Skip to contentSkip to conversation

coredump - as root not only if euid switched

PATCH Peter Waechtler 4 participants 5 messages Apr 22, 2004

Subsystems (3)
lore
Share & tools
lore
#0 lore raw
Quoted 8 lines · Andrew Morton
Quoted 6 lines · Peter Waechtler

But I agree that sys_unlink should be the fast call and dumping core is the exception :)

would fastcall do_unlink() help? I guess the arg is then passed in a register

Quoted 8 lines

I've never been able to measure any size or space benefit for fastcall, and we do it via compiler options kernel-wide nowadays.

The above will work fine. You can probably just open-code it at the place where you're unlinking the file.

(why are you trying to unlink the old file anyway?)

For security measure :O
I tried on solaris: touch the core file as user, open it and wait, dump core as root -> nope, couldn't read the damn core - it was unlinked and created! So do I want.
I will sent the new patch from home.

#1 lore raw
Quoted 7 lines · Peter Waechtler
Quoted 3 lines

(why are you trying to unlink the old file anyway?)

For security measure :O
I tried on solaris: touch the core file as user, open it and wait, dump core as root -> nope, couldn't read the damn core - it was unlinked and created!

hm, OK. There's a window in which someone can come in and recreate the file, but the open is using O_EXCL|O_CREATE so that seems safe enough.

#2 lore raw

Am Do, 2004-04-22 um 11.56 schrieb Andrew Morton:

Quoted 11 lines
Quoted 7 lines · Peter Waechtler
Quoted 3 lines

(why are you trying to unlink the old file anyway?)

For security measure :O
I tried on solaris: touch the core file as user, open it and wait, dump core as root -> nope, couldn't read the damn core - it was unlinked and created!

hm, OK. There's a window in which someone can come in and recreate the file, but the open is using O_EXCL|O_CREATE so that seems safe enough.

So here is the updated patch with an open coded call to sys_unlink

fs/exec.c +21 -3
diff -Nur -X dontdiff linux-2.6.5/fs/exec.c linux-2.6/fs/exec.c
@@ -1378,6 +1378,10 @@
 		goto fail;
 	}
 	mm->dumpable = 0;
+	if (mm->dump_as_root){
+		current->fsuid = 0;
+		current->fsgid = 0;
+	}
 	init_completion(&mm->core_done);
 	current->signal->group_exit = 1;
 	current->signal->group_exit_code = exit_code;
@@ -1387,9 +1391,23 @@
 		goto fail_unlock;
 
  	format_corename(corename, core_pattern, signr);
-	file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
-	if (IS_ERR(file))
-		goto fail_unlock;
+	file = filp_open(corename, O_CREAT | O_EXCL | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
+	if (IS_ERR(file)){
+        long error;
+        mm_segment_t old_fs = get_fs();
+
+        set_fs(KERNEL_DS);
+        error = sys_unlink(corename);
+        set_fs(old_fs);
+        if (error)
+			goto fail_unlock;
+		file = filp_open(corename, O_CREAT | O_EXCL | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
+		if (IS_ERR(file)){
+			printk(KERN_WARNING "pid %d doesn't dump core to %s because of creation race\n",
+				current->pid, corename);
+			goto fail_unlock;
+		}
+	}
 	inode = file->f_dentry->d_inode;
 	if (inode->i_nlink > 1)
 		goto close_fail;	/* multiple links - don't dump */
include/linux/sched.h +1 -0
diff -Nur -X dontdiff linux-2.6.5/include/linux/sched.h linux-2.6/include/linux/sched.h
@@ -211,6 +211,7 @@
 	unsigned long saved_auxv[40]; /* for /proc/PID/auxv */
 
