On 2/28/21 8:39 AM, Brian Foster wrote:
On Thu, Feb 25, 2021 at 03:28:00PM -0700, Allison Henderson wrote:
On 2/25/21 7:22 AM, Brian Foster wrote:
On Thu, Feb 25, 2021 at 12:01:10AM -0700, Allison Henderson wrote:
On 2/24/21 11:45 AM, Brian Foster wrote:
On Thu, Feb 18, 2021 at 09:53:37AM -0700, Allison Henderson wrote:
This patch modifies the attr remove routines to be delay ready. This
means they no longer roll or commit transactions, but instead return
-EAGAIN to have the calling routine roll and refresh the transaction. In
this series, xfs_attr_remove_args has become xfs_attr_remove_iter, which
uses a sort of state machine like switch to keep track of where it was
when EAGAIN was returned. xfs_attr_node_removename has also been
modified to use the switch, and a new version of xfs_attr_remove_args
consists of a simple loop to refresh the transaction until the operation
is completed. A new XFS_DAC_DEFER_FINISH flag is used to finish the
transaction where ever the existing code used to.
Calls to xfs_attr_rmtval_remove are replaced with the delay ready
version __xfs_attr_rmtval_remove. We will rename
__xfs_attr_rmtval_remove back to xfs_attr_rmtval_remove when we are
done.
xfs_attr_rmtval_remove itself is still in use by the set routines (used
during a rename). For reasons of preserving existing function, we
modify xfs_attr_rmtval_remove to call xfs_defer_finish when the flag is
set. Similar to how xfs_attr_remove_args does here. Once we transition
the set routines to be delay ready, xfs_attr_rmtval_remove is no longer
used and will be removed.
This patch also adds a new struct xfs_delattr_context, which we will use
to keep track of the current state of an attribute operation. The new
xfs_delattr_state enum is used to track various operations that are in
progress so that we know not to repeat them, and resume where we left
off before EAGAIN was returned to cycle out the transaction. Other
members take the place of local variables that need to retain their
values across multiple function recalls. See xfs_attr.h for a more
detailed diagram of the states.
Signed-off-by: Allison Henderson <allison.henderson@xxxxxxxxxx>
---
fs/xfs/libxfs/xfs_attr.c | 223 +++++++++++++++++++++++++++++-----------
fs/xfs/libxfs/xfs_attr.h | 100 ++++++++++++++++++
fs/xfs/libxfs/xfs_attr_leaf.c | 2 +-
fs/xfs/libxfs/xfs_attr_remote.c | 48 +++++----
fs/xfs/libxfs/xfs_attr_remote.h | 2 +-
fs/xfs/xfs_attr_inactive.c | 2 +-
6 files changed, 294 insertions(+), 83 deletions(-)
diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
index 56d4b56..d46b92a 100644
--- a/fs/xfs/libxfs/xfs_attr.c
+++ b/fs/xfs/libxfs/xfs_attr.c
...
@@ -1285,51 +1365,74 @@ xfs_attr_node_remove_step(
*
* This routine will find the blocks of the name to remove, remove them and
* shrink the tree if needed.
+ *
+ * This routine is meant to function as either an inline or delayed operation,
+ * and may return -EAGAIN when the transaction needs to be rolled. Calling
+ * functions will need to handle this, and recall the function until a
+ * successful error code is returned.
*/
STATIC int
-xfs_attr_node_removename(
- struct xfs_da_args *args)
+xfs_attr_node_removename_iter(
+ struct xfs_delattr_context *dac)
{
- struct xfs_da_state *state = NULL;
- int retval, error;
- struct xfs_inode *dp = args->dp;
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_da_state *state = NULL;
+ int retval, error;
+ struct xfs_inode *dp = args->dp;
trace_xfs_attr_node_removename(args);
- error = xfs_attr_node_removename_setup(args, &state);
- if (error)
- goto out;
-
- error = xfs_attr_node_remove_step(args, state);
- if (error)
- goto out;
-
- retval = xfs_attr_node_remove_cleanup(args, state);
-
- /*
- * Check to see if the tree needs to be collapsed.
- */
- if (retval && (state->path.active > 1)) {
- error = xfs_da3_join(state);
- if (error)
- goto out;
- error = xfs_defer_finish(&args->trans);
+ if (!dac->da_state) {
+ error = xfs_attr_node_removename_setup(dac);
if (error)
goto out;
+ }
+ state = dac->da_state;
+
+ switch (dac->dela_state) {
+ case XFS_DAS_UNINIT:
/*
- * Commit the Btree join operation and start a new trans.
+ * repeatedly remove remote blocks, remove the entry and join.
+ * returns -EAGAIN or 0 for completion of the step.
