Skip to contentSkip to conversation

proc: fix comm_write return value when truncated or error

PATCH In review Shengzhuo Wei 3 participants 6 messages Apr 23 – Apr 24

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 proc: fix comm_write return value when truncated or error

When count exceeds TASK_COMM_LEN-1, comm_write() copies at most TASK_COMM_LEN-1 bytes but returns the original count. This violates write(2) semantics, which require returning the number of bytes actually written.

The count parameter is size_t and should not be repurposed to carry a negative error code on the same_thread_group() failure path.

Introduce a local len for the truncated length and a separate ssize_t ret for the return value.

Fixes 4614a696bd1c ("procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm")
---
 fs/proc/base.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
fs/proc/base.c +7 -4
index d9acfa89c894bd1608580331e1d5b3018c59123b..5d34590dbe9d9f05147c3e6b34c615cbf0984b1c 100644
@@ -1727,8 +1727,10 @@ static ssize_t comm_write(struct file *file, const char __user *buf,
 	struct task_struct *p;
 	char buffer[TASK_COMM_LEN] = {};
 	const size_t maxlen = sizeof(buffer) - 1;
+	size_t len = count > maxlen ? maxlen : count;
+	ssize_t ret;
 
-	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
+	if (copy_from_user(buffer, buf, len))
 		return -EFAULT;
 
 	p = get_proc_task(inode);
@@ -1738,13 +1740,14 @@ static ssize_t comm_write(struct file *file, const char __user *buf,
 	if (same_thread_group(current, p)) {
 		set_task_comm(p, buffer);
 		proc_comm_connector(p);
+		ret = len;
+	} else {
+		ret = -EINVAL;
 	}
-	else
-		count = -EINVAL;
 
 	put_task_struct(p);
 
-	return count;
+	return ret;
 }
 
 static int comm_show(struct seq_file *m, void *v)

---
base-commit: 2e68039281932e6dc37718a1ea7cbb8e2cda42e6
change-id: 20260424-fix_proc_write_return-cd48edb86600

Best regards,

Signature · 1 line
Shengzhuo Wei <[email protected]>
#1 lore raw
Quoted 10 lines · On Fri, 24 Apr 2026 04:06:21 +0800 "Shengzhuo Wei" <[email protected]> wrote:

When count exceeds TASK_COMM_LEN-1, comm_write() copies at most TASK_COMM_LEN-1 bytes but returns the original count. This violates write(2) semantics, which require returning the number of bytes actually written.

The count parameter is size_t and should not be repurposed to carry a negative error code on the same_thread_group() failure path.

Introduce a local len for the truncated length and a separate ssize_t ret for the return value.

Looks right to me.

Quoted 1 line
Fixes 4614a696bd1c ("procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm")

December 2009.

Hopefully no userspace is depending on the broken return value of a write to /proc/pid/comm.

Arguably we should leave the code as-is and add an apologetic comment explaining the situation.

#2 lore raw
Quoted 17 lines · Andrew Morton
Quoted 10 lines · On Fri, 24 Apr 2026 04:06:21 +0800 "Shengzhuo Wei" <[email protected]> wrote:

When count exceeds TASK_COMM_LEN-1, comm_write() copies at most TASK_COMM_LEN-1 bytes but returns the original count. This violates write(2) semantics, which require returning the number of bytes actually written.

The count parameter is size_t and should not be repurposed to carry a negative error code on the same_thread_group() failure path.

Introduce a local len for the truncated length and a separate ssize_t ret for the return value.

Looks right to me.

Quoted 1 line
Fixes 4614a696bd1c ("procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm")
Quoted 7 lines

December 2009.

Hopefully no userspace is depending on the broken return value of a write to /proc/pid/comm.

Arguably we should leave the code as-is and add an apologetic comment explaining the situation.

Yes, this issue must all over virtual filesystems.

Patch may break stuff in the other direction too:

if process is doing "full write" loop, and write hook gets string which is too long, then the last piece will be written, not truncated first part.

/proc/alexey

#3 lore raw
Quoted 10 lines · Alexey Dobriyan
Quoted 9 lines
Quoted 1 line
Fixes 4614a696bd1c ("procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm")

December 2009.

Hopefully no userspace is depending on the broken return value of a write to /proc/pid/comm.

Arguably we should leave the code as-is and add an apologetic comment explaining the situation.

First, there's a plain bug regardless: assigning -EINVAL to a size_t count and returning it as ssize_t produces a bogus positive value on the error path. Introducing a dedicated ssize_t ret fixes this without changing any success-path behavior, the change is on a permission failure path, so it's hard to imagine anyone depending on the broken behavior, but worth mentioning

Quoted 7 lines

Yes, this issue must all over virtual filesystems.

Patch may break stuff in the other direction too:

if process is doing "full write" loop, and write hook gets string which is too long, then the last piece will be written, not truncated first part.

For the truncation case, I agree returning a short count breaks the full-write idiom. But sliently truncating isnt't great either. Maybe we should reject overlong writes and return -EINVAL instead? That's also an ABI change and breaks the userspace, so it needs discussion, but at least it fails loudly.

Quoted 2 lines

/proc/alexey

Very happy to hear from your opinions and sorry for my poor English. Best regards.

#4 lore raw
Quoted 8 lines · Shengzhuo Wei
Quoted 7 lines · Alexey Dobriyan
Quoted 7 lines
Quoted 1 line
Fixes 4614a696bd1c ("procfs: allow threads to rename siblings via /proc/pid/tasks/tid/comm")

December 2009.

Hopefully no userspace is depending on the broken return value of a write to /proc/pid/comm.

Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

Arguably we should leave the code as-is and add an apologetic comment explaining the situation.

First, there's a plain bug regardless: assigning -EINVAL to a size_t count and returning it as ssize_t produces a bogus positive value on the error path.

There is no bug, write() will return -EINVAL as it should.

Quoted 3 lines

Introducing a dedicated ssize_t ret fixes this without changing any success-path behavior, the change is on a permission failure path, so it's hard to imagine anyone depending on the broken behavior, but worth mentioning

#5 lore raw
In reply to Shengzhuo Wei
Quoted 10 lines · On Fri, 24 Apr 2026 04:06:21 +0800 "Shengzhuo Wei" <[email protected]> wrote:

When count exceeds TASK_COMM_LEN-1, comm_write() copies at most TASK_COMM_LEN-1 bytes but returns the original count. This violates write(2) semantics, which require returning the number of bytes actually written.

The count parameter is size_t and should not be repurposed to carry a negative error code on the same_thread_group() failure path.

Introduce a local len for the truncated length and a separate ssize_t ret for the return value.

AI review asks a good question:

	https://sashiko.dev/#/patchset/[email protected]