On 4/15/20 4:46 AM, Brian Foster wrote:
On Tue, Apr 14, 2020 at 02:35:43PM -0700, Allison Collins wrote:
On 4/13/20 5:30 AM, Brian Foster wrote:
On Fri, Apr 03, 2020 at 03:12:27PM -0700, Allison Collins 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.
Calls to xfs_attr_rmtval_remove are replaced with the delay ready
counter parts: xfs_attr_rmtval_invalidate (appearing in the setup
helper) and then __xfs_attr_rmtval_remove. We will rename
__xfs_attr_rmtval_remove back to xfs_attr_rmtval_remove when we are
done.
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.
Below is a state machine diagram for attr remove operations. The
XFS_DAS_* states indicate places where the function would return
-EAGAIN, and then immediately resume from after being recalled by the
calling function. States marked as a "subroutine state" indicate that
they belong to a subroutine, and so the calling function needs to pass
them back to that subroutine to allow it to finish where it left off.
But they otherwise do not have a role in the calling function other than
just passing through.
xfs_attr_remove_iter()
XFS_DAS_RM_SHRINK ─┐
(subroutine state) │
│
XFS_DAS_RMTVAL_REMOVE ─┤
(subroutine state) │
└─>xfs_attr_node_removename()
│
v
need to remove
┌─n── rmt blocks?
│ │
│ y
│ │
│ v
│ ┌─>XFS_DAS_RMTVAL_REMOVE
│ │ │
│ │ v
│ └──y── more blks
│ to remove?
│ │
│ n
│ │
│ v
│ need to
└─────> shrink tree? ─n─┐
│ │
y │
│ │
v │
XFS_DAS_RM_SHRINK │
│ │
v │
done <─────┘
Signed-off-by: Allison Collins <allison.henderson@xxxxxxxxxx>
---
All in all this is starting to look much more simple to me, at least in
the remove path. ;P There's only a few states and the markers that are
introduced are fairly straightforward, etc. Comments to follow..
fs/xfs/libxfs/xfs_attr.c | 168 ++++++++++++++++++++++++++++++++++++-----------
fs/xfs/libxfs/xfs_attr.h | 38 +++++++++++
2 files changed, 168 insertions(+), 38 deletions(-)
diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
index d735570..f700976 100644
--- a/fs/xfs/libxfs/xfs_attr.c
+++ b/fs/xfs/libxfs/xfs_attr.c
@@ -45,7 +45,7 @@ STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
*/
STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
-STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
+STATIC int xfs_attr_leaf_removename(struct xfs_delattr_context *dac);
STATIC int xfs_attr_leaf_hasname(struct xfs_da_args *args, struct xfs_buf **bp);
/*
@@ -53,12 +53,21 @@ STATIC int xfs_attr_leaf_hasname(struct xfs_da_args *args, struct xfs_buf **bp);
*/
STATIC int xfs_attr_node_get(xfs_da_args_t *args);
STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
-STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
+STATIC int xfs_attr_node_removename(struct xfs_delattr_context *dac);
STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
struct xfs_da_state **state);
STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
+STATIC void
+xfs_delattr_context_init(
+ struct xfs_delattr_context *dac,
+ struct xfs_da_args *args)
+{
+ memset(dac, 0, sizeof(struct xfs_delattr_context));
+ dac->da_args = args;
+}
+
int
xfs_inode_hasattr(
struct xfs_inode *ip)
@@ -356,20 +365,66 @@ xfs_has_attr(
*/
int
xfs_attr_remove_args(
- struct xfs_da_args *args)
+ struct xfs_da_args *args)
{
+ int error = 0;
+ struct xfs_delattr_context dac;
+
+ xfs_delattr_context_init(&dac, args);
+
+ do {
+ error = xfs_attr_remove_iter(&dac);
+ if (error != -EAGAIN)
+ break;
+
+ if (dac.flags & XFS_DAC_DEFER_FINISH) {
+ dac.flags &= ~XFS_DAC_DEFER_FINISH;
+ error = xfs_defer_finish(&args->trans);
+ if (error)
+ break;
+ }
+
+ error = xfs_trans_roll_inode(&args->trans, args->dp);
+ if (error)
+ break;
+ } while (true);
+
+ return error;
+}
+
+/*
+ * Remove the attribute specified in @args.
