On 11/13/20 6:35 PM, Darrick J. Wong wrote:
On Thu, Nov 12, 2020 at 06:38:10PM -0700, Allison Henderson wrote:
On 11/10/20 4:10 PM, Darrick J. Wong wrote:
On Thu, Oct 22, 2020 at 11:34:28PM -0700, Allison Henderson wrote:
This patch modifies the attr set 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_set_args has become xfs_attr_set_iter, which uses a
state machine like switch to keep track of where it was when EAGAIN was
returned. See xfs_attr.h for a more detailed diagram of the states.
Two new helper functions have been added: xfs_attr_rmtval_set_init and
xfs_attr_rmtval_set_blk. They provide a subset of logic similar to
xfs_attr_rmtval_set, but they store the current block in the delay attr
context to allow the caller to roll the transaction between allocations.
This helps to simplify and consolidate code used by
xfs_attr_leaf_addname and xfs_attr_node_addname. xfs_attr_set_args has
now become a simple loop to refresh the transaction until the operation
is completed. Lastly, xfs_attr_rmtval_remove is no longer used, and is
removed.
Signed-off-by: Allison Henderson <allison.henderson@xxxxxxxxxx>
---
fs/xfs/libxfs/xfs_attr.c | 370 ++++++++++++++++++++++++++--------------
fs/xfs/libxfs/xfs_attr.h | 126 +++++++++++++-
fs/xfs/libxfs/xfs_attr_remote.c | 99 +++++++----
fs/xfs/libxfs/xfs_attr_remote.h | 4 +
fs/xfs/xfs_trace.h | 1 -
5 files changed, 439 insertions(+), 161 deletions(-)
diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
index 6ca94cb..95c98d7 100644
--- a/fs/xfs/libxfs/xfs_attr.c
+++ b/fs/xfs/libxfs/xfs_attr.c
@@ -44,7 +44,7 @@ STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
* Internal routines when attribute list is one block.
*/
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_addname(struct xfs_delattr_context *dac);
STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
STATIC int xfs_attr_leaf_hasname(struct xfs_da_args *args, struct xfs_buf **bp);
@@ -52,12 +52,15 @@ STATIC int xfs_attr_leaf_hasname(struct xfs_da_args *args, struct xfs_buf **bp);
* Internal routines when attribute list is more than one block.
*/
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_addname(struct xfs_delattr_context *dac);
STATIC int xfs_attr_node_removename_iter(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 int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
+STATIC int xfs_attr_set_iter(struct xfs_delattr_context *dac,
+ struct xfs_buf **leaf_bp);
int
xfs_inode_hasattr(
@@ -218,8 +221,11 @@ xfs_attr_is_shortform(
/*
* Attempts to set an attr in shortform, or converts short form to leaf form if
- * there is not enough room. If the attr is set, the transaction is committed
- * and set to NULL.
+ * there is not enough room. This function is meant to operate as a helper
+ * routine to the delayed attribute functions. It returns -EAGAIN to indicate
+ * that the calling function should roll the transaction, and then proceed to
+ * add the attr in leaf form. This subroutine does not expect to be recalled
+ * again like the other delayed attr routines do.
*/
STATIC int
xfs_attr_set_shortform(
@@ -227,16 +233,16 @@ xfs_attr_set_shortform(
struct xfs_buf **leaf_bp)
{
struct xfs_inode *dp = args->dp;
- int error, error2 = 0;
+ int error = 0;
/*
* Try to add the attr to the attribute list in the inode.
*/
error = xfs_attr_try_sf_addname(dp, args);
+
+ /* Should only be 0, -EEXIST or ENOSPC */
Nit: "...or -ENOSPC"
Also, this comment could go a couple of lines up:
Sure
/*
* Try to add the attr to the attribute list in the inode.
* This should only return 0, -EEXIST, or -ENOSPC.
*/
error = xfs_attr_try_sf_addname(dp, args);
if (error != -ENOSPC)
return error;
if (error != -ENOSPC) {
- error2 = xfs_trans_commit(args->trans);
- args->trans = NULL;
- return error ? error : error2;
+ return error;
}
/*
* It won't fit in the shortform, transform to a leaf block. GROT:
@@ -249,18 +255,10 @@ xfs_attr_set_shortform(
/*
* Prevent the leaf buffer from being unlocked so that a concurrent AIL
* push cannot grab the half-baked leaf buffer and run into problems
- * with the write verifier. Once we're done rolling the transaction we
- * can release the hold and add the attr to the leaf.
+ * with the write verifier.
