This patch set is my attempt to make quota code more scalable.
Main goal of this patch-set is to split global locking to per-sb basis.
Actually it consists of several parts
* Fixes : trivial fixes which i hope will be accepted w/o any complain
* Splitup global locks: Imho this part clean and simple. I hope it is
also a sane candidate for for_testing branch.
* More scalability for single sb : Some of this patches was already
submitted previously, some wasn't. This part is just my first vision
of the way we can move. This way result in real speedup, but i'm not
shure about design solutions, please do not punch me too strong
if you dont like that direction.
This patch-set survived after some stress testing
* parallel quota{on,off}
* fssress
* triggering ENOSPC
More info here: download.openvz.org/~dmonakhov/quota.html
Changes from V2
* Move getfmt call to dquot (suggested by hch@)
* Use global hash with per backet lock (suggested by viro@)
* Protect dqget with SRCU to prevent race on quota_info ptr
* Add dquot_init optimization
* Remove data_lock for ocfs2 where possible.
* I've remove dq_count optimization patch because it was buggy,
and in fact it should belongs to another patch-set.
* Bug fixes
** Fix deadlock on dquot transfer due to previous ENOSPC
Changes from V1
* random fixes according to Jan's comments
+ fix spelling
+ fix deadlock on dquot_transfer, and lock_dep issues
- list_lock patches split is still the same as before.
* move quota data from sb to dedicated pointer.
* Basic improvements fore per-sb scalability
patch against 2.6.36-rc5, linux-fs-2.6.git for_testing branch
<Out of tree patches from other developers>
kernel: add bl_list
<Cleanups and Fixes>
quota: protect getfmt call with dqonoff_mutex lock
quota: Wrap common expression to helper function
<Split-up global locks>
quota: protect dqget() from parallels quotaoff via SRCU
quota: mode quota internals from sb to quota_info
quota: Remove state_lock
quota: add quota format lock
quota: make dquot lists per-sb
quota: optimize quota_initialize
quota: user per-backet hlist lock for dquot_hash
quota: remove global dq_list_lock
<More scalability for single sb>
quota: rename dq_lock
quota: make per-sb dq_data_lock
quota: protect dquot mem info with object's lock
quota: drop dq_data_lock where possible
quota: relax dq_data_lock dq_lock locking consistency
quota: Some stylistic cleanup for dquot interface
fs: add unlocked helpers
quota: protect i_dquot with i_lock instead of dqptr_sem
Signed-off-by: Dmitry Monakhov <dmonakhov@xxxxxxxxxx>
---
Makefile | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/Makefile b/Makefile
index 471c49f..4e7602b 100644
--- a/Makefile
+++ b/Makefile
@@ -1,7 +1,7 @@
VERSION = 2
PATCHLEVEL = 6
SUBLEVEL = 36
-EXTRAVERSION = -rc6
+EXTRAVERSION = -rc6-quota
NAME = Sheep on Meth
# *DOCUMENTATION*
--
1.6.5.2
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
