For archival purposes, forwarding an incoming command email to
linux-kernel@vger.kernel.org, syzkaller-bugs@googlegroups.com.

***

Subject: [PATCH] PCI/proc: validate user buffer before touching config space
Author: kartikey406@gmail.com

#syz test: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master



proc_bus_pci_write() invokes __get_user() repeatedly without checking
its return value.  When the user pointer faults, the extable fixup
leaves the destination indeterminate but the function still hands the
value to pci_user_write_config_*(), writing fixup state to PCI
configuration space.

syzbot triggers this with a writev() whose iov_base is NULL on
/proc/bus/pci/00/03.0 (the virtio-blk controller in the syzkaller VM).
Every __get_user() faults, val ends up as fixup-zero, and zero is
written to config space offsets 0..6 -- including the Command register
at offset 4, clearing Bus Master and Memory Space Enable.  The disk
goes silent mid-flight, in-flight journal bios never complete, and
jbd2 hangs in wait_on_buffer() indefinitely:

  INFO: task jbd2/sda1-8:4955 blocked in I/O wait for more than 143 seconds.
   __wait_on_buffer fs/buffer.c:123
   jbd2_journal_commit_transaction+0x388a/0x6870 fs/jbd2/commit.c:837
   kjournald2 fs/jbd2/journal.c:201

proc_bus_pci_read() has the symmetric problem: an unchecked __put_user()
into a faulting user pointer silently drops config-space data.

Fix both by staging through a kernel buffer.  The write path uses
memdup_user() to copy the user buffer up front; a bad user pointer now
returns -EFAULT before any device state is touched, and the function
no longer races a concurrent userspace mutation of the buffer across
successive __get_user()s.  The read path collects config-space data
into a kmalloc'd buffer and copy_to_user()s it once at the end.

Reported-by: syzbot+c7604c9fdd7580cca4e0@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=c7604c9fdd7580cca4e0
Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
---
 drivers/pci/proc.c | 78 ++++++++++++++++++++++++----------------------
 1 file changed, 41 insertions(+), 37 deletions(-)

diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
index ce36e35681e8..aa54c4e1445d 100644
--- a/drivers/pci/proc.c
+++ b/drivers/pci/proc.c
@@ -31,13 +31,14 @@ static ssize_t proc_bus_pci_read(struct file *file, char __user *buf,
 	struct pci_dev *dev = pde_data(file_inode(file));
 	unsigned int pos = *ppos;
 	unsigned int cnt, size;
+	void *kbuf;
+	u8 *kp;
 
 	/*
 	 * Normal users can read only the standardized portion of the
 	 * configuration space as several chips lock up when trying to read
 	 * undefined locations (think of Intel PIIX4 as a typical example).
 	 */
-
 	if (capable(CAP_SYS_ADMIN))
 		size = dev->cfg_size;
 	else if (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
@@ -53,57 +54,61 @@ static ssize_t proc_bus_pci_read(struct file *file, char __user *buf,
 		nbytes = size - pos;
 	cnt = nbytes;
 
-	if (!access_ok(buf, cnt))
-		return -EINVAL;
+	kbuf = kmalloc(cnt, GFP_KERNEL);
+	if (!kbuf)
+		return -ENOMEM;
+	kp = kbuf;
 
 	pci_config_pm_runtime_get(dev);
 
 	if ((pos & 1) && cnt) {
-		unsigned char val;
-		pci_user_read_config_byte(dev, pos, &val);
-		__put_user(val, buf);
-		buf++;
+		pci_user_read_config_byte(dev, pos, kp);
+		kp++;
 		pos++;
 		cnt--;
 	}
 
 	if ((pos & 3) && cnt > 2) {
-		unsigned short val;
+		u16 val;
 		pci_user_read_config_word(dev, pos, &val);
-		__put_user(cpu_to_le16(val), (__le16 __user *) buf);
-		buf += 2;
+		*(__le16 *)kp = cpu_to_le16(val);
+		kp += 2;
 		pos += 2;
 		cnt -= 2;
 	}
 
