| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
When we're doing btree updates from journal flush, this becomes a
locking inversion
|
|
|
|
|
|
|
|
|
|
the assertion was meant to check that bch2_journal_reclaim_fast() was
always being called, but since the atomic dec can happen outside of
j->lock the assertion itself can race
|
|
|
|
|
|
|
|
|
|
Checking if we can do the insert after getting the journal reservation
means potentially wasting space in the journal, which will break the new
pre reservation mechanism
|
|
|
|
|
|
|
|
the btree trans iterator stuff relies on a dirty horrible hack that uses
a short backtrace to generate tokens, and gcc uninlining these functions
breaks that hack. ugh.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This fixes a lockdep recursion when using __vmalloc from places that
aren't GFP_KERNEL safe.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
