Created attachment 994408 [details]
call fflush() before checking ferror() on job file descriptor

Description of problem:

at does not properly call fflush() before checking ferror on job file. 

If filesystem (/var) is full or almost full when setting at job is attempted, a 0byte or truncated file is created.

This causes a spew of 

File <filename> is in wrong format - aborting 

to syslog which can fill up /var again very quickly.

Version-Release number of selected component (if applicable):

3.1.10

How reproducible:

Happens if filesystem containing /var/spool/at is full.

Steps to Reproduce:
1. Fill up /var
2. set a small job with `at`
3. observe that the file created is 0bytes long
4. wait for the scheduled time

Actual results:

at does not properly remove at file when write error to the file occurs. This leaves behind a 0byte or potentially truncated job file.

0byte job files are created and when scheduled time is reached, atd complains profusely with:

File <filename> is in wrong format - aborting

If written to /var/log/messages (which is default), file grows rapidly until jobs are removed with atrm.

If /var is already full, and messages is cleared out without cleaning up the bad at jobs, /var fills up again very soon.

Expected results:

at should reach 

    fprintf(fp, "\n");
    if (ferror(fp))
        panic("Output error");

so panic() cleans up the malformed atfile. 

Need to fflush() before checking ferror() in case the write() returns an error.

Additional info:

Patch attached, built against 3.1.10 and tested on Scientific Linux 6.6.