 13, 2008 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 CRED: Fixup credentials build breakage

A recent patch titled:

CRED: Separate task security context from task_struct

removed task security context from task_struct, but did not update all locations to use the new struct cred that was introduced.

The change to task_struct broke perfmon and ia32 syscalls on ia64. This patch fixes the build.
---
All things considered, I'd prefer to see this patch folded into 7931c65268777ed10cab22486de149d742a1f269 so we can keep bisectability. Would that be possible, given that these changes are "only" in linux-next and haven't hit Linus's tree yet?

My hope for folding in this patch into the bigger patch is the reason why I didn't refer to the commit by SHA1 in the changelog...

Also, I'm not exactly sure why wrappers were provided for task_gid, but then later removed. Additionally, I wasn't sure why no wrapper was provided for task_suid and friends. But I admit that I didn't read the patch series in depth; only enough to fix the build.

 arch/ia64/ia32/sys_ia32.c  |    8 ++++----
 arch/ia64/kernel/perfmon.c |   24 ++++++++++++------------
 2 files changed, 16 insertions(+), 16 deletions(-)
arch/ia64/ia32/sys_ia32.c +4 -4
index 465116a..950b63a 100644
@@ -2089,20 +2089,20 @@ sys32_getgroups16 (int gidsetsize, short __user *grouplist)
 	if (gidsetsize < 0)
 		return -EINVAL;
 
-	get_group_info(current->group_info);
-	i = current->group_info->ngroups;
+	get_group_info(current->cred->group_info);
+	i = current->cred->group_info->ngroups;
 	if (gidsetsize) {
 		if (i > gidsetsize) {
 			i = -EINVAL;
 			goto out;
 		}
-		if (groups16_to_user(grouplist, current->group_info)) {
+		if (groups16_to_user(grouplist, current->cred->group_info)) {
 			i = -EFAULT;
 			goto out;
 		}
 	}
 out:
-	put_group_info(current->group_info);
+	put_group_info(current->cred->group_info);
 	return i;
 }
 
arch/ia64/kernel/perfmon.c +12 -12
index ffe6de0..7df49bc 100644
@@ -2410,18 +2410,18 @@ pfm_bad_permissions(struct task_struct *task)
 	DPRINT(("cur: uid=%d gid=%d task: euid=%d suid=%d uid=%d egid=%d sgid=%d\n",
 		uid,
 		gid,
-		task->euid,
-		task->suid,
-		task->uid,
-		task->egid,
-		task->sgid));
-
-	return (uid != task->euid)
-	    || (uid != task->suid)
-	    || (uid != task->uid)
-	    || (gid != task->egid)
-	    || (gid != task->sgid)
-	    || (gid != task->gid)) && !capable(CAP_SYS_PTRACE);
+		task_euid(task),
+		task->cred->suid,
+		task_uid(task),
+		task->cred->egid,
+		task->cred->sgid));
+
+	return ((uid != task_euid(task))
+	    || (uid != task->cred->suid)
+	    || (uid != task_uid(task))
+	    || (gid != task->cred->egid)
+	    || (gid != task->cred->sgid)
+	    || (gid != task->cred->gid)) && !capable(CAP_SYS_PTRACE);
 }
 
 static int
Signature · 1 line
1.5.3.1.gbed62
#1 lore raw
Quoted 10 lines · Alex Chiang

A recent patch titled:

CRED: Separate task security context from task_struct

removed task security context from task_struct, but did not update all locations to use the new struct cred that was introduced.

The change to task_struct broke perfmon and ia32 syscalls on ia64. This patch fixes the build.

I've submitted a patch to fix this.

Quoted 4 lines

All things considered, I'd prefer to see this patch folded into 7931c65268777ed10cab22486de149d742a1f269 so we can keep bisectability. Would that be possible, given that these changes are "only" in linux-next and haven't hit Linus's tree yet?

Probably. I'll have to talk to Stephen about how to do this. I can't maintain my patches on top of linux-next now:-/

Quoted 5 lines

Also, I'm not exactly sure why wrappers were provided for task_gid, but then later removed. Additionally, I wasn't sure why no wrapper was provided for task_suid and friends. But I admit that I didn't read the patch series in depth; only enough to fix the build.

Generally, accesses to task->gid are accompanied by accesses to task->uid and/or task->groups, in which case using multiple wrappers in series introduces excess locking and superfluous memory barriers. To access task->gid, one must take the RCU read lock and then use rcu_dereference().

However, I can't use the cred struct until the patch in which it is introduced. The wrappers, however, allow me to grep for things I've missed, and thus make it easier to do things in stages.

Your patch also has a couple of issues:

Review

Quoted 1 line
(quoted diff) +1 -0
+	get_group_info(current->cred->group_info);

The call to get_group_info() is not necessary. Only current may change its own groups.