*/
- error = xfs_trans_roll_inode(&args->trans, dp);
+ error = xfs_attr_node_remove_step(dac);
if (error)
- goto out;
- }
+ break;
Hmm.. so re: my comment further down on xfs_attr_rmtval_remove(),
wouldn't that change semantics here? I.e., once remote blocks are
removed this would previously carry on with a clean transaction. Now it
looks like we'd carry on with the dirty transaction that removed the
last remote extent. This suggests that perhaps we should return once
more and fall into a new state to remove the name..?
I suspect the diff might be making this a bit difficult to see. The roll
that you see being removed here belongs to the transaction we hoisted up in
patch 3 which happens after the clean up below, and we have the
corresponding EAGAIN fot that one. I think the diff gets things a little
interlaced here because the switch adds another level of indentation.
Hmm.. the roll in patch 3 appears to be related to the _cleanup()
helper. What I'm referring to here is the state of the transaction after
the final remote block is removed from the attr. I'm not sure we're
talking about the same thing here..
some times i do like to I use a graphical diffviewer like diffuse when
patches get weird like this. Something like this:
git config --global diff.tool diffuse
git difftool 3c53e49 e201c09
You'd need to download the branch and also the diffuse tool, but sometimes i
think it makes some of these diffs a bit easier to see
I think it's easier just to refer to the code directly. The current
upstream code flows down into:
...
xfs_attr_node_removename()
xfs_attr_node_remove_rmt()
xfs_attr_rmtval_remove()
... which then implements the following loop:
do {
retval = __xfs_attr_rmtval_remove(args);
if (retval && retval != -EAGAIN)
return retval;
/*
* Close out trans and start the next one in the chain.
*/
error = xfs_trans_roll_inode(&args->trans, args->dp);
if (error)
return error;
} while (retval == -EAGAIN);
This rolls the transaction when retval == -EAGAIN or retval == 0, thus
always returns with a clean transaction after the remote block removal
completes.
The code as of this patch does:
...
xfs_attr_node_removename_iter()
xfs_attr_node_remove_step()
xfs_attr_node_remove_rmt()
__xfs_attr_rmtval_remove()
... which either returns -EAGAIN (since the roll is now implemented at
the very top) or 0 when done == true. The transaction might be dirty in
the latter case, but xfs_attr_node_removename_iter() moves right on to
xfs_attr_node_remove_cleanup() which can now do more work in that same
transaction. Am I following that correctly?
Also, it would be
nice to remove the several seemingly unnecessary layers of indirection
here. For example, something like the following (also considering my
comment above wrt to xfs_attr_remove_iter() and UNINIT):
case UNINIT:
...
/* fallthrough */
case RMTBLK:
if (args->rmtblkno > 0) {
dac->dela_state = RMTBLK;
error = __xfs_attr_rmtval_remove(dac);
if (error)
break;
ASSERT(args->rmtblkno == 0);
xfs_attr_refillstate(state);
dac->flags |= XFS_DAC_DEFER_FINISH;
dac->dela_state = RMNAME;
return -EAGAIN;
}
Ok, this looks to me like we've hoisted both xfs_attr_node_remove_rmt and
xfs_attr_node_remove_step into this scope, but I still think this adds an
extra roll where non previously was. With out that extra EAGAIN, I think we
are fine to have all that just under the UNINIT case. I also think it's
also worth noteing here that this is kind of a reverse of patch 1, which I
think we put in for reasons of trying to modularize the higher level
functions as much as possible.
I suspect some of where you were going with this may have been influenced by
the earlier diff confusion too. Maybe take a second look there before we go
too much down this change....
I can certainly be getting lost somewhere in all the refactoring. If so,
can you point out where in the flow described above?
Ok, I think see it. So basically I think this means we cant have the
helpers because it's ambiguos as to if the transaction is dirty or not. I
dont see that there's anything in the review history where we rationalized
that away, so I think we just overlooked it. So I think what this means is
that we need to reverse apply commit 72b97ea40d (which is where we added
xfs_attr_node_remove_rmt), then drop patch 1 which leaves no need for patch
3, since the transaction will have not moved. Then add state RMTBLK? I
think that arrives at what you have here.
It's not clear to me if anything needs to change before this patch or
the changes can just fold into this patch itself. You probably have a
better sense of that than I do atm.
I can unfold it here in this patch too if people prefer. I'm getting the
impression that others are having a hard time keeping up with the
refactoring, so maybe that might be better.
From my perspective, I think we want
that transaction to roll after the final remote extent removal unless we
had some reason to explicitly change existing behavior. This used to be
handled by the old loop that rolled the transaction down in the remote
block removal code. ISTM that the proper way to maintain the same
behavior in the new state machine code is to unconditionally fall out of
a RMTBLKREMOVE state with an -EAGAIN from _iter().
Sure, I think that describes what you have up there
IOW, __xfs_attr_rmtval_remove() returns -EAGAIN when it has more work to
do. _iter() returns -EAGAIN when __xfs_attr_rmtval_remove() was called,
finished its work, but we need to roll the transaction before the next
step of the operation..
Right, ok, will make those modifcations here then. Thanks for the
reviews! I know it's a lot.