*/
xfs_trans_bhold(args->trans, *leaf_bp);
- error = xfs_defer_finish(&args->trans);
- xfs_trans_bhold_release(args->trans, *leaf_bp);
- if (error) {
- xfs_trans_brelse(args->trans, *leaf_bp);
- return error;
- }
-
- return 0;
+ return -EAGAIN;
What state are we in when return -EAGAIN here? Are we still in
XFS_DAS_UNINIT, but with an attr fork that is no longer in local format,
which means that we skip the xfs_attr_is_shortform branch next time
around?
Yes, that's correct. I think I used to have an explicit state for it, but
it's really not needed for this reason. Though I think they do add some
degree of readability. Maybe we could add a comment?
/* Restart attr operation in leaf format */
?
Or even mention the DAS state explicitly, e.g.
/*
* We're still in XFS_DAS_UNINIT state here. We've converted the attr
* fork to leaf format and will restart with the leaf add.
*/
Hmm, second question: Could you add some tracepoints that would fire
every time we either change the DAS state or return -EAGAIN to trigger a
roll? I bet that will make debugging the attr code easier in the future.
Sure, I will look into adding some tracing here.
}
/*
@@ -268,7 +266,7 @@ xfs_attr_set_shortform(
* also checks for a defer finish. Transaction is finished and rolled as
* needed, and returns true of false if the delayed operation should continue.
*/
-int
+STATIC int
xfs_attr_trans_roll(
struct xfs_delattr_context *dac)
{
@@ -297,61 +295,130 @@ int
xfs_attr_set_args(
struct xfs_da_args *args)
{
- struct xfs_inode *dp = args->dp;
- struct xfs_buf *leaf_bp = NULL;
- int error = 0;
+ struct xfs_buf *leaf_bp = NULL;
+ int error = 0;
+ struct xfs_delattr_context dac = {
+ .da_args = args,
+ };
+
+ do {
+ error = xfs_attr_set_iter(&dac, &leaf_bp);
+ if (error != -EAGAIN)
+ break;
+
+ error = xfs_attr_trans_roll(&dac);
+ if (error)
+ return error;
+
+ if (leaf_bp) {
+ xfs_trans_bjoin(args->trans, leaf_bp);
+ xfs_trans_bhold(args->trans, leaf_bp);
+ }
+
+ } while (true);
+
+ return error;
+}
+
+/*
+ * Set the attribute specified in @args.
+ * This routine is meant to function as a 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_set_iter(
+ struct xfs_delattr_context *dac,
+ struct xfs_buf **leaf_bp)
+{
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_inode *dp = args->dp;
+ int error = 0;
+
+ /* State machine switch */
+ switch (dac->dela_state) {
+ case XFS_DAS_FLIP_LFLAG:
+ case XFS_DAS_FOUND_LBLK:
Do we need to catch XFS_DAS_RM_LBLK here?
I think we fall into the correct code path without it, but I think it's
better to have it here for consistency. Will add.
+ goto das_leaf;
+ case XFS_DAS_FOUND_NBLK:
+ case XFS_DAS_FLIP_NFLAG:
+ case XFS_DAS_ALLOC_NODE:
+ goto das_node;
+ default:
+ break;
+ }
/*
* If the attribute list is already in leaf format, jump straight to
* leaf handling. Otherwise, try to add the attribute to the shortform
* list; if there's no room then convert the list to leaf format and try
- * again.
+ * again. No need to set state as we will be in leaf form when we come
+ * back
*/
if (xfs_attr_is_shortform(dp)) {
/*
- * If the attr was successfully set in shortform, the
- * transaction is committed and set to NULL. Otherwise, is it
- * converted from shortform to leaf, and the transaction is
- * retained.
+ * If the attr was successfully set in shortform, no need to
+ * continue. Otherwise, is it converted from shortform to leaf
+ * and -EAGAIN is returned.
*/
- error = xfs_attr_set_shortform(args, &leaf_bp);
- if (error || !args->trans)
- return error;
+ error = xfs_attr_set_shortform(args, leaf_bp);
+ if (error == -EAGAIN)
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+
+ return error;
}
- if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
- error = xfs_attr_leaf_addname(args);
- if (error != -ENOSPC)
- return error;
+ /*
+ * After a shortform to leaf conversion, we need to hold the leaf and
+ * cycle out the transaction. When we get back, we need to release
+ * the leaf.
"...to release the hold on the leaf buffer."
Sure, will expand
+ */
+ if (*leaf_bp != NULL) {
+ xfs_trans_bhold_release(args->trans, *leaf_bp);
+ *leaf_bp = NULL;
+ }
- /*
- * Promote the attribute list to the Btree format.
- */
- error = xfs_attr3_leaf_to_node(args);
- if (error)
- return error;
+ if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
+ error = xfs_attr_leaf_try_add(args, *leaf_bp);
+ switch (error) {
+ case -ENOSPC:
+ /*
+ * Promote the attribute list to the Btree format.
