On 4/5/21 6:17 AM, Brian Foster wrote:
Mkay, so generic/026 picked up a hang with this. It's because the condidtion that breaks this loop isnt args->rmtblkno > 0, it just error != -EAGAIN, and then falling through. I think had a comment somewhere in the last patch that confused people, I will update that.On Fri, Apr 02, 2021 at 02:01:20AM -0700, Allison Henderson wrote:On 4/1/21 9:57 AM, Brian Foster wrote:On Thu, Mar 25, 2021 at 05:33:08PM -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_find_space 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> Reviewed-by: Chandan Babu R <chandanrlinux@xxxxxxxxx> Reviewed-by: Darrick J. Wong <djwong@xxxxxxxxxx> --- fs/xfs/libxfs/xfs_attr.c | 446 ++++++++++++++++++++++++---------------- fs/xfs/libxfs/xfs_attr.h | 241 +++++++++++++++++++++- fs/xfs/libxfs/xfs_attr_remote.c | 100 ++++++--- fs/xfs/libxfs/xfs_attr_remote.h | 5 +- fs/xfs/xfs_trace.h | 1 - 5 files changed, 582 insertions(+), 211 deletions(-) diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c index 4a73691..6a86b62 100644 --- a/fs/xfs/libxfs/xfs_attr.c +++ b/fs/xfs/libxfs/xfs_attr.c...@@ -246,29 +247,55 @@ xfs_attr_trans_roll( return error; } +/* + * Set the attribute specified in @args. + */ +int +xfs_attr_set_args( + struct xfs_da_args *args) +{ + 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; + } while (true); + + return error; +} + STATIC int xfs_attr_set_fmt( - struct xfs_da_args *args) + struct xfs_delattr_context *dac, + struct xfs_buf **leaf_bp) { - struct xfs_buf *leaf_bp = NULL; - struct xfs_inode *dp = args->dp; - int error2, error = 0; + struct xfs_da_args *args = dac->da_args; + struct xfs_inode *dp = args->dp; + int error = 0; /* * Try to add the attr to the attribute list in the inode. */ error = xfs_attr_try_sf_addname(dp, args); - if (error != -ENOSPC) { - error2 = xfs_trans_commit(args->trans); - args->trans = NULL; - return error ? error : error2; - } + + /* Should only be 0, -EEXIST or -ENOSPC */ + if (error != -ENOSPC) + return error;Ok, so it looks like the commit that goes away here is replaced by one up the call stack./* * It won't fit in the shortform, transform to a leaf block. * GROT: another possible req'mt for a double-split btree op. */ - error = xfs_attr_shortform_to_leaf(args, &leaf_bp); + error = xfs_attr_shortform_to_leaf(args, leaf_bp); if (error) return error; @@ -277,95 +304,139 @@ xfs_attr_set_fmt( * concurrent AIL push cannot grab the half-baked leaf buffer * and run into problems 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);What about the xfs_trans_brelse() in the error case that seems to disappear?Hmm, maybe we could add some handler in xfs_attr_set_args such that if the roll fails, we release the leaf_bp if it is not null?I suspect we need it somewhere. I don't recall the details of this code off the top of my head, but I remember it being tricky so we should probably work to preserve current logic unless there's a clear reason to do otherwise.- return error; - } + xfs_trans_bhold(args->trans, *leaf_bp); + /* + * 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. + */ + dac->flags |= XFS_DAC_DEFER_FINISH; return -EAGAIN; } /* * 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. */ int -xfs_attr_set_args( - struct xfs_da_args *args) +xfs_attr_set_iter( + struct xfs_delattr_context *dac, + struct xfs_buf **leaf_bp) { - struct xfs_inode *dp = args->dp; - struct xfs_buf *bp = NULL; - struct xfs_da_state *state = NULL; - int forkoff, error = 0; + struct xfs_da_args *args = dac->da_args; + struct xfs_inode *dp = args->dp; + struct xfs_buf *bp = NULL; + struct xfs_da_state *state = NULL; + int forkoff, error = 0; - /* - * 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. - */ - if (xfs_attr_is_shortform(dp)) { - error = xfs_attr_set_fmt(args); - if (error != -EAGAIN) - return error; - } + /* State machine switch */ + switch (dac->dela_state) { + case XFS_DAS_UNINIT: + if (xfs_attr_is_shortform(dp)) + return xfs_attr_set_fmt(dac, leaf_bp); - if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) { - error = xfs_attr_leaf_try_add(args, bp); - if (error == -ENOSPC) { - /* - * Promote the attribute list to the Btree format. - */ - error = xfs_attr3_leaf_to_node(args); + /* + * 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. + */ + if (*leaf_bp != NULL) { + xfs_trans_bhold_release(args->trans, *leaf_bp); + *leaf_bp = NULL; + } + + if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) { + error = xfs_attr_leaf_try_add(args, *leaf_bp); + if (error == -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. + * + * At this point, we are still in + * XFS_DAS_UNINIT, but when we come back, we'll + * be a node, so we'll fall down into the node + * handling code below + */ + dac->flags |= XFS_DAC_DEFER_FINISH; + return -EAGAIN; + } + else if (error)Nit: } else if (error)Sure, will fix+ return error; + } + else {Nit: } else {will fix+ error = xfs_attr_node_addname_find_attr(dac); 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); + error = xfs_attr_node_addname(dac); if (error) return error;Ok, so these couple of node calls get peeled out of the loop that existed prior to this patch and xfs_attr_node_addname() returns -EAGAIN to trigger reentry, if necessary.Right, that is the idea/* - * Commit the current trans (including the inode) and - * start a new one. + * If addname was successful, and we dont need to alloc + * anymore blks, we're done. */ - error = xfs_trans_roll_inode(&args->trans, dp); - if (error) + if (!args->rmtblkno && !args->rmtblkno2) return error;Is this check new? What about clearing flags and whatnot?gosh, I remember putting this in, and now I'm struggling to remember the why that lead to it. It's needed later in the set, but at this point its sort of an optimization. In the xfs_attr_leaf_try_add, in the case of a rename, the blocks are saved for later processing. If no blocks were saved, there's nothing to flip, so the idea is we can stop here. The extra go around isnt a big deal at this point, but it's a problem in delayed attrs, because you end up with an extra empty log entry. in the extended set, we fail an assertion with out it: Assertion failed: !list_empty(&cil->xc_cil), file: fs/xfs/xfs_log_cil.c, line: 907 On my set up, we seem to get away without it at this point in the set, but it does have to go in eventually. I think I must have worked through this bug at one time, and then placed the fix in this patch, as it seems to be a requirement for becoming "delay ready"Hmm, that seems like an odd failure to associate with this code. I guess it's hard to grok because the context (i.e. delayed attrs) comes a bit later. If possible, it might be wise to defer this hunk until where it's necessary so the context/purpose is more clear on review (it also might be helpful to explain the purpose in a bit more detail in the comment).- goto node; - } - else if (error) { - return error; + dac->dela_state = XFS_DAS_FOUND_NBLK; + return -EAGAIN; } - /* - * 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; + dac->dela_state = XFS_DAS_FOUND_LBLK; + return -EAGAIN;Is there a reason the node branch sets the state in the branch but the leaf branch falls out to here? I.e., could we not be consistent and have if (one_block()) { ... dac->dela_state = XFS_DAS_FOUND_LBLK; } else { ... dac->dela_state = XFS_DAS_FOUND_NBLK; } return -EAGAIN;It's just sort of left over from its originally linear code flow that just sort of fell through in the the leaf logic. I think what you are proposing is logically analogous tho. Does your example mean to add an extra if/else at the end here? Or to tuck the return up into the the existing if/else? Both work, I am fine with either.Yes, the suggestion was intended to be logically equivalent. I think it improves readability and is slightly less fragile to make the duplicated code (i.e. return -EAGAIN) common and the state assignment as part of the associated branch. Brian+ case XFS_DAS_FOUND_LBLK: /* * If there was an out-of-line value, allocate the blocks we * identified for its storage and copy the value. This is done * 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; + + return -EAGAIN; } + 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 @@ -394,22 +465,26 @@ xfs_attr_set_args( * 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; + case XFS_DAS_FLIP_LFLAG: /* * Dismantle the "old" attribute/value pair by removing a * "remote" value (if it exists). */ xfs_attr_restore_rmt_blk(args); - if (args->rmtblkno) { - error = xfs_attr_rmtval_invalidate(args); - if (error) - return error; + error = xfs_attr_rmtval_invalidate(args); + if (error) + return error; + + /* Set state in case xfs_attr_rmtval_remove returns -EAGAIN */ + dac->dela_state = XFS_DAS_RM_LBLK; - error = xfs_attr_rmtval_remove(args); + /* fallthrough */ + case XFS_DAS_RM_LBLK: + if (args->rmtblkno) { + error = __xfs_attr_rmtval_remove(dac); if (error) return error; }This looks like a similar scenario as before where xfs_attr_rmtval_remove() would have returned with a clean transaction after the final unmap, but here __xfs_attr_rmtval_remove() just returns 0 if done == 1. We probably need to roll one more time out of this branch and land in a subsequent state..?I see, ok will add extra -EAGAIN here.@@ -434,91 +509,114 @@ xfs_attr_set_args( /* bp is gone due to xfs_da_shrink_inode */ return error; - } -node: + case XFS_DAS_FOUND_NBLK: + /* + * If there was an out-of-line value, allocate the blocks we + * identified for its storage and copy the value. This is done + * 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) { + /* + * Open coded xfs_attr_rmtval_set without trans + * handling + */ + error = xfs_attr_rmtval_find_space(dac); + if (error) + return error; - do { - error = xfs_attr_node_addname_find_attr(args, &state); - if (error) - return error; - error = xfs_attr_node_addname(args, state); - } while (error == -EAGAIN); - if (error) - return error; + /* + * Roll through the "value", allocating blocks on disk + * as required. Set the state in case of -EAGAIN return + * code + */ + dac->dela_state = XFS_DAS_ALLOC_NODE; + } - /* - * 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; + /* fallthrough */ + case XFS_DAS_ALLOC_NODE: + if (args->rmtblkno > 0) { + if (dac->blkcnt > 0) { + error = xfs_attr_rmtval_set_blk(dac); + if (error) + return error; - /* - * If there was an out-of-line value, allocate the blocks we - * identified for its storage and copy the value. This is done - * 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); - if (error) - return error; - } + return -EAGAIN; + } + + 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. + */ + if (args->rmtblkno > 0) + error = xfs_attr3_leaf_clearflag(args); + goto out; + } - if (!(args->op_flags & XFS_DA_OP_RENAME)) { /* - * Added a "remote" value, just clear the incomplete flag. + * If this is an atomic rename operation, we must "flip" the + * incomplete flags on the "new" and "old" attribute/value pairs + * so that one disappears and one appears atomically. Then we + * must remove the "old" attribute/value pair. + * + * In a separate transaction, set the incomplete flag on the + * "old" attr and clear the incomplete flag on the "new" attr. */ - if (args->rmtblkno > 0) - error = xfs_attr3_leaf_clearflag(args); - goto out; - } - - /* - * If this is an atomic rename operation, we must "flip" the incomplete - * flags on the "new" and "old" attribute/value pairs so that one - * disappears and one appears atomically. Then we must remove the "old" - * attribute/value pair. - * - * 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) - goto out; - /* - * 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; + error = xfs_attr3_leaf_flipflags(args); + if (error) + goto out; + /* + * Commit the flag value change and start the next trans in + * series + */ + dac->dela_state = XFS_DAS_FLIP_NFLAG; + return -EAGAIN; - /* - * Dismantle the "old" attribute/value pair by removing a "remote" value - * (if it exists). - */ - xfs_attr_restore_rmt_blk(args); + case XFS_DAS_FLIP_NFLAG: + /* + * Dismantle the "old" attribute/value pair by removing a + * "remote" value (if it exists). + */ + xfs_attr_restore_rmt_blk(args); - if (args->rmtblkno) { error = xfs_attr_rmtval_invalidate(args); if (error) return error; - error = xfs_attr_rmtval_remove(args); - if (error) - return error; - } + /* Set state in case xfs_attr_rmtval_remove returns -EAGAIN */ + dac->dela_state = XFS_DAS_RM_NBLK; + + /* fallthrough */ + case XFS_DAS_RM_NBLK: + if (args->rmtblkno) { + error = __xfs_attr_rmtval_remove(dac); + if (error) + return error; + }Similar thing here with __xfs_attr_rmtval_remove()..?Sure, will add one more -EAGAIN
In anycase, if we want another -EAGAIN, it means we have to stick in another state. And ditto for the other two calls to this function. Not a big deal, just an fyi so that it's not a suprise in the next revision.
