Skip to contentSkip to conversation

hidraw.c: double mutex_lock

BUG iceberg 2 participants 2 messages Oct 12, 2009 HID

lore
Share & tools
lore

iceberg

author
#0 lore raw
	KERNEL_VERSION: 2.6.31
	DESCRIBE:
	In driver ./drivers/hid/hidraw.c in function hidraw_read may be 

double mutex_lock:

Path:
1. line 50: begin first iteration of "while(ret==0)"
2. line 52: first call to mutex_lock
3. inner loop "while (list->head == list->tail)" does not change state of mutex, because mutex_lock immediatelly follows mutex_unlock
4. if we go to the second iteration of "while(ret == 0)" in line 50 then there are second call to mutex_lock in line 52 (mutex aquired twice).

Second iteration of loop "while(ret==0)" is possible if local variable ret is not changed at line 94: ret+=len - i.e. len==0; Variable len may be zero if hidraw_read is called with count==0 or list->buffer[list->tail].len == 0.

static ssize_t hidraw_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
44{

  45        struct hidraw_list *list = file->private_data;
  46        int ret = 0, len;
  47        char *report;
  48        DECLARE_WAITQUEUE(wait, current);

49
50 while (ret == 0) {
51
52 mutex_lock(&list->read_mutex);
53
54 if (list->head == list->tail) {
55 add_wait_queue(&list->hidraw->wait, &wait);
56 set_current_state(TASK_INTERRUPTIBLE);
57
58 while (list->head == list->tail) {
59 if (file->f_flags & O_NONBLOCK) {
60 ret = -EAGAIN;
61 break;
62 }
63 if (signal_pending(current)) {
64 ret = -ERESTARTSYS;
65 break;
66 }
67 if (!list->hidraw->exist) {
68 ret = -EIO;
69 break;
70 }
71
72 /* allow O_NONBLOCK to work well
from other threads */

  73                                mutex_unlock(&list->read_mutex);
  74                                schedule();
  75                                mutex_lock(&list->read_mutex);
  76                                set_current_state

(TASK_INTERRUPTIBLE);
77 }
78
79 set_current_state(TASK_RUNNING);
80 remove_wait_queue(&list->hidraw->wait,
&wait);
81 }
82
83 if (ret)
84 goto out;
85
86 report = list->buffer[list->tail].value;
87 len = list->buffer[list->tail].len > count ?
88 count : list->buffer[list->tail].len;
89
90 if (copy_to_user(buffer, list->buffer[list-

Quoted 1 line

tail].value, len)) {

91 ret = -EFAULT;
92 goto out;
93 }
94 ret += len;
95
96 kfree(list->buffer[list->tail].value);
97 list->tail = (list->tail + 1) & (HIDRAW_BUFFER_SIZE
- 1);
98 }
99out:
100 mutex_unlock(&list->read_mutex);
101 return ret;
102}

#1 lore raw
Quoted 14 lines · iceberg
	KERNEL_VERSION: 2.6.31
	DESCRIBE:
	In driver ./drivers/hid/hidraw.c in function hidraw_read may be 

double mutex_lock:

Path:
1. line 50: begin first iteration of "while(ret==0)"
2. line 52: first call to mutex_lock
3. inner loop "while (list->head == list->tail)" does not change state of mutex, because mutex_lock immediatelly follows mutex_unlock
4. if we go to the second iteration of "while(ret == 0)" in line 50 then there are second call to mutex_lock in line 52 (mutex aquired twice).

Quoted 4 lines

Second iteration of loop "while(ret==0)" is possible if local variable ret is not changed at line 94: ret+=len - i.e. len==0; Variable len may be zero if hidraw_read is called with count==0 or list->buffer[list->tail].len == 0.

Good catch. I will fix that up by moving the mutex_lock() so that it's locked before the loop is entered.

Thanks,

Signature · 2 lines
Jiri Kosina
SUSE Labs, Novell Inc.