+ */
+ error = xfs_attr3_leaf_to_node(args);
+ if (error)
+ return error;
- /*
- * Finish any deferred work items and roll the transaction once
- * more. The goal here is to call node_addname with the inode
- * and transaction in the same state (inode locked and joined,
- * transaction clean) no matter how we got to this step.
- */
- error = xfs_defer_finish(&args->trans);
- if (error)
+ /*
+ * Finish any deferred work items and roll the
+ * transaction once more. The goal here is to call
+ * node_addname with the inode and transaction in the
+ * same state (inode locked and joined, transaction
+ * clean) no matter how we got to this step.
+ */
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+ return -EAGAIN;
What state should we be in at this -EAGAIN return? Is it
XFS_DAS_UNINIT, but with more than one extent in the attr fork?
It could be UNINIT, if the attr was already a leaf at the time we started.
If we had to promote from a block to a leaf, and STILL counldnt fit in leaf
form, then we're probably in some state reminiscent of the leaf routines.
But because xfs_attr3_leaf_to_node just turned us into a node, we fall into
the node path upon return.
I know that's confusing... which leads to your next question of.....
/me is wishing these would get turned into explicit states, since afaict
we don't unlock the inode and so we should find it in /exactly/ the
state that the delattr_context says it should be in.
IIRC it used to have an explicit XFS_DC_LEAF_TO_NODE state, but I think we
simplified it away at some point in the reviewing in an effort to simplify
the statemachine as much as possible. v8 I think. But yes, I do think
there is a trade off between removing the states where they can be, but then
reducing the readability of where we are in the attr process. Because now
your state isnt exactly represented by dela_state anymore, it's the
combination of dela_state and the state of the tree.
I think I've been over this code so much by now, I can follow it either way,
but if it's confusing to others, maybe we should put it back? Or maybe just
a comment if that helps?
A comment laying out which states we could be in and how we got there. :)
Sure, will add in comments and traceing too. I'm thinking the comment
from below should probably come up here. Jump my next comment below...
+ case 0:
+ dac->dela_state = XFS_DAS_FOUND_LBLK;
+ return -EAGAIN;
+ default:
return error;
+ }
+das_leaf:
The only way to get to this block of code is by jumping to das_leaf,
from the switch statement above, right? If so, then shouldn't it be up
there in the switch statement?
We could, though I think we were just trying to be consistent in that the
switch is sort of a dispatcher for gotos? Otherwise we end up with a switch
with giant cases. It's the same difference I suppose.
With this comment in particular, I had to dig through the switch
statement in the previous code block to figure out that it's not
possible to fall into das_leaf from above.
+ error = xfs_attr_leaf_addname(dac);
+ if (error == -ENOSPC)
+ /*
+ * No need to set state. We will be in node form when
+ * we are recalled
+ */
+ return -EAGAIN;
How do we get to node form?
Hmm, I thought xfs_attr_leaf_addname did promote to node if theres not
enough space, but now that you point it out, i'm not seeing it. We may have
to put the LEAF_TO_NODE state back anyway.
maybe i can add a test case too, it doesnt look like any of the existing
cases run across it.
Hm. At least in the old code, I thought it was xfs_attr_set_args that
would call xfs_attr_leaf_addname and if it returned ENOSPC, it would
then call xfs_attr3_leaf_to_node...
(Gosh, I can't even tell where we are anymore. :()
Ok, sorry, so what happens is:
We pull xfs_attr_leaf_try_add out of xfs_attr_leaf_addname, and the
ENOSPC handler with it. But I think that's the only place in
xfs_attr_leaf_addname where ENOSPC used to come from. So I think this
old ENOSPC handler here can go away, I dont see that theres any other
way an ENOSPC comes back, or if it does its not expected. That's likely
why this ENOSPC handler isnt getting triggered in the test cases. The
ENOSPC condition is already getting handled in the case above.
So I think this line can just turn into
"return xfs_attr_leaf_addname(dac);"
Sorry for the condufsion!! :-(
Going back to your request for comments above, how about we expand the
existing coment up there and add:
/*
* We do not need to set a state here, because when we come back, we
* will be in node form. dela_state is still XFS_DAS_UNINIT at this time
*/
- /*
- * Commit the current trans (including the inode) and
- * start a new one.
- */
- error = xfs_trans_roll_inode(&args->trans, dp);
- if (error)
- return error;
+ return error;
}
-
- error = xfs_attr_node_addname(args);
+das_node:
+ error = xfs_attr_node_addname(dac);
return error;
Similarly, I think the only way get to this block of code is if we're in
the initial state (XFS_DAS_UNINIT?) and the inode wasn't in short
format; or if we jumped here via DAS_{FOUND_NBLK,FLIP_NFLAG,ALLOC_NODE},
right?
