libnvdimm/pmem: Bypass CONFIG_HARDENED_USERCOPY overhead
commite71556e3684d6b11592bbdbe5406bbe1d73ab706
authorDan Williams <dan.j.williams@intel.com>
Fri, 17 May 2019 00:05:21 +0000 (16 17:05 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 31 May 2019 13:47:11 +0000 (31 06:47 -0700)
treeb62731887ed93fb503ac1c94cc9f804f55681aa4
parent74d76947b1dc6389941e90e4a8452b89b8b239ce
libnvdimm/pmem: Bypass CONFIG_HARDENED_USERCOPY overhead

commit 52f476a323f9efc959be1c890d0cdcf12e1582e0 upstream.

Jeff discovered that performance improves from ~375K iops to ~519K iops
on a simple psync-write fio workload when moving the location of 'struct
page' from the default PMEM location to DRAM. This result is surprising
because the expectation is that 'struct page' for dax is only needed for
third party references to dax mappings. For example, a dax-mapped buffer
passed to another system call for direct-I/O requires 'struct page' for
sending the request down the driver stack and pinning the page. There is
no usage of 'struct page' for first party access to a file via
read(2)/write(2) and friends.

However, this "no page needed" expectation is violated by
CONFIG_HARDENED_USERCOPY and the check_copy_size() performed in
copy_from_iter_full_nocache() and copy_to_iter_mcsafe(). The
check_heap_object() helper routine assumes the buffer is backed by a
slab allocator (DRAM) page and applies some checks.  Those checks are
invalid, dax pages do not originate from the slab, and redundant,
dax_iomap_actor() has already validated that the I/O is within bounds.
Specifically that routine validates that the logical file offset is
within bounds of the file, then it does a sector-to-pfn translation
which validates that the physical mapping is within bounds of the block
device.

Bypass additional hardened usercopy overhead and call the 'no check'
versions of the copy_{to,from}_iter operations directly.

Fixes: 0aed55af8834 ("x86, uaccess: introduce copy_from_iter_flushcache...")
Cc: <stable@vger.kernel.org>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <willy@infradead.org>
Reported-and-tested-by: Jeff Smits <jeff.smits@intel.com>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/nvdimm/pmem.c