Skip to contentSkip to conversation

fs/proc/: fix input sanity check.

PATCH In review Vincent Li 3 participants 4 messages Aug 11, 2009 – Aug 12, 2009

Subsystems (2)
lore
Share & tools
lore

Vincent Li

author
#0 lore raw

PATCH 1/1 fs/proc/: fix input sanity check.

fix fs/proc/task_mmu.c clear_refs_write(), fs/proc/base.c proc_fault_inject_write()
and proc_fault_inject_operations() input sanity check by following the disccusion of
http://marc.info/?l=linux-mm&m=124938168905463&w=2.

---
 fs/proc/base.c     |   20 ++++++++------------
 fs/proc/task_mmu.c |   11 +++++------
 2 files changed, 13 insertions(+), 18 deletions(-)
fs/proc/base.c +8 -12
index 04d29a0..44054d2 100644
@@ -1190,17 +1190,16 @@ static ssize_t proc_fault_inject_write(struct file * file,
 		count = sizeof(buffer) - 1;
 	if (copy_from_user(buffer, buf, count))
 		return -EFAULT;
-	make_it_fail = simple_strtol(buffer, &end, 0);
-	if (*end == '\n')
-		end++;
+	make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
+	if (*end)
+		return -EINVAL;
 	task = get_proc_task(file->f_dentry->d_inode);
 	if (!task)
 		return -ESRCH;
 	task->make_it_fail = make_it_fail;
 	put_task_struct(task);
-	if (end - buffer == 0)
-		return -EIO;
-	return end - buffer;
+
+	return count;
 }
 
 static const struct file_operations proc_fault_inject_operations = {
@@ -2253,18 +2252,15 @@ static ssize_t proc_coredump_filter_write(struct file *file,
 		goto out_no_task;
 
 	ret = -EINVAL;
-	val = (unsigned int)simple_strtoul(buffer, &end, 0);
-	if (*end == '\n')
-		end++;
-	if (end - buffer == 0)
-		goto out_no_task;
+	val = (unsigned int)simple_strtoul(strstrip(buffer), &end, 0);
+	if (*end)
+		return ret;
 
 	ret = -ESRCH;
 	task = get_proc_task(file->f_dentry->d_inode);
 	if (!task)
 		goto out_no_task;
 
-	ret = end - buffer;
 	mm = get_task_mm(task);
 	if (!mm)
 		goto out_no_mm;
fs/proc/task_mmu.c +5 -6
index 6f61b7c..957b266 100644
@@ -477,10 +477,10 @@ static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 		count = sizeof(buffer) - 1;
 	if (copy_from_user(buffer, buf, count))
 		return -EFAULT;
-	if (!simple_strtol(buffer, &end, 0))
+	if (!simple_strtol(strstrip(buffer), &end, 0))
+		return -EINVAL;
+	if (*end)
 		return -EINVAL;
-	if (*end == '\n')
-		end++;
 	task = get_proc_task(file->f_path.dentry->d_inode);
 	if (!task)
 		return -ESRCH;
@@ -502,9 +502,8 @@ static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 		mmput(mm);
 	}
 	put_task_struct(task);
-	if (end - buffer == 0)
-		return -EIO;
-	return end - buffer;
+
+	return count;
 }
 
 const struct file_operations proc_clear_refs_operations = {
Signature · 1 line
1.6.0.4
maintainer
#1 lore raw
Quoted 3 lines · On Tue, 11 Aug 2009 14:34:26 -0700 Vincent Li <[email protected]> wrote:

fix fs/proc/task_mmu.c clear_refs_write(), fs/proc/base.c proc_fault_inject_write()
and proc_fault_inject_operations() input sanity check by following the disccusion of
http://marc.info/?l=linux-mm&m=124938168905463&w=2.

Would prefer that the changelog not consist of a link to another patch which is already quite poorly changelogged, please!

What, precisely was wrong with the old code and how does the patch change its behaviour.

This matters. The patch potentially changes the kernel<->userspace interface and we must fully understand the nature of that change so as to minimise the risk of breaking existing userspace application code.

Thanks.

author
#2 lore raw
Quoted 9 lines · Andrew Morton
Quoted 3 lines · On Tue, 11 Aug 2009 14:34:26 -0700 Vincent Li <[email protected]> wrote:

fix fs/proc/task_mmu.c clear_refs_write(), fs/proc/base.c proc_fault_inject_write()
and proc_fault_inject_operations() input sanity check by following the disccusion of
http://marc.info/?l=linux-mm&m=124938168905463&w=2.

Would prefer that the changelog not consist of a link to another patch which is already quite poorly changelogged, please!

It make sense to me only after you point it out to me, That is process of learning :)

Quoted 4 lines

What, precisely was wrong with the old code and how does the patch change its behaviour.

It is after I read your comment about the strange handling of userspace string input http://marc.info/?l=linux-mm&m=124890923527186&w=2 and Kosaki's "fix oom_adjust_write() input sanity check" patch to match your comment http://marc.info/?l=linux-mm&m=124938168905463&w=2.

I re-read the clear_refs_write() in fs/proc/task_mmu.c, If I understand it correctly, The old code takes non-zero string
from userspace and convert it to unsigned long interger value. for example,It accepts:

echo 1 > /proc/pid/clear_refs
echo 12foo > /proc/pid/clear_refs

The new code add the stripping of leading and trailing whitespace, remove the obfuscated check for zero-length input at the end of the function as you commented. Will this new code break the existing application behaviour? David Rientjes is the author of clear_refs, I cced him to see if he has any comment.

I have not read the proc_fault_inject_write() and proc_fault_inject_operations() yet, I will drop the changes to those two functions

Quoted 5 lines

This matters. The patch potentially changes the kernel<->userspace interface and we must fully understand the nature of that change so as to minimise the risk of breaking existing userspace application code.

Point taken, I am in a rush to post the patch without fully understanding what the functions do and the impact of patch changes.

Quoted 1 line

Thanks.

You are welcome.

Vincent Li
Biomedical Research Center
University of British Columbia

#3 lore raw
Quoted 3 lines · Vincent Li

I re-read the clear_refs_write() in fs/proc/task_mmu.c, If I understand it correctly, The old code takes non-zero string
from userspace and convert it to unsigned long interger value. for

strict_strtol() converts to long.

Quoted 5 lines

example,It accepts:

echo 1 > /proc/pid/clear_refs
echo 12foo > /proc/pid/clear_refs

http://userweb.kernel.org/~akpm/mmotm/broken-out/pagemap-clear_refs-modify-to-specify-anon-or-mapped-vma-clearing.patch
changes that, so you'll need to rebase on mmotm.