From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andreas Gruenbacher Date: Tue, 26 Oct 2021 11:44:30 +0200 Subject: [Cluster-devel] [PATCH v8 00/17] gfs2: Fix mmap + page fault deadlocks In-Reply-To: References: <20211019134204.3382645-1-agruenba@redhat.com> Message-ID: <20211026094430.3669156-1-agruenba@redhat.com> List-Id: To: cluster-devel.redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Ted, here's an updated version of Dave Hansen's original commit, but note that generic/208 won't run on ext4 with data journaling enabled: $ MOUNT_OPTIONS='-o data=journal' TEST_DIR=/mnt/test TEST_DEV=/dev/vdb ./tests/generic/208 QA output created by 208 208 not run: ext4 data journaling doesn't support O_DIRECT Thanks, Andreas -- Based on commit 998ef75ddb57 ("fs: do not prefault sys_write() user buffer pages") by Dave Hansen , but: * Fix generic_perform_write as well as iomap_write_iter. * copy_page_from_iter_atomic() doesn't trigger page faults, so there's no need to disable page faults around it [see commit 9e8c2af96e0d ("callers of iov_copy_from_user_atomic() don't need pagecache_disable()")]. * If fault_in_iov_iter_readable() fails to fault in the entire buffer, we still want to read everything up to the fault position. This depends on commit a6294593e8a1 ("iov_iter: Turn iov_iter_fault_in_readable into fault_in_iov_iter_readable"). Signed-off-by: Andreas Gruenbacher --- fs/iomap/buffered-io.c | 20 +++++++------------- mm/filemap.c | 20 +++++++------------- 2 files changed, 14 insertions(+), 26 deletions(-) diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c index 1753c26c8e76..d8809cd9ab31 100644 --- a/fs/iomap/buffered-io.c +++ b/fs/iomap/buffered-io.c @@ -744,17 +744,6 @@ static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i) if (bytes > length) bytes = length; - /* - * Bring in the user page that we'll copy from _first_. - * Otherwise there's a nasty deadlock on copying from the - * same page as we're writing to, without it being marked - * up-to-date. - */ - if (unlikely(fault_in_iov_iter_readable(i, bytes))) { - status = -EFAULT; - break; - } - status = iomap_write_begin(iter, pos, bytes, &page); if (unlikely(status)) break; @@ -777,9 +766,14 @@ static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i) * halfway through, might be a race with munmap, * might be severe memory pressure. */ - if (copied) + if (copied) { bytes = copied; - goto again; + goto again; + } + if (fault_in_iov_iter_readable(i, bytes) != bytes) + goto again; + status = -EFAULT; + break; } pos += status; written += status; diff --git a/mm/filemap.c b/mm/filemap.c index 4dd5edcd39fd..467cdb7d086d 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -3751,17 +3751,6 @@ ssize_t generic_perform_write(struct file *file, iov_iter_count(i)); again: - /* - * Bring in the user page that we will copy from _first_. - * Otherwise there's a nasty deadlock on copying from the - * same page as we're writing to, without it being marked - * up-to-date. - */ - if (unlikely(fault_in_iov_iter_readable(i, bytes))) { - status = -EFAULT; - break; - } - if (fatal_signal_pending(current)) { status = -EINTR; break; @@ -3794,9 +3783,14 @@ ssize_t generic_perform_write(struct file *file, * halfway through, might be a race with munmap, * might be severe memory pressure. */ - if (copied) + if (copied) { bytes = copied; - goto again; + goto again; + } + if (fault_in_iov_iter_readable(i, bytes) != bytes) + goto again; + status = -EFAULT; + break; } pos += status; written += status; -- 2.26.3