пн, 25 янв. 2021 г. в 09:07, Shyam Prasad N <nspmangalore@xxxxxxxxx>: > > From my readings so far, -ERESTARTSYS assumes that the syscall is idempotent. > Can we safely make such assumptions for all VFS file operations? For > example, what happens if a close gets restarted, and we already > scheduled a delayed close because the original close failed with > ERESTARTSYS? I don't think we can assume all system calls to be idempotent. We should probably examine them one-by-one and return -ERESTARTSYS for some and -EINTR for the others. > > Also, I ran a quick grep for EINTR, and it looks like __cifs_readv and > __cifs_writev return EINTR too. Should those be replaced by > ERESTARTSYS too? > They return -EINTR after receiving a kill signal (see wait_for_completion_killable around those lines). It doesn't seem there is any point in returning -ERESTARTSYS after a kill signal anyway, so, I think the code is correct there. -- Best regards, Pavel Shilovsky
