Bug 30223

This isn't a bug in the threads library, it is a bug
in the driver :

/*
 * mydev_write takes in a character string and saves it
 * to buf for later accessing.
 */
int
mydev_write(dev_t dev, struct uio *uio, int ioflag)
{
    int err = 0;

    printf("mydev_write: dev_t=%d, uio=%p, ioflag=%d\n",
        dev2udev(dev), uio, ioflag);

    err = copyinstr(uio->uio_iov->iov_base, &buf, 512, &len);
    if (err != 0) {
        printf("Write to \"cdev\" failed.\n");
    }
    return(err);
}

if err is 0 it mydev_write should be returning len, the 
number of bytes copied.

It isn't returning because the thread library does a 
non-blocking write and simulates a blocking write by 
doing writes piecemeal:

                  /*
                   * Loop while no error occurs and until the expected number
                   * of bytes are written if performing a blocking write:
                   */
                  while (ret == 0) {
                          /* Perform a non-blocking write syscall: */
                          n = __sys_write(fd, buf + num, nbytes - num);

sys_write is returning 0 so it keeps on writing the
string over and over again.
write(5,0x804988c,14)                            = 0
(0x0)
poll(0x8059000,0x1,0x0)                          = 1
(0x1)
write(5,0x804988c,14)                            = 0
(0x0)
poll(0x8059000,0x1,0x0)                          = 1
(0x1)

The following patch will make the driver work:
> diff -Naur cdev.c.bak cdev.c
--- cdev.c.bak  Sun Jan  6 18:54:54 2002
+++ cdev.c      Sun Jan  6 20:03:51 2002
@@ -149,16 +149,14 @@
 int
 mydev_write(dev_t dev, struct uio *uio, int ioflag)
 {
-    int err = 0;
+    int towrite = 0;
 
     printf("mydev_write: dev_t=%d, uio=%p, ioflag=%d\n",
        dev2udev(dev), uio, ioflag);
 
-    err = copyinstr(uio->uio_iov->iov_base, &buf, 512, &len);
-    if (err != 0) {
-       printf("Write to \"cdev\" failed.\n");
-    }
-    return(err);
+    towrite = (min(uio->uio_resid, sizeof(buf)));
+    len = towrite;
+    return(uiomove(buf, towrite, uio));
 }
 
 /*

Although the ioctl is still getting called incorrectly
twice when compiled with pthread:
once after open:
open("/dev/cdev",0x2,00)                         = 5
(0x5)
fcntl(0x5,0x3,0x0)                               = 2
(0x2)
fcntl(0x5,0x4,0x6)                              
ERR#22 'Invalid argument'

and once before exit:
fcntl(0x5,0x4,0x2)                              
ERR#22 'Invalid argument'
exit(0x0)                                      
process exit, rval = 0

What is doing is setting the nonblocking flag on open,
and unsetting on exit and this trivial driver doesn't
support having flags set on it:
        /*
         * Enter a loop to set all file descriptors to blocking
         * if they were not created as non-blocking:
         */
        for (i = 0; i < _thread_dtablesize; i++) {
                /* Check if this file descriptor is in use: */
                if (_thread_fd_table[i] != NULL &&
                        !(_thread_fd_table[i]->flags & O_NONBLOCK)) {
                        /* Get the current flags: */
                        flags = _thread_sys_fcntl(i, F_GETFL, NULL);
                        /* Clear the nonblocking file descriptor flag: */
                        _thread_sys_fcntl(i, F_SETFL, flags & ~O_NONBLOCK);
                }
        }

        /* Call the _exit syscall: */
        _thread_sys__exit(status);




__________________________________________________
Do You Yahoo!?
Send FREE video emails in Yahoo! Mail!
http://promo.yahoo.com/videomail/

Comment 2 Remko Lodder 2007-03-07 21:19:43 UTC

State Changed
From-To: open->closed

It was mentioned that this was a bug in the kld code itself (from the 
submitter), no feedback after that so I am assuming this was correct. 
Please prove me my mistake if needed...