 	unsigned dumpable:1;
+	unsigned dump_as_root:1;
 #ifdef CONFIG_HUGETLB_PAGE
 	int used_hugetlb;
 #endif
kernel/sys.c +10 -10
diff -Nur -X dontdiff linux-2.6.5/kernel/sys.c linux-2.6/kernel/sys.c
@@ -573,7 +573,7 @@
 	}
 	if (new_egid != old_egid)
 	{
-		current->mm->dumpable = 0;
+		current->mm->dump_as_root = 1;
 		wmb();
 	}
 	if (rgid != (gid_t) -1 ||
@@ -603,7 +603,7 @@
 	{
 		if(old_egid != gid)
 		{
-			current->mm->dumpable=0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->gid = current->egid = current->sgid = current->fsgid = gid;
@@ -612,7 +612,7 @@
 	{
 		if(old_egid != gid)
 		{
-			current->mm->dumpable=0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->egid = current->fsgid = gid;
@@ -641,7 +641,7 @@
 
 	if(dumpclear)
 	{
-		current->mm->dumpable = 0;
+		current->mm->dump_as_root = 1;
 		wmb();
 	}
 	current->uid = new_ruid;
@@ -698,7 +698,7 @@
 
 	if (new_euid != old_euid)
 	{
-		current->mm->dumpable=0;
+		current->mm->dump_as_root = 1;
 		wmb();
 	}
 	current->fsuid = current->euid = new_euid;
@@ -746,7 +746,7 @@
 
 	if (old_euid != uid)
 	{
-		current->mm->dumpable = 0;
+		current->mm->dump_as_root = 1;
 		wmb();
 	}
 	current->fsuid = current->euid = uid;
@@ -789,7 +789,7 @@
 	if (euid != (uid_t) -1) {
 		if (euid != current->euid)
 		{
-			current->mm->dumpable = 0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->euid = euid;
@@ -837,7 +837,7 @@
 	if (egid != (gid_t) -1) {
 		if (egid != current->egid)
 		{
-			current->mm->dumpable = 0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->egid = egid;
@@ -882,7 +882,7 @@
 	{
 		if (uid != old_fsuid)
 		{
-			current->mm->dumpable = 0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->fsuid = uid;
@@ -910,7 +910,7 @@
 	{
 		if (gid != old_fsgid)
 		{
-			current->mm->dumpable = 0;
+			current->mm->dump_as_root = 1;
 			wmb();
 		}
 		current->fsgid = gid;
#3 lore raw
Quoted 6 lines · * Peter Wächtler ([email protected]) wrote:

Am Do, 2004-04-22 um 11.56 schrieb Andrew Morton:

Quoted 5 lines
Quoted 4 lines · Peter Waechtler
Quoted 3 lines

(why are you trying to unlink the old file anyway?)

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

For security measure :O
I tried on solaris: touch the core file as user, open it and wait, dump core as root -> nope, couldn't read the damn core - it was unlinked and created!

hm, OK. There's a window in which someone can come in and recreate the file, but the open is using O_EXCL|O_CREATE so that seems safe enough.

So here is the updated patch with an open coded call to sys_unlink

This patch breaks various ptrace() checks.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
#4 lore raw
In reply to Peter Wächtler
Quoted 5 lines · Peter Wächtler
Quoted 3 lines

hm, OK. There's a window in which someone can come in and recreate the file, but the open is using O_EXCL|O_CREATE so that seems safe enough.

So here is the updated patch with an open coded call to sys_unlink

Aughr.

Wouldn't it be much nicer to just refuse to overwrite files owned by anybody else?

In other words, I'd much rather see a patch that is a much simpler one, which just says: if we opened an existing file, we won't touch it if we weren't the owners of it.

That should be safe for root _and_ it should be safe for people who already had a file descriptor open previously (hey, if the previous root-owned core-file was world readable, then what else is new?)

Tell me why this isn't simpler?

Linus

---

fs/exec.c +2 -0
@@ -1378,6 +1378,8 @@
 	inode = file->f_dentry->d_inode;
 	if (inode->i_nlink > 1)
 		goto close_fail;	/* multiple links - don't dump */
+	if (inode->i_uid != current->euid || inode->i_gid != current->egid)
+		goto close_fail;
 	if (d_unhashed(file->f_dentry))
 		goto close_fail;