I think you could straighten this out a bit further (I left out the
comments):
switch (dac->dela_state) {
case XFS_DAS_FLIP_LFLAG:
case XFS_DAS_FOUND_LBLK:
error = xfs_attr_leaf_addname(dac);
if (error == -ENOSPC)
return -EAGAIN;
return error;
case XFS_DAS_FOUND_NBLK:
case XFS_DAS_FLIP_NFLAG:
case XFS_DAS_ALLOC_NODE:
return xfs_attr_node_addname(dac);
case XFS_DAS_UNINIT:
break;
default:
...assert on the XFS_DAS_RM_* flags...
}
if (xfs_attr_is_shortform(dp))
return xfs_attr_set_shortform(args, leaf_bp);
if (*leaf_bp != NULL) {
...release bhold...
}
if (!xfs_bmap_one_block(...))
return xfs_attr_node_addname(dac);
error = xfs_attr_leaf_try_add(args, *leaf_bp);
switch (error) {
...handle -ENOSPC and 0...
}
return error;
Ok, I'll see if I can get something like that through the test cases. If if
doesnt work out, I'll make a note of it.
<nod>
}
@@ -723,28 +790,30 @@ xfs_attr_leaf_try_add(
*
* This leaf block cannot have a "remote" value, we only call this routine
* if bmap_one_block() says there is only one block (ie: no remote blks).
+ *
+ * This routine is meant to function as a 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_leaf_addname(
- struct xfs_da_args *args)
+ struct xfs_delattr_context *dac)
{
- int error, forkoff;
- struct xfs_buf *bp = NULL;
- struct xfs_inode *dp = args->dp;
-
- trace_xfs_attr_leaf_addname(args);
-
- error = xfs_attr_leaf_try_add(args, bp);
- if (error)
- return error;
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_buf *bp = NULL;
+ int error, forkoff;
+ struct xfs_inode *dp = args->dp;
- /*
- * Commit the transaction that added the attr name so that
- * later routines can manage their own transactions.
- */
- error = xfs_trans_roll_inode(&args->trans, dp);
- if (error)
- return error;
+ /* State machine switch */
+ switch (dac->dela_state) {
+ case XFS_DAS_FLIP_LFLAG:
+ goto das_flip_flag;
+ case XFS_DAS_RM_LBLK:
+ goto das_rm_lblk;
+ default:
+ break;
+ }
/*
* If there was an out-of-line value, allocate the blocks we
@@ -752,12 +821,34 @@ xfs_attr_leaf_addname(
* after we create 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_rmtval_set(args);
+
+ /* Open coded xfs_attr_rmtval_set without trans handling */
+ if ((dac->flags & XFS_DAC_LEAF_ADDNAME_INIT) == 0) {
+ dac->flags |= XFS_DAC_LEAF_ADDNAME_INIT;
+ if (args->rmtblkno > 0) {
+ error = xfs_attr_rmtval_find_space(dac);
+ if (error)
+ return error;
+ }
+ }
+
+ /*
+ * Roll through the "value", allocating blocks on disk as
+ * required.
+ */
+ if (dac->blkcnt > 0) {
+ error = xfs_attr_rmtval_set_blk(dac);
if (error)
return error;
+
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+ return -EAGAIN;
What state are we in here? FOUND_LBLK, with blkcnt slowly decreasing?
I used to have an ALLOC_LEAF state for this one. Used to look something
like this:
+alloc_leaf:
Aha, that's where ALLOC_LEAF went.
+ while (args->dac.blkcnt > 0) {
+ error = xfs_attr_rmtval_set_blk(args);
+ if (error)
+ return error;
+
+ args->dac.flags |= XFS_DAC_FINISH_TRANS;
+ args->dac.dela_state = XFS_DAS_ALLOC_LEAF;
+ return -EAGAIN;
+ }
Again, it's not really needed, as we will fall into this logic with or with
out the state. And the while loop doesnt really loop, though I guess it
does sort of help the reader to understand that this is supposed to function
like a loop. I think it's easy to see something like that, and then want to
simplify away the extra semantics, but then on a second look, it's not quite
as obvious why with out the recollection of what it once was. Maybe a
comment is in order?
/* Repeat this until we have set all rmt blks */
?
Well there already is a comment that we're repeating until we've set all
the remote blocks, but it should capture which DAS state(s) we could be
in, because I quickly get lost, especially in the attr set code.
