| Message ID | e5946d67-4e5e-b056-ba80-656bab12d9f6@kernel.dk (mailing list archive) |
|---|---|
| State | Mainlined, archived |
| Headers | show |
| Series | pipe: check for IOCB_NOWAIT alongside O_NONBLOCK | expand |
On Tue, 09 May 2023 09:12:24 -0600, Jens Axboe wrote: > Pipe reads or writes need to enable nonblocking attempts, if either > O_NONBLOCK is set on the file, or IOCB_NOWAIT is set in the iocb being > passed in. The latter isn't currently true, ensure we check for both > before waiting on data or space. > > Thanks. I picked this up and pushed a tag. I'll get this to Linus before -rc2, Jens. tree: git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git tag: vfs/v6.4-rc1/pipe [1/1] pipe: check for IOCB_NOWAIT alongside O_NONBLOCK commit: d326147107eb7152941ab3b04ed4ffba8ce93f64
diff --git a/fs/pipe.c b/fs/pipe.c index ceb17d2dfa19..c88611c612ba 100644 --- a/fs/pipe.c +++ b/fs/pipe.c @@ -342,7 +342,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to) break; if (ret) break; - if (filp->f_flags & O_NONBLOCK) { + if (filp->f_flags & O_NONBLOCK || iocb->ki_flags & IOCB_NOWAIT) { ret = -EAGAIN; break; } @@ -547,7 +547,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from) continue; /* Wait for buffer space to become available. */ - if (filp->f_flags & O_NONBLOCK) { + if (filp->f_flags & O_NONBLOCK || iocb->ki_flags & IOCB_NOWAIT) { if (!ret) ret = -EAGAIN; break;
Pipe reads or writes need to enable nonblocking attempts, if either O_NONBLOCK is set on the file, or IOCB_NOWAIT is set in the iocb being passed in. The latter isn't currently true, ensure we check for both before waiting on data or space. Fixes: afed6271f5b0 ("pipe: set FMODE_NOWAIT on pipes") Signed-off-by: Jens Axboe <axboe@kernel.dk> --- With the pipe nonblock rework, these two hunsk from one of the other patches was inadvertently dropped. We obviously need to check both, or files without O_NONBLOCK but IOCB_NOWAIT set in the iocb will still block on space/data when writing/reading to/from a pipe. I can just include this in the io_uring changes as it directly impacts io_uring (breaking a few test cases), or you guys can queue it up on the VFS side. Would be great to get this in -rc2 as it's stalling the regression testing with the vanilla kernels.