Allison
Brian
Allison
Brian
/* fallthrough */
case RMNAME:
...
...
- /*
- * If the result is small enough, push it all into the inode.
- */
- if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
- error = xfs_attr_node_shrink(args, state);
+ retval = xfs_attr_node_remove_cleanup(args, state);
...
I think the overlooked EAGAIN was in this area that got clipped out.....
diff --git a/fs/xfs/libxfs/xfs_attr_remote.c b/fs/xfs/libxfs/xfs_attr_remote.c
index 48d8e9c..f09820c 100644
--- a/fs/xfs/libxfs/xfs_attr_remote.c
+++ b/fs/xfs/libxfs/xfs_attr_remote.c
...
@@ -685,31 +687,29 @@ c(
* Keep de-allocating extents until the remote-value region is gone.
*/
do {
- retval = __xfs_attr_rmtval_remove(args);
- if (retval && retval != -EAGAIN)
- return retval;
+ error = __xfs_attr_rmtval_remove(&dac);
+ if (error != -EAGAIN)
+ break;
Previously this would roll once and exit the loop on retval == 0. Now it
looks like we break out of the loop immediately. Why the change?
Gosh, I think sometime in reviewing v9, we had come up with a
"xfs_attr_roll_again" helper that took the error code as a paramater and
decided whether or not to roll. And then in v10 i think people thought that
was weird and we turned it into xfs_attr_trans_roll. I think I likley
forgot to restore the orginal retval handling here. This whole function
disappears in the next patch, but the original error handling should be
restored to keep things consistent. Thx for the catch!
Thx for the reviews!! I know it's complicated! I've chased my tail many
times with it myself :-)
Allison
Brian
- /*
- * Close out trans and start the next one in the chain.
- */
- error = xfs_trans_roll_inode(&args->trans, args->dp);
+ error = xfs_attr_trans_roll(&dac);
if (error)
return error;
- } while (retval == -EAGAIN);
+ } while (true);
- return 0;
+ return error;
}
/*
* Remove the value associated with an attribute by deleting the out-of-line
- * buffer that it is stored on. Returns EAGAIN for the caller to refresh the
+ * buffer that it is stored on. Returns -EAGAIN for the caller to refresh the
* transaction and re-call the function
*/
int
__xfs_attr_rmtval_remove(
- struct xfs_da_args *args)
+ struct xfs_delattr_context *dac)
{
- int error, done;
+ struct xfs_da_args *args = dac->da_args;
+ int error, done;
/*
* Unmap value blocks for this attr.
@@ -719,12 +719,20 @@ __xfs_attr_rmtval_remove(
if (error)
return error;
- error = xfs_defer_finish(&args->trans);
- if (error)
- return error;
-
- if (!done)
+ /*
+ * We dont need an explicit state here to pick up where we left off. We
+ * can figure it out using the !done return code. Calling function only
+ * needs to keep recalling this routine until we indicate to stop by
+ * returning anything other than -EAGAIN. The actual value of
+ * attr->xattri_dela_state may be some value reminicent of the calling
+ * function, but it's value is irrelevant with in the context of this
+ * function. Once we are done here, the next state is set as needed
+ * by the parent
+ */
+ if (!done) {
+ dac->flags |= XFS_DAC_DEFER_FINISH;
return -EAGAIN;
+ }
return error;
}
diff --git a/fs/xfs/libxfs/xfs_attr_remote.h b/fs/xfs/libxfs/xfs_attr_remote.h
index 9eee615..002fd30 100644
--- a/fs/xfs/libxfs/xfs_attr_remote.h
+++ b/fs/xfs/libxfs/xfs_attr_remote.h
@@ -14,5 +14,5 @@ int xfs_attr_rmtval_remove(struct xfs_da_args *args);
int xfs_attr_rmtval_stale(struct xfs_inode *ip, struct xfs_bmbt_irec *map,
xfs_buf_flags_t incore_flags);
int xfs_attr_rmtval_invalidate(struct xfs_da_args *args);
-int __xfs_attr_rmtval_remove(struct xfs_da_args *args);
+int __xfs_attr_rmtval_remove(struct xfs_delattr_context *dac);
#endif /* __XFS_ATTR_REMOTE_H__ */
diff --git a/fs/xfs/xfs_attr_inactive.c b/fs/xfs/xfs_attr_inactive.c
index bfad669..aaa7e66 100644
--- a/fs/xfs/xfs_attr_inactive.c
+++ b/fs/xfs/xfs_attr_inactive.c
@@ -15,10 +15,10 @@
#include "xfs_da_format.h"
#include "xfs_da_btree.h"
#include "xfs_inode.h"
+#include "xfs_attr.h"
#include "xfs_attr_remote.h"
#include "xfs_trans.h"
#include "xfs_bmap.h"
-#include "xfs_attr.h"
#include "xfs_attr_leaf.h"
#include "xfs_quota.h"
#include "xfs_dir2.h"
--
2.7.4