OK, how about:
/*
* We dont set a state because we can use dac->blkcnt to know if we need
* to continue setting blocks. dac.dela_state is still XFS_DAS_UNINIT
*/
To directly answer your question though, I think the state is still UNINIT
at this point, since any of the other states would have branched off before
this. It's important to note though that the functions that have states are
meant to sort of take ownership the statemachine. IOW, if the state coming
in does not apply to the scope of this function, or any of the subroutines
there in, then the state is simply overwritten as this function decides
appropriate. It doesnt throw an error if it is passed a state that used to
belong to it's parent. Calling functions should understand that they have
sort of "surrendered" the statemachine to this subfunction until it returns
something other than EAGAIN.
That will become very obvious once we've arrived at the end of the
series and everyone must use defer ops. :)
At least that's the idea. Honnestly, the only
reason I have UNINIT at all is because we get warnings about setting the
state to 0 when the enum needs to start at something other than 0.
Hope that helps?
Yeah.
}
+ error = xfs_attr_rmtval_set_value(args);
+ if (error)
+ return error;
+
if (!(args->op_flags & XFS_DA_OP_RENAME)) {
/*
* Added a "remote" value, just clear the incomplete flag.
@@ -777,29 +868,29 @@ xfs_attr_leaf_addname(
* In a separate transaction, set the incomplete flag on the "old" attr
* and clear the incomplete flag on the "new" attr.
*/
-
error = xfs_attr3_leaf_flipflags(args);
if (error)
return error;
/*
* Commit the flag value change and start the next trans in series.
*/
- error = xfs_trans_roll_inode(&args->trans, args->dp);
- if (error)
- return error;
-
+ dac->dela_state = XFS_DAS_FLIP_LFLAG;
+ return -EAGAIN;
+das_flip_flag:
/*
* Dismantle the "old" attribute/value pair by removing a "remote" value
* (if it exists).
*/
xfs_attr_restore_rmt_blk(args);
+ error = xfs_attr_rmtval_invalidate(args);
+ if (error)
+ return error;
+das_rm_lblk:
if (args->rmtblkno) {
- error = xfs_attr_rmtval_invalidate(args);
- if (error)
- return error;
-
- error = xfs_attr_rmtval_remove(args);
+ error = __xfs_attr_rmtval_remove(dac);
+ if (error == -EAGAIN)
+ dac->dela_state = XFS_DAS_RM_LBLK;
if (error)
return error;
}
@@ -965,23 +1056,38 @@ xfs_attr_node_hasname(
*
* "Remote" attribute values confuse the issue and atomic rename operations
* add a whole extra layer of confusion on top of that.
+ *
+ * This routine is meant to function as a 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_addname(
- struct xfs_da_args *args)
+ struct xfs_delattr_context *dac)
{
- struct xfs_da_state *state;
- struct xfs_da_state_blk *blk;
- struct xfs_inode *dp;
- int retval, error;
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_da_state *state = NULL;
+ struct xfs_da_state_blk *blk;
+ int retval = 0;
+ int error = 0;
trace_xfs_attr_node_addname(args);
- /*
- * Fill in bucket of arguments/results/context to carry around.
- */
- dp = args->dp;
-restart:
+ /* State machine switch */
+ switch (dac->dela_state) {
+ case XFS_DAS_FLIP_NFLAG:
+ goto das_flip_flag;
+ case XFS_DAS_FOUND_NBLK:
+ goto das_found_nblk;
+ case XFS_DAS_ALLOC_NODE:
+ goto das_alloc_node;
+ case XFS_DAS_RM_NBLK:
+ goto das_rm_nblk;
+ default:
+ break;
+ }
+
/*
* Search to see if name already exists, and get back a pointer
* to where it should go.
@@ -1027,19 +1133,13 @@ xfs_attr_node_addname(
error = xfs_attr3_leaf_to_node(args);
if (error)
goto out;
- error = xfs_defer_finish(&args->trans);
- if (error)
- goto out;
/*
- * Commit the node conversion and start the next
- * trans in the chain.
+ * Restart routine from the top. No need to set the
+ * state
*/
- error = xfs_trans_roll_inode(&args->trans, dp);
- if (error)
- goto out;
-
- goto restart;
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+ return -EAGAIN;
What state are we in here? Are we still in the same state that we were
at the start of the function, but ready to try xfs_attr3_leaf_add again?
To directly answer the question: we may be in UNINIT if we were already a
node when we started the attr op. If we had to promote from leaf to node,
it may be some state left over from the leaf routines.
Again though, in so far as this routine is concerned, the idea is that the
state either one of the cases in the switch up top, or it's not.
<nod> Comment please. :)
Sure, how about:
/*
* Now that we have converted the leaf to a node, we can roll the
* transaction, and try xfs_attr3_leaf_add again on re-entry. No need
* to set dela_state to do this. dela_state is still unset by this
* function at this point.
