Commit d5b81454 authored by Andreas Gruenbacher's avatar Andreas Gruenbacher
Browse files

Revert "gfs2: Fix mmap locking for write faults" 

This reverts commit b7f55d92.

As explained by Linus in [*], write faults on a mmap region are reads
from a filesysten point of view, so taking the inode glock exclusively
on write faults is incorrect.

Instead, when a page is marked writable, the .page_mkwrite vm operation
will be called, which is where the exclusive lock taking needs to
happen.  I got this wrong because of a broken test case that made me
believe .page_mkwrite isn't getting called when it actually is.

[*] https://lore.kernel.org/lkml/CAHk-=wj8EWr_D65i4oRSj2FTbrc6RdNydNNCGxeabRnwtoU=3Q@mail.gmail.com/



Signed-off-by: default avatarAndreas Gruenbacher <agruenba@redhat.com>
parent c2131f7e

fs/gfs2/file.c

+1 −3
Original line number Diff line number Diff line
@@ -540,11 +540,9 @@ static vm_fault_t gfs2_fault(struct vm_fault *vmf) 
	struct gfs2_inode *ip = GFS2_I(inode);

	struct gfs2_holder gh;

	vm_fault_t ret;

	u16 state;

	int err;



	state = (vmf->flags & FAULT_FLAG_WRITE) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;

	gfs2_holder_init(ip->i_gl, state, 0, &gh);

	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);

	err = gfs2_glock_nq(&gh);

	if (err) {

		ret = block_page_mkwrite_return(err);