usb: mon: Fix a deadlock in usbmon between mmap and read
authorPete Zaitcev <zaitcev@redhat.com>
Thu, 5 Dec 2019 02:39:41 +0000 (20:39 -0600)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 10 Dec 2019 10:41:20 +0000 (11:41 +0100)
The problem arises because our read() function grabs a lock of the
circular buffer, finds something of interest, then invokes copy_to_user()
straight from the buffer, which in turn takes mm->mmap_sem. In the same
time, the callback mon_bin_vma_fault() is invoked under mm->mmap_sem.
It attempts to take the fetch lock and deadlocks.

This patch does away with protecting of our page list with any
semaphores, and instead relies on the kernel not close the device
while mmap is active in a process.

In addition, we prohibit re-sizing of a buffer while mmap is active.
This way, when (now unlocked) fault is processed, it works with the
page that is intended to be mapped-in, and not some other random page.
Note that this may have an ABI impact, but hopefully no legitimate
program is this wrong.

Signed-off-by: Pete Zaitcev <zaitcev@redhat.com>
Reported-by: syzbot+56f9673bb4cdcbeb0e92@syzkaller.appspotmail.com
Reviewed-by: Alan Stern <stern@rowland.harvard.edu>
Fixes: 46eb14a6e158 ("USB: fix usbmon BUG trigger")
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20191204203941.3503452b@suzdal.zaitcev.lan
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/usb/mon/mon_bin.c

index ac2b4fcc265f65c10a128bcf092de3a134319dc8..f48a23adbc35ddbbc66c5227d8c59e3413791b05 100644 (file)
@@ -1039,12 +1039,18 @@ static long mon_bin_ioctl(struct file *file, unsigned int cmd, unsigned long arg
 
                mutex_lock(&rp->fetch_lock);
                spin_lock_irqsave(&rp->b_lock, flags);
-               mon_free_buff(rp->b_vec, rp->b_size/CHUNK_SIZE);
-               kfree(rp->b_vec);
-               rp->b_vec  = vec;
-               rp->b_size = size;
-               rp->b_read = rp->b_in = rp->b_out = rp->b_cnt = 0;
-               rp->cnt_lost = 0;
+               if (rp->mmap_active) {
+                       mon_free_buff(vec, size/CHUNK_SIZE);
+                       kfree(vec);
+                       ret = -EBUSY;
+               } else {
+                       mon_free_buff(rp->b_vec, rp->b_size/CHUNK_SIZE);
+                       kfree(rp->b_vec);
+                       rp->b_vec  = vec;
+                       rp->b_size = size;
+                       rp->b_read = rp->b_in = rp->b_out = rp->b_cnt = 0;
+                       rp->cnt_lost = 0;
+               }
                spin_unlock_irqrestore(&rp->b_lock, flags);
                mutex_unlock(&rp->fetch_lock);
                }
@@ -1216,13 +1222,21 @@ mon_bin_poll(struct file *file, struct poll_table_struct *wait)
 static void mon_bin_vma_open(struct vm_area_struct *vma)
 {
        struct mon_reader_bin *rp = vma->vm_private_data;
+       unsigned long flags;
+
+       spin_lock_irqsave(&rp->b_lock, flags);
        rp->mmap_active++;
+       spin_unlock_irqrestore(&rp->b_lock, flags);
 }
 
 static void mon_bin_vma_close(struct vm_area_struct *vma)
 {
+       unsigned long flags;
+
        struct mon_reader_bin *rp = vma->vm_private_data;
+       spin_lock_irqsave(&rp->b_lock, flags);
        rp->mmap_active--;
+       spin_unlock_irqrestore(&rp->b_lock, flags);
 }
 
 /*
@@ -1234,16 +1248,12 @@ static vm_fault_t mon_bin_vma_fault(struct vm_fault *vmf)
        unsigned long offset, chunk_idx;
        struct page *pageptr;
 
-       mutex_lock(&rp->fetch_lock);
        offset = vmf->pgoff << PAGE_SHIFT;
-       if (offset >= rp->b_size) {
-               mutex_unlock(&rp->fetch_lock);
+       if (offset >= rp->b_size)
                return VM_FAULT_SIGBUS;
-       }
        chunk_idx = offset / CHUNK_SIZE;
        pageptr = rp->b_vec[chunk_idx].pg;
        get_page(pageptr);
-       mutex_unlock(&rp->fetch_lock);
        vmf->page = pageptr;
        return 0;
 }