Allison
+ + error = xfs_attr_node_addname_clear_incomplete(dac); - error = xfs_attr_node_addname_clear_incomplete(args); out: - if (state) - xfs_da_state_free(state); - return error; + if (state) + xfs_da_state_free(state); + return error;Can we avoid this out label landing inside the switch statement? That looks like a landmine. Even if we just duplicated an 'done_out' path after the last return in the function, I think that would be preferable.Sure, can do, that seems like a simple thing to tack onAll previous feedback aside, I think this patch now looks much more digestable in general. Most of the state code is isolated to the _iter() function and so it's much easier to follow along and compare against the current code flow. I did still have some thoughts with regard to further cleanups, possibly clearing up some the logic and/or tweaking the states and whatnot, but I think this is at a point where it might be reasonable to make such changes on top of this patch instead of continuing to make significant changes to it. If I get a chance perhaps I'll take a closer look at that once the remaining kinks are worked out.. BrianOk, I will get these last bits updated here. I still need to check into the issues Darrick is seeing on his set up, but it sounds like we've found an arrangement people like. And yes, I think switching to cleanups on top is a good next step. Thanks for the reviews! Allison+ + default: + ASSERT(dac->dela_state != XFS_DAS_RM_SHRINK); + break; + } + return error; } + /* * Return EEXIST if attr is found, or ENOATTR if not */ @@ -984,18 +1082,18 @@ xfs_attr_node_hasname( STATIC int xfs_attr_node_addname_find_attr( - struct xfs_da_args *args, - struct xfs_da_state **state) + struct xfs_delattr_context *dac) { - int retval; + struct xfs_da_args *args = dac->da_args; + int retval; /* * Search to see if name already exists, and get back a pointer * to where it should go. */ - retval = xfs_attr_node_hasname(args, state); + retval = xfs_attr_node_hasname(args, &dac->da_state); if (retval != -ENOATTR && retval != -EEXIST) - goto error; + return retval; if (retval == -ENOATTR && (args->attr_flags & XATTR_REPLACE)) goto error; @@ -1021,8 +1119,8 @@ xfs_attr_node_addname_find_attr( return 0; error: - if (*state) - xfs_da_state_free(*state); + if (dac->da_state) + xfs_da_state_free(dac->da_state); return retval; } @@ -1035,20 +1133,24 @@ xfs_attr_node_addname_find_attr( * * "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_da_state *state) + struct xfs_delattr_context *dac) { - struct xfs_da_state_blk *blk; - struct xfs_inode *dp; - int error; + struct xfs_da_args *args = dac->da_args; + struct xfs_da_state *state = dac->da_state; + struct xfs_da_state_blk *blk; + int error; trace_xfs_attr_node_addname(args); - dp = args->dp; - blk = &state->path.blk[state->path.active-1]; + blk = &state->path.blk[ state->path.active-1 ]; ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC); error = xfs_attr3_leaf_add(blk->bp, state->args); @@ -1064,18 +1166,15 @@ 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. + * 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. */ - error = xfs_trans_roll_inode(&args->trans, dp); - if (error) - goto out; - + dac->flags |= XFS_DAC_DEFER_FINISH; return -EAGAIN; } @@ -1088,9 +1187,7 @@ 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. @@ -1105,8 +1202,9 @@ xfs_attr_node_addname( STATIC int xfs_attr_node_addname_clear_incomplete( - struct xfs_da_args *args) + struct xfs_delattr_context *dac) { + struct xfs_da_args *args = dac->da_args; struct xfs_da_state *state = NULL; struct xfs_da_state_blk *blk; int retval = 0; diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h index 92a6a50..4e4233d 100644 --- a/fs/xfs/libxfs/xfs_attr.h +++ b/fs/xfs/libxfs/xfs_attr.h @@ -159,6 +159,233 @@ struct xfs_attr_list_context { * v * done * + * + * Below is a state machine diagram for attr set operations. + * + * It seems the challenge with understanding this system comes from trying to + * absorb the state machine all at once, when really one should only be looking + * at it with in the context of a single function. Once a state sensitive + * function is called, the idea is that it "takes ownership" of the + * state machine. It isn't concerned with the states that may have belonged to + * it's calling parent. Only the states relevant to itself or any other + * subroutines there in. Once a calling function hands off the state machine to + * a subroutine, it needs to respect the simple rule that it doesn't "own" the + * state machine anymore, and it's the responsibility of that calling function + * to propagate the -EAGAIN back up the call stack. Upon reentry, it is + * committed to re-calling that subroutine until it returns something other than + * -EAGAIN. Once that subroutine signals completion (by returning anything other + * than -EAGAIN), the calling function can resume using the state machine. + * + * xfs_attr