Review

Quoted 4 lines
(quoted diff) +4 -0
+	    || (uid != task->cred->suid)
+	    || (gid != task->cred->egid)
+	    || (gid != task->cred->sgid)
+	    || (gid != task->cred->gid))

This is incorrect. You must hold the RCU read lock whilst you make this access, and you must pass task->cred through rcu_dereference(), or better still, use __task_cred(task) to get at it.

My patch is below. Note that there is a problem with it, presumably something to do with GIT's merging (the ad1848 files being deleted).

David
---
Fix the IA64 arch's use of COW credentials.

---

 arch/ia64/ia32/sys_ia32.c     |    7 +++----
 arch/ia64/kernel/perfmon.c    |   32 ++++++++++++++++++++------------
 include/sound/ad1848.h        |    0 
 sound/isa/ad1848/ad1848_lib.c |    0 
 4 files changed, 23 insertions(+), 16 deletions(-)
 delete mode 100644 include/sound/ad1848.h
 delete mode 100644 sound/isa/ad1848/ad1848_lib.c
arch/ia64/ia32/sys_ia32.c +3 -4
index 465116a..7f0704f 100644
@@ -2084,25 +2084,24 @@ groups16_from_user(struct group_info *group_info, short __user *grouplist)
 asmlinkage long
 sys32_getgroups16 (int gidsetsize, short __user *grouplist)
 {
+	const struct cred *cred = current_cred();
 	int i;
 
 	if (gidsetsize < 0)
 		return -EINVAL;
 
-	get_group_info(current->group_info);
-	i = current->group_info->ngroups;
+	i = cred->group_info->ngroups;
 	if (gidsetsize) {
 		if (i > gidsetsize) {
 			i = -EINVAL;
 			goto out;
 		}
-		if (groups16_to_user(grouplist, current->group_info)) {
+		if (groups16_to_user(grouplist, cred->group_info)) {
 			i = -EFAULT;
 			goto out;
 		}
 	}
 out:
-	put_group_info(current->group_info);
 	return i;
 }
 
arch/ia64/kernel/perfmon.c +20 -12
index ffe6de0..a1aead7 100644
@@ -2403,25 +2403,33 @@ error_kmem:
 static int
 pfm_bad_permissions(struct task_struct *task)
 {
+	const struct cred *tcred;
 	uid_t uid = current_uid();
 	gid_t gid = current_gid();
+	int ret;
+
+	rcu_read_lock();
+	tcred = __task_cred(task);
 
 	/* inspired by ptrace_attach() */
 	DPRINT(("cur: uid=%d gid=%d task: euid=%d suid=%d uid=%d egid=%d sgid=%d\n",
 		uid,
 		gid,
-		task->euid,
-		task->suid,
-		task->uid,
-		task->egid,
-		task->sgid));
-
-	return (uid != task->euid)
-	    || (uid != task->suid)
-	    || (uid != task->uid)
-	    || (gid != task->egid)
-	    || (gid != task->sgid)
-	    || (gid != task->gid)) && !capable(CAP_SYS_PTRACE);
+		tcred->euid,
+		tcred->suid,
+		tcred->uid,
+		tcred->egid,
+		tcred->sgid));
+
+	ret = ((uid != tcred->euid)
+	       || (uid != tcred->suid)
+	       || (uid != tcred->uid)
+	       || (gid != tcred->egid)
+	       || (gid != tcred->sgid)
+	       || (gid != tcred->gid)) && !capable(CAP_SYS_PTRACE);
+
+	rcu_read_unlock();
+	return ret;
 }
 
 static int
include/sound/ad1848.h deleted +0 -0
deleted file mode 100644
index e69de29..0000000
sound/isa/ad1848/ad1848_lib.c deleted +0 -0
deleted file mode 100644
index e69de29..0000000
#2 lore raw

* David Howells <[email protected]>:

Quoted 6 lines
Quoted 4 lines · Alex Chiang

A recent patch titled:

CRED: Separate task security context from task_struct

Quoted 8 lines
Quoted 6 lines

removed task security context from task_struct, but did not update all locations to use the new struct cred that was introduced.

The change to task_struct broke perfmon and ia32 syscalls on ia64. This patch fixes the build.

I've submitted a patch to fix this.

Thanks, clearly your patch is correct and mine was not. I was just trying to get my build going again. I can apply yours by hand to my own tree here to get going.

Quoted 8 lines
Quoted 5 lines

All things considered, I'd prefer to see this patch folded into 7931c65268777ed10cab22486de149d742a1f269 so we can keep bisectability. Would that be possible, given that these changes are "only" in linux-next and haven't hit Linus's tree yet?

Probably. I'll have to talk to Stephen about how to do this. I can't maintain my patches on top of linux-next now:-/

Ok, I still think this would be a good idea, so if it's possible, that would be great.

Thanks,

/ac