+ *
+ * This function may return -EAGAIN to signal that the transaction needs to be
+ * rolled. Callers should continue calling this function until they receive a
+ * return value other than -EAGAIN.
+ */
+int
+xfs_attr_remove_iter(
+ struct xfs_delattr_context *dac)
+{
+ struct xfs_da_args *args = dac->da_args;
struct xfs_inode *dp = args->dp;
int error;
+ /* State machine switch */
+ switch (dac->dela_state) {
+ case XFS_DAS_RM_SHRINK:
+ case XFS_DAS_RMTVAL_REMOVE:
+ return xfs_attr_node_removename(dac);
+ default:
+ break;
+ }
+
Hmm.. so we're duplicating the call instead of using labels..?
Yes, this was a suggestion made during v7. I suspect Dave may have been
wanting to simplify things by escaping the use of labels. At least in so
far as the remove path is concerned. Though he may not have realized this
would create a duplication call? I will cc him here; the conditions for
calling xfs_attr_node_removename are: the below if/else sequence exhausts
with no successes, and defaults into the else case (ie: the entry
condition), OR one of the above states is set (which is a re-entry
condition)
Ok.
I'm
wondering if this can be elegantly combined with the if/else branches
below, particularly since node format is the only situation that seems
to require a roll here.
if (!xfs_inode_hasattr(dp)) {
error = -ENOATTR;
} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
error = xfs_attr_shortform_remove(args);
} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
- error = xfs_attr_leaf_removename(args);
+ error = xfs_attr_leaf_removename(dac);
} else {
- error = xfs_attr_node_removename(args);
+ error = xfs_attr_node_removename(dac);
}
return error;
If we want to try and combine this into if/elses with no duplication, I
believe the simplest arrangement would look something like this:
int
xfs_attr_remove_iter(
struct xfs_delattr_context *dac)
{
struct xfs_da_args *args = dac->da_args;
struct xfs_inode *dp = args->dp;
if (dac->dela_state != XFS_DAS_RM_SHRINK &&
dac->dela_state != XFS_DAS_RMTVAL_REMOVE) {
if (!xfs_inode_hasattr(dp)) {
return -ENOATTR;
} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
return xfs_attr_shortform_remove(args);
} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
return xfs_attr_leaf_removename(dac);
}
}
return xfs_attr_node_removename(dac);
}
Let me know what folks think of that. I'm not always clear on where people
stand with aesthetics. (IE, is it better to have a duplicate call if it gets
rid of a label? Is the solution with the least amount of LOC always
preferable?) This area seems simple enough maybe we can get it ironed out
here with out another version.
IMHO I think the above code sort of obfuscates that the code flow is really
just one if/else switch with one function that has the statemachine
behavior. But its not bad either if that's what people prefer. I'd like to
find something every can be sort of happy with. :-)
If you want my .02, some combination of the above is cleanest from an
aesthetic pov:
{
...
if (RM_SHRINK || RMTVAL_REMOVE)
goto node;
if (!hasattr)
return -ENOATTR;
else if (local)
return shortform_remove();
else if (oneblock)
return leaf_removename();
node:
return node_removename();
}
I find that easiest to read at a glance, but I don't feel terribly
strongly about it I guess.
I am entirely fine with that if everyone else is :-)
@@ -794,11 +849,12 @@ xfs_attr_leaf_hasname(
*/
STATIC int
xfs_attr_leaf_removename(
- struct xfs_da_args *args)
+ struct xfs_delattr_context *dac)
{
- struct xfs_inode *dp;
- struct xfs_buf *bp;
- int error, forkoff;
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_inode *dp;
+ struct xfs_buf *bp;
+ int error, forkoff;
trace_xfs_attr_leaf_removename(args);
@@ -825,9 +881,8 @@ xfs_attr_leaf_removename(
/* bp is gone due to xfs_da_shrink_inode */
if (error)
return error;
- error = xfs_defer_finish(&args->trans);
- if (error)
- return error;
+
+ dac->flags |= XFS_DAC_DEFER_FINISH;
There's no -EAGAIN return here, is this an exit path for the remove?