*/
}
/*
@@ -1051,9 +1151,7 @@ xfs_attr_node_addname(
error = xfs_da3_split(state);
if (error)
goto out;
- error = xfs_defer_finish(&args->trans);
- if (error)
- goto out;
+ dac->flags |= XFS_DAC_DEFER_FINISH;
} else {
/*
* Addition succeeded, update Btree hashvals.
@@ -1068,13 +1166,9 @@ xfs_attr_node_addname(
xfs_da_state_free(state);
state = NULL;
- /*
- * Commit the leaf addition or btree split and start the next
- * trans in the chain.
- */
- error = xfs_trans_roll_inode(&args->trans, dp);
- if (error)
- goto out;
+ dac->dela_state = XFS_DAS_FOUND_NBLK;
+ return -EAGAIN;
+das_found_nblk:
/*
* If there was an out-of-line value, allocate the blocks we
@@ -1083,7 +1177,27 @@ xfs_attr_node_addname(
* maximum size of a transaction and/or hit a deadlock.
*/
if (args->rmtblkno > 0) {
- error = xfs_attr_rmtval_set(args);
+ /* Open coded xfs_attr_rmtval_set without trans handling */
+ error = xfs_attr_rmtval_find_space(dac);
+ if (error)
+ return error;
+
+ /*
+ * Roll through the "value", allocating blocks on disk as
+ * required.
+ */
+das_alloc_node:
+ if (dac->blkcnt > 0) {
+ error = xfs_attr_rmtval_set_blk(dac);
+ if (error)
+ return error;
+
+ dac->flags |= XFS_DAC_DEFER_FINISH;
+ dac->dela_state = XFS_DAS_ALLOC_NODE;
+ return -EAGAIN;
+ }
+
+ error = xfs_attr_rmtval_set_value(args);
if (error)
return error;
}
@@ -1113,22 +1227,28 @@ xfs_attr_node_addname(
/*
* Commit the flag value change and start the next trans in series
*/
- error = xfs_trans_roll_inode(&args->trans, args->dp);
- if (error)
- goto out;
-
+ dac->dela_state = XFS_DAS_FLIP_NFLAG;
+ return -EAGAIN;
+das_flip_flag:
/*
* Dismantle the "old" attribute/value pair by removing a "remote" value
* (if it exists).
*/
xfs_attr_restore_rmt_blk(args);
+ error = xfs_attr_rmtval_invalidate(args);
+ if (error)
+ return error;
+
+das_rm_nblk:
if (args->rmtblkno) {
- error = xfs_attr_rmtval_invalidate(args);
- if (error)
- return error;
+ error = __xfs_attr_rmtval_remove(dac);
+
+ if (error == -EAGAIN) {
+ dac->dela_state = XFS_DAS_RM_NBLK;
+ return -EAGAIN;
+ }
- error = xfs_attr_rmtval_remove(args);
if (error)
return error;
}
diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
index 64dcf0f..501f9df 100644
--- a/fs/xfs/libxfs/xfs_attr.h
+++ b/fs/xfs/libxfs/xfs_attr.h
@@ -106,6 +106,118 @@ struct xfs_attr_list_context {
* v │
* done <─────┘
*
+ *
+ * Below is a state machine diagram for attr set operations.
+ *
+ * xfs_attr_set_iter()
+ * │
+ * v
I think this diagram is missing the part where we attempt to add a
shortform attr?
I left if out because the short form doesnt make use of states. I can
doodle that in though if you prefer:
┌───n── is shortform?
│ |
│ y
│ |
│ V
│ xfs_attr_set_shortform
│ |
│ V
├───n─── had enough
│ space?
│ │
│ y
│ │
│ V
│ done
└────────────┐
│
V
Yes, please do capture the entire mechanism so that 2025 us aren't
sitting here muttering about why we didn't do that when everything was
still warm in our L1 brain cache. ;)
--D
Alrighty, will do :-)
Thank you!!!
Allison
--D
Thx for the thorough reviews!
Allison
+ * ┌───n── fork has
+ * │ only 1 blk?
+ * │ │
+ * │ y
+ * │ │
+ * │ v
+ * │ xfs_attr_leaf_try_add()
+ * │ │
+ * │ v
+ * │ had enough
+ * ├───n────space?
