From Bugzilla Helper: User-Agent: Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1) Description of problem: We do a non-timeout select on an fd pointing to a pipe. After select returns, we do an fstat on that fd to determine the number of bytes available on the pipe. The st_size member of the structure contains 0. This works properly on Red Hat Linux 6.2, Solaris 8, Irix 6.5, HP-UX 11.0, and Tru64 Unix 4.0f. This problem makes it impossible to run our application on 7.2. Version-Release number of selected component (if applicable): How reproducible: Always Steps to Reproduce: 1.cc -o parent parent.c 2.cc -o child child.c 3../parent Actual Results: The "after select fstat reports" value is zero. Expected Results: The "after select fstat reports" value should be positive, as it is on all the other tested systems. Additional info:
Created attachment 63884 [details] gzipped tar file containing two c programs.
Your application uses undefined behavior as per posix. Please fix it! Older linux version accidentally had values in the stat fields that APPEARED to be what others did, however these were the result of a rather serious bug in the VFS. Having said that, for 7.2 we made an erratum kernel that deliberately puts this random value back into the field for the case your application depends on. However this is a temporary hack and it will go away in the future, esp when the zero-copy pipes get implemented. I'm closing this "fixed in errata" since we released several errata kernels with the workaround that everyone should be using anyway due to the VERY important security fixes in them.
Thanks for taking the time to look at this. Two questions: 1. What is the correct way to find out how much data is waiting on a pipe? 2. Where do I get the errata kernel? Dick Jay