I think so, maybe I can remove this and the other one you pointed out in
patch 12 along with the other unneeded transaction handling.
}
return 0;
}
@@ -1128,12 +1183,13 @@ xfs_attr_node_addname(
*/
STATIC int
xfs_attr_node_shrink(
- struct xfs_da_args *args,
- struct xfs_da_state *state)
+ struct xfs_delattr_context *dac,
+ struct xfs_da_state *state)
{
- struct xfs_inode *dp = args->dp;
- int error, forkoff;
- struct xfs_buf *bp;
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_inode *dp = args->dp;
+ int error, forkoff;
+ struct xfs_buf *bp;
/*
* Have to get rid of the copy of this dabuf in the state.
@@ -1153,9 +1209,7 @@ xfs_attr_node_shrink(
if (error)
return error;
- error = xfs_defer_finish(&args->trans);
- if (error)
- return error;
+ dac->flags |= XFS_DAC_DEFER_FINISH;
Same question here.
} else
xfs_trans_brelse(args->trans, bp);
@@ -1194,13 +1248,15 @@ xfs_attr_leaf_mark_incomplete(
/*
* Initial setup for xfs_attr_node_removename. Make sure the attr is there and
- * the blocks are valid. Any remote blocks will be marked incomplete.
+ * the blocks are valid. Any remote blocks will be marked incomplete and
+ * invalidated.
*/
STATIC
int xfs_attr_node_removename_setup(
- struct xfs_da_args *args,
- struct xfs_da_state **state)
+ struct xfs_delattr_context *dac,
+ struct xfs_da_state **state)
{
+ struct xfs_da_args *args = dac->da_args;
int error;
struct xfs_da_state_blk *blk;
@@ -1212,10 +1268,21 @@ int xfs_attr_node_removename_setup(
ASSERT(blk->bp != NULL);
ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
+ /*
+ * Store blk and state in the context incase we need to cycle out the
+ * transaction
+ */
+ dac->blk = blk;
+ dac->da_state = *state;
+
if (args->rmtblkno > 0) {
error = xfs_attr_leaf_mark_incomplete(args, *state);
if (error)
return error;
+
+ error = xfs_attr_rmtval_invalidate(args);
+ if (error)
+ return error;
Seems like this moves code, which should probably happen in a separate
patch.
Ok, this pairs with the xfs_attr_rmtval_remove to __xfs_attr_rmtval_remove
below. Basically xfs_attr_rmtval_remove is the combination of
xfs_attr_rmtval_invalidate and __xfs_attr_rmtval_remove. So thats why we see
xfs_attr_rmtval_remove going away and xfs_attr_rmtval_invalidate +
__xfs_attr_rmtval_remove coming in.
How about a patch that pulls xfs_attr_rmtval_invalidate out of
xfs_attr_rmtval_remove and into the calling functions? I think that might
be more clear.
Yes, separate patch please. I think that if the earlier refactoring
parts of the series are split out properly (i.e., no dependencies on
subsequent patches) and reviewed, perhaps we can start getting some of
those patches merged while the latter bits are worked out.
Ok then, will do
}
return 0;
@@ -1228,7 +1295,10 @@ xfs_attr_node_removename_rmt (
{
int error = 0;
- error = xfs_attr_rmtval_remove(args);
+ /*
+ * May return -EAGAIN to request that the caller recall this function
+ */
+ error = __xfs_attr_rmtval_remove(args);
if (error)
return error;
@@ -1249,19 +1319,37 @@ xfs_attr_node_removename_rmt (
* This will involve walking down the Btree, and may involve joining
* leaf nodes and even joining intermediate nodes up to and including
* the root node (a special case of an intermediate node).