+ * │ │
+ * │ y
+ * │ │
+ * │ v
+ * │ XFS_DAS_FOUND_LBLK ──┐
+ * │ │
+ * │ XFS_DAS_FLIP_LFLAG ──┤
+ * │ (subroutine state) │
+ * │ │
+ * │ └─>xfs_attr_leaf_addname()
+ * │ │
+ * │ v
+ * │ was this
+ * │ a rename? ──n─┐
+ * │ │ │
+ * │ y │
+ * │ │ │
+ * │ v │
+ * │ flip incomplete │
+ * │ flag │
+ * │ │ │
+ * │ v │
+ * │ XFS_DAS_FLIP_LFLAG │
+ * │ │ │
+ * │ v │
+ * │ remove │
+ * │ XFS_DAS_RM_LBLK ─> old name │
+ * │ ^ │ │
+ * │ │ v │
+ * │ └──────y── more to │
+ * │ remove │
+ * │ │ │
+ * │ n │
+ * │ │ │
+ * │ v │
+ * │ done <──────┘
+ * └──> XFS_DAS_FOUND_NBLK ──┐
+ * (subroutine state) │
+ * │
+ * XFS_DAS_ALLOC_NODE ──┤
+ * (subroutine state) │
+ * │
+ * XFS_DAS_FLIP_NFLAG ──┤
+ * (subroutine state) │
+ * │
+ * └─>xfs_attr_node_addname()
+ * │
+ * v
+ * find space to store
+ * attr. Split if needed
+ * │
+ * v
+ * XFS_DAS_FOUND_NBLK
+ * │
+ * v
+ * ┌─────n── need to
+ * │ alloc blks?
+ * │ │
+ * │ y
+ * │ │
+ * │ v
+ * │ ┌─>XFS_DAS_ALLOC_NODE
+ * │ │ │
+ * │ │ v
+ * │ └──y── need to alloc
+ * │ more blocks?
+ * │ │
+ * │ n
+ * │ │
+ * │ v
+ * │ was this
+ * └────────> a rename? ──n─┐
+ * │ │
+ * y │
+ * │ │
+ * v │
+ * flip incomplete │
+ * flag │
+ * │ │
+ * v │
+ * XFS_DAS_FLIP_NFLAG │
+ * │ │
+ * v │
+ * remove │
+ * XFS_DAS_RM_NBLK ─> old name │
+ * ^ │ │
+ * │ v │
+ * └──────y── more to │
+ * remove │
+ * │ │
+ * n │
+ * │ │
+ * v │
+ * done <──────┘
+ *
*/
/*
@@ -120,6 +232,13 @@ struct xfs_attr_list_context {
enum xfs_delattr_state {
XFS_DAS_UNINIT = 0, /* No state has been set yet */
XFS_DAS_RM_SHRINK, /* We are shrinking the tree */
+ XFS_DAS_FOUND_LBLK, /* We found leaf blk for attr */
+ XFS_DAS_FOUND_NBLK, /* We found node blk for attr */
+ XFS_DAS_FLIP_LFLAG, /* Flipped leaf INCOMPLETE attr flag */
+ XFS_DAS_RM_LBLK, /* A rename is removing leaf blocks */
+ XFS_DAS_ALLOC_NODE, /* We are allocating node blocks */
+ XFS_DAS_FLIP_NFLAG, /* Flipped node INCOMPLETE attr flag */
+ XFS_DAS_RM_NBLK, /* A rename is removing node blocks */
};
/*
@@ -127,6 +246,7 @@ enum xfs_delattr_state {
*/
#define XFS_DAC_DEFER_FINISH 0x01 /* finish the transaction */
#define XFS_DAC_NODE_RMVNAME_INIT 0x02 /* xfs_attr_node_removename init */
+#define XFS_DAC_LEAF_ADDNAME_INIT 0x04 /* xfs_attr_leaf_addname init*/
/*
* Context used for keeping track of delayed attribute operations
@@ -134,6 +254,11 @@ enum xfs_delattr_state {
struct xfs_delattr_context {
struct xfs_da_args *da_args;
+ /* Used in xfs_attr_rmtval_set_blk to roll through allocating blocks */
+ struct xfs_bmbt_irec map;
+ xfs_dablk_t lblkno;
+ int blkcnt;
+
/* Used in xfs_attr_node_removename to roll through removing blocks */
struct xfs_da_state *da_state;
@@ -160,7 +285,6 @@ 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);
-int xfs_attr_trans_roll(struct xfs_delattr_context *dac);
bool xfs_attr_namecheck(const void *name, size_t length);
void xfs_delattr_context_init(struct xfs_delattr_context *dac,
struct xfs_da_args *args);
diff --git a/fs/xfs/libxfs/xfs_attr_remote.c b/fs/xfs/libxfs/xfs_attr_remote.c
index 1426c15..5b445e7 100644
--- a/fs/xfs/libxfs/xfs_attr_remote.c
+++ b/fs/xfs/libxfs/xfs_attr_remote.c
@@ -441,7 +441,7 @@ xfs_attr_rmtval_get(
* Find a "hole" in the attribute address space large enough for us to drop the
* new attribute's value into
*/
-STATIC int
+int
xfs_attr_rmt_find_hole(
struct xfs_da_args *args)
{
@@ -468,7 +468,7 @@ xfs_attr_rmt_find_hole(
return 0;
}
-STATIC int
+int
xfs_attr_rmtval_set_value(