 	while (cnt >= 4) {
-		unsigned int val;
+		u32 val;
 		pci_user_read_config_dword(dev, pos, &val);
-		__put_user(cpu_to_le32(val), (__le32 __user *) buf);
-		buf += 4;
+		*(__le32 *)kp = cpu_to_le32(val);
+		kp += 4;
 		pos += 4;
 		cnt -= 4;
 		cond_resched();
 	}
 
 	if (cnt >= 2) {
-		unsigned short val;
+		u16 val;
 		pci_user_read_config_word(dev, pos, &val);
-		__put_user(cpu_to_le16(val), (__le16 __user *) buf);
-		buf += 2;
+		*(__le16 *)kp = cpu_to_le16(val);
+		kp += 2;
 		pos += 2;
 		cnt -= 2;
 	}
 
 	if (cnt) {
-		unsigned char val;
-		pci_user_read_config_byte(dev, pos, &val);
-		__put_user(val, buf);
+		pci_user_read_config_byte(dev, pos, kp);
 		pos++;
 	}
 
 	pci_config_pm_runtime_put(dev);
 
+	if (copy_to_user(buf, kbuf, nbytes)) {
+		kfree(kbuf);
+		return -EFAULT;
+	}
+	kfree(kbuf);
+
 	*ppos = pos;
 	return nbytes;
 }
@@ -116,6 +121,8 @@ static ssize_t proc_bus_pci_write(struct file *file, const char __user *buf,
 	int pos = *ppos;
 	int size = dev->cfg_size;
 	int cnt, ret;
+	const u8 *kp;
+	u8 *kbuf;
 
 	ret = security_locked_down(LOCKDOWN_PCI_ACCESS);
 	if (ret)
@@ -129,56 +136,53 @@ static ssize_t proc_bus_pci_write(struct file *file, const char __user *buf,
 		nbytes = size - pos;
 	cnt = nbytes;
 
-	if (!access_ok(buf, cnt))
-		return -EINVAL;
+	kbuf = memdup_user(buf, cnt);
+	if (IS_ERR(kbuf))
+		return PTR_ERR(kbuf);
+	kp = kbuf;
 
 	pci_config_pm_runtime_get(dev);
 
 	if ((pos & 1) && cnt) {
-		unsigned char val;
-		__get_user(val, buf);
-		pci_user_write_config_byte(dev, pos, val);
-		buf++;
+		pci_user_write_config_byte(dev, pos, *kp);
+		kp++;
 		pos++;
 		cnt--;
 	}
 
 	if ((pos & 3) && cnt > 2) {
-		__le16 val;
-		__get_user(val, (__le16 __user *) buf);
+		__le16 val = *(const __le16 *)kp;
 		pci_user_write_config_word(dev, pos, le16_to_cpu(val));
-		buf += 2;
+		kp += 2;
 		pos += 2;
 		cnt -= 2;
 	}
 
 	while (cnt >= 4) {
-		__le32 val;
-		__get_user(val, (__le32 __user *) buf);
+		__le32 val = *(const __le32 *)kp;
 		pci_user_write_config_dword(dev, pos, le32_to_cpu(val));
-		buf += 4;
+		kp += 4;
 		pos += 4;
 		cnt -= 4;
 	}
 
 	if (cnt >= 2) {
-		__le16 val;
-		__get_user(val, (__le16 __user *) buf);
+		__le16 val = *(const __le16 *)kp;
 		pci_user_write_config_word(dev, pos, le16_to_cpu(val));
-		buf += 2;
+		kp += 2;
 		pos += 2;
 		cnt -= 2;
 	}
 
 	if (cnt) {
-		unsigned char val;
-		__get_user(val, buf);
-		pci_user_write_config_byte(dev, pos, val);
+		pci_user_write_config_byte(dev, pos, *kp);
 		pos++;
 	}
 
 	pci_config_pm_runtime_put(dev);
 
+	kfree(kbuf);
+
 	*ppos = pos;
 	i_size_write(ino, dev->cfg_size);
 	return nbytes;
-- 
2.43.0