+ *
+ * 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)
+ struct xfs_delattr_context *dac)
{
+ struct xfs_da_args *args = dac->da_args;
struct xfs_da_state *state;
struct xfs_da_state_blk *blk;
int retval, error;
struct xfs_inode *dp = args->dp;
trace_xfs_attr_node_removename(args);
+ state = dac->da_state;
+ blk = dac->blk;
+
+ /* State machine switch */
+ switch (dac->dela_state) {
+ case XFS_DAS_RMTVAL_REMOVE:
+ goto das_rmtval_remove;
+ case XFS_DAS_RM_SHRINK:
+ goto das_rm_shrink;
+ default:
+ break;
+ }
- error = xfs_attr_node_removename_setup(args, &state);
+ error = xfs_attr_node_removename_setup(dac, &state);
if (error)
goto out;
@@ -1270,10 +1358,16 @@ xfs_attr_node_removename(
* This is done before we remove the attribute so that we don't
* overflow the maximum size of a transaction and/or hit a deadlock.
*/
+
+das_rmtval_remove:
+
I wonder if we need this label just to protect the setup. Perhaps if we
had something like:
/* set up the remove only once... */
if (dela_state == 0)
error = xfs_attr_node_removename_setup(...);
... we could reduce another state.
We could also accomplish the same thing with an explicit state to
indicate the setup already occurred or a new dac flag, though I'm not
sure a flag is appropriate if it would only be used here.
Brian
Mmmm, dela_state == 0 will conflict a bit when we get into fully delayed
attrs. Basically when this is getting called from the delayed operations
path, it sets dela_state to a new XFS_DAS_INIT. Because we have to set up
args mid fight, we need the extra state to not do that twice.
Can we address that when the conflict is introduced?
Sure, i was just looking ahead, but focus should stay here
But even without getting into that right away, what you're proposing only
gets rid of the label. It doesnt get rid of the state. We still have to
set the state to not be zero (or what ever the initial value is). So we
still need the unique value of XFS_DAS_RMTVAL_REMOVE
Yeah, I was partly thinking of the setup call being tied to a flag
rather than a state. That way the logic is something like the typical:
if (!setup)
do_setup();
...
... and it's one less bit of code tied into the state machine. All in
all, it's more that having a label right at the top of a function like
that kind of looks like it's asking for some form of simplification.
Ok, this is similar to the discussion going on in the next patch. To be
clear, if we add a flag, we need to keep it in the delay_attr_context so
that it persists across re-calls. We dont want multiple functions
shareing the same setup flag, so really we're talking about a new flag
scheme like XFS_DAC_SETUP_* for any function that needs a set up flag?
Is that what you had in mind?
Really what you would need here in order to do what you are describeing is
dela_state != XFS_DAS_RMTVAL_REMOVE. If I assume to simplify away to the
lease amount of LOC we get this:
STATIC int
xfs_attr_node_removename(
struct xfs_delattr_context *dac)
{
struct xfs_da_args *args = dac->da_args;
struct xfs_da_state *state;
struct xfs_da_state_blk *blk;
int retval, error;
struct xfs_inode *dp = args->dp;
trace_xfs_attr_node_removename(args);
state = dac->da_state;
blk = dac->blk;
if (dac->dela_state == XFS_DAS_RM_SHRINK) {
goto das_rm_shrink;
} else if (dac->dela_state != XFS_DAS_RMTVAL_REMOVE) {
error = xfs_attr_node_removename_setup(dac, &state);
if (error)
goto out;
}
/*
* If there is an out-of-line value, de-allocate the blocks.
* This is done before we remove the attribute so that we don't
* overflow the maximum size of a transaction and/or hit a deadlock.
*/
if (args->rmtblkno > 0) {
error = xfs_attr_node_removename_rmt(args, state);
if (error) {
if (error == -EAGAIN)
dac->dela_state = XFS_DAS_RMTVAL_REMOVE;
return error;
}
}
.....
