Possible improper locking issues

[ryancaicse]

HI, in this function, it is possible that the lock lock (Line 71) is not released properly because the execution does not enter the while(j < num_urls)? Also, it also seems that the lock lock could be released multiple times in the while loop at Line 97.

Thanks for checking.

curl/docs/examples/smooth-gtk-thread.c

Lines 68 to 113 in 3085ccf

	void *pull_one_url(void *NaN)
	{
	/* Stop threads from entering unless j is incremented */
	pthread_mutex_lock(&lock);
	while(j < num_urls) {
	CURL *curl;
	gchar *http;
	printf("j = %d\n", j);
	http =
	g_strdup_printf("xoap.weather.com/weather/local/%s?cc=*&dayf=5&unit=i\n",
	urls[j]);
	printf("http %s", http);
	curl = curl_easy_init();
	if(curl) {
	FILE *outfile = fopen(urls[j], "wb");
	/* Set the URL and transfer type */
	curl_easy_setopt(curl, CURLOPT_URL, http);
	/* Write to the file */
	curl_easy_setopt(curl, CURLOPT_WRITEDATA, outfile);
	curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, write_file);
	j++; /* critical line */
	pthread_mutex_unlock(&lock);
	curl_easy_perform(curl);
	fclose(outfile);
	printf("fclose\n");
	curl_easy_cleanup(curl);
	}
	g_free(http);
	/* Adds more latency, testing the mutex.*/
	sleep(1);
	} /* end while */
	return NULL;
	}

[bagder]

It does indeed look very wrong. You up for writing a pull request and fix it?

[ryancaicse]

Sorry, I don't know which is the best way to fix them due to being unaware of the shared data protected by the lock. But, here is my try. Is this enough to fix this issue?

 void *pull_one_url(void *NaN) 
 { 
   /* Stop threads from entering unless j is incremented */ 
   pthread_mutex_lock(&lock); 
   while(j < num_urls) { 
     ...;
     if(curl) { 
      ...;
       pthread_mutex_unlock(&lock); 
  
       ...;
       pthread_mutex_lock(&lock); // new added
     } 
     ...;
   } /* end while */ 
    pthread_mutex_unlock(&lock); // new added
   return NULL; 
 }

[bagder]

I think it needs more than that. See #7931 for my proposed fix.

[ryancaicse]

OK, thanks