tanya@adds.newyork.NCR.COM (Tanya Katz) (06/02/90)
Calling poll(2) with nfds > NPOLLFILE causes corruption of the pollfd structure at offsets > NPOLLFILE. The O/S we are using is Unix v.3 Release 1.01 for the NCR Tower 32/700. If nfds > NPOLLFILE, the kernel code loops through the pollfd array in chunks of NPOLLFILE. The problem arises when copyin() is called to copy user data into kernel space. If nfds > NPOLLFILE the pointer to the user pollfd structure is never advanced and the copyin() routine always copies the same user data into the kernel structure. When copyout() is called the user data, beginning with offsets > NPOLLFILE, is overwritten with the data from the beginning of the table. Anyone have similar experiences? -Tanya #-------------------------------------------------# | Tanya Katz (516) 231-5400 x430 | | | | ...uunet!ncrlnk!adds!tanya | | tanya.katz@adds.newyork.ncr.com | | | | ADDS Inc, 100 Marcus Blvd, Hauppauge, NY 11788 | #-------------------------------------------------#
sar0@cbnewsl.att.com (stephen.a.rago) (06/10/90)
In article <1300@adds.newyork.NCR.COM>, tanya@adds.newyork.NCR.COM (Tanya Katz) writes: > > Calling poll(2) with nfds > NPOLLFILE causes corruption of the pollfd > structure at offsets > NPOLLFILE. > The O/S we are using is Unix v.3 Release 1.01 for the NCR Tower 32/700. > If nfds > NPOLLFILE, the kernel code loops through the pollfd array > in chunks of NPOLLFILE. The problem arises when copyin() is called > to copy user data into kernel space. > If nfds > NPOLLFILE the pointer to the user pollfd structure is never > advanced and the copyin() routine always copies the same user data > into the kernel structure. > When copyout() is called the user data, beginning with offsets > > NPOLLFILE, is overwritten with the data from the beginning of the table. > Anyone have similar experiences? Yeah, SVR3.2 fixes this problem. Steve Rago sar@attunix.att.com