Let me know what folks think of this. Again, I think I like the switches
and the labels just because it makes it more clear where the jump points
are, even if its more LOC. But again, this isnt bad either if this is more
preferable to folks. If there's another arrangment that is preferable, let
me know, it's not difficult to run it through the test cases to make sure
it's functional. It may be a faster way to hash out what people want to
see.
I prefer to see the state management stuff as boilerplate as possible.
The above pattern of creating separate reentry calls to the same
functions is not nearly as clear to me, particularly in this instance
where we have multiple branches of reentry logic (as opposed to the
earlier example of only one).
IOW, I agree that the jumps are preferable and more intuitive. I'm just
trying to be reductive by considering what could be factored out vs.
trying to fundamentally rework the approach or aggressively reduce LOC
or anything like that. IMO, simplicity of the code is usually top
priority.
Brian
Ok, lets firm up what we want this setup flag to look like, and then we
can simplify the current label and goto scheme :-)
Thank you!!
Allison
Thank you again for all the reviewing!!!
Allison
if (args->rmtblkno > 0) {
error = xfs_attr_node_removename_rmt(args, state);
- if (error)
- goto out;
+ if (error) {
+ if (error == -EAGAIN)
+ dac->dela_state = XFS_DAS_RMTVAL_REMOVE;
+ return error;
+ }
}
/*
@@ -1291,22 +1385,20 @@ xfs_attr_node_removename(
error = xfs_da3_join(state);
if (error)
goto out;
- error = xfs_defer_finish(&args->trans);
- if (error)
- goto out;
- /*
- * Commit the Btree join operation and start a new trans.
- */
- error = xfs_trans_roll_inode(&args->trans, dp);
- if (error)
- goto out;
+
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+ dac->dela_state = XFS_DAS_RM_SHRINK;
+ return -EAGAIN;
}
+das_rm_shrink:
+ dac->dela_state = XFS_DAS_RM_SHRINK;
+
/*
* 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);
+ error = xfs_attr_node_shrink(dac, state);
error = 0;
out:
diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
index 66575b8..0e8ae1a 100644
--- a/fs/xfs/libxfs/xfs_attr.h
+++ b/fs/xfs/libxfs/xfs_attr.h
@@ -74,6 +74,43 @@ struct xfs_attr_list_context {
};
+/*
+ * ========================================================================
+ * Structure used to pass context around among the delayed routines.
+ * ========================================================================
+ */
+
+/*
+ * Enum values for xfs_delattr_context.da_state
+ *
+ * These values are used by delayed attribute operations to keep track of where
+ * they were before they returned -EAGAIN. A return code of -EAGAIN signals the
+ * calling function to roll the transaction, and then recall the subroutine to
+ * finish the operation. The enum is then used by the subroutine to jump back
+ * to where it was and resume executing where it left off.
+ */
+enum xfs_delattr_state {
+ /* Zero is uninitalized */
+ XFS_DAS_RM_SHRINK = 1, /* We are shrinking the tree */
+ XFS_DAS_RMTVAL_REMOVE, /* We are removing remote value blocks */
+};
+
+/*
+ * Defines for xfs_delattr_context.flags
+ */
+#define XFS_DAC_DEFER_FINISH 0x1 /* indicates to finish the transaction */
+
+/*
+ * Context used for keeping track of delayed attribute operations
+ */
+struct xfs_delattr_context {
+ struct xfs_da_args *da_args;
+ struct xfs_da_state *da_state;
+ struct xfs_da_state_blk *blk;
+ unsigned int flags;
+ enum xfs_delattr_state dela_state;
+};
+
/*========================================================================
* Function prototypes for the kernel.
*========================================================================*/
@@ -91,6 +128,7 @@ int xfs_attr_set(struct xfs_da_args *args);
int xfs_attr_set_args(struct xfs_da_args *args);
int xfs_has_attr(struct xfs_da_args *args);
int xfs_attr_remove_args(struct xfs_da_args *args);
+int xfs_attr_remove_iter(struct xfs_delattr_context *dac);
bool xfs_attr_namecheck(const void *name, size_t length);
#endif /* __XFS_ATTR_H__ */
--
2.7.4
