gfs2: No short reads or writes upon glock contention
authorAndreas Gruenbacher <agruenba@redhat.com>
Thu, 28 Apr 2022 12:51:33 +0000 (14:51 +0200)
committerAndreas Gruenbacher <agruenba@redhat.com>
Thu, 28 Apr 2022 13:14:48 +0000 (15:14 +0200)
commit296abc0d91d8b65d42224dd33452ace14491ad08
treeae0e376e354a496575c2d53623bee98a4a768fdb
parent4fad37d595b9d9a2996467d780cb2e7a1b08b2c0
gfs2: No short reads or writes upon glock contention

Commit 00bfe02f4796 ("gfs2: Fix mmap + page fault deadlocks for buffered
I/O") changed gfs2_file_read_iter() and gfs2_file_buffered_write() to
allow dropping the inode glock while faulting in user buffers.  When the
lock was dropped, a short result was returned to indicate that the
operation was interrupted.

As pointed out by Linus (see the link below), this behavior is broken
and the operations should always re-acquire the inode glock and resume
the operation instead.

Link: https://lore.kernel.org/lkml/CAHk-=whaz-g_nOOoo8RRiWNjnv2R+h6_xk2F1J4TuSRxk1MtLw@mail.gmail.com/
Fixes: 00bfe02f4796 ("gfs2: Fix mmap + page fault deadlocks for buffered I/O")
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
fs/gfs2/file.c