struct xfs_da_args *args)
{
@@ -628,6 +628,69 @@ xfs_attr_rmtval_set(
}
/*
+ * Find a hole for the attr and store it in the delayed attr context. This
+ * initializes the context to roll through allocating an attr extent for a
+ * delayed attr operation
+ */
+int
+xfs_attr_rmtval_find_space(
+ struct xfs_delattr_context *dac)
+{
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_bmbt_irec *map = &dac->map;
+ int error;
+
+ dac->lblkno = 0;
+ dac->blkcnt = 0;
+ args->rmtblkcnt = 0;
+ args->rmtblkno = 0;
+ memset(map, 0, sizeof(struct xfs_bmbt_irec));
+
+ error = xfs_attr_rmt_find_hole(args);
+ if (error)
+ return error;
+
+ dac->blkcnt = args->rmtblkcnt;
+ dac->lblkno = args->rmtblkno;
+
+ return 0;
+}
+
+/*
+ * Write one block of the value associated with an attribute into the
+ * out-of-line buffer that we have defined for it. This is similar to a subset
+ * of xfs_attr_rmtval_set, but records the current block to the delayed attr
+ * context, and leaves transaction handling to the caller.
+ */
+int
+xfs_attr_rmtval_set_blk(
+ struct xfs_delattr_context *dac)
+{
+ struct xfs_da_args *args = dac->da_args;
+ struct xfs_inode *dp = args->dp;
+ struct xfs_bmbt_irec *map = &dac->map;
+ int nmap;
+ int error;
+
+ nmap = 1;
+ error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)dac->lblkno,
+ dac->blkcnt, XFS_BMAPI_ATTRFORK, args->total,
+ map, &nmap);
+ if (error)
+ return error;
+
+ ASSERT(nmap == 1);
+ ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
+ (map->br_startblock != HOLESTARTBLOCK));
+
+ /* roll attribute extent map forwards */
+ dac->lblkno += map->br_blockcount;
+ dac->blkcnt -= map->br_blockcount;
+
+ return 0;
+}
+
+/*
* Remove the value associated with an attribute by deleting the
* out-of-line buffer that it is stored on.
*/
@@ -669,38 +732,6 @@ xfs_attr_rmtval_invalidate(
}
/*
- * Remove the value associated with an attribute by deleting the
- * out-of-line buffer that it is stored on.
- */
-int
-xfs_attr_rmtval_remove(
- struct xfs_da_args *args)
-{
- int error;
- struct xfs_delattr_context dac = {
- .da_args = args,
- };
-
- trace_xfs_attr_rmtval_remove(args);
-
- /*
- * Keep de-allocating extents until the remote-value region is gone.
- */
- do {
- error = __xfs_attr_rmtval_remove(&dac);
- if (error != -EAGAIN)
- break;
-
- error = xfs_attr_trans_roll(&dac);
- if (error)
- return error;
-
- } while (true);
-
- 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
* transaction and re-call the function
diff --git a/fs/xfs/libxfs/xfs_attr_remote.h b/fs/xfs/libxfs/xfs_attr_remote.h
index 002fd30..84e2700 100644
--- a/fs/xfs/libxfs/xfs_attr_remote.h
+++ b/fs/xfs/libxfs/xfs_attr_remote.h
@@ -15,4 +15,8 @@ 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_delattr_context *dac);
+int xfs_attr_rmt_find_hole(struct xfs_da_args *args);
+int xfs_attr_rmtval_set_value(struct xfs_da_args *args);
+int xfs_attr_rmtval_set_blk(struct xfs_delattr_context *dac);
+int xfs_attr_rmtval_find_space(struct xfs_delattr_context *dac);
#endif /* __XFS_ATTR_REMOTE_H__ */
diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
index 8695165..e9dde4e 100644
--- a/fs/xfs/xfs_trace.h
+++ b/fs/xfs/xfs_trace.h
@@ -1925,7 +1925,6 @@ DEFINE_ATTR_EVENT(xfs_attr_refillstate);
DEFINE_ATTR_EVENT(xfs_attr_rmtval_get);
DEFINE_ATTR_EVENT(xfs_attr_rmtval_set);
-DEFINE_ATTR_EVENT(xfs_attr_rmtval_remove);
#define DEFINE_DA_EVENT(name) \
DEFINE_EVENT(xfs_da_class, name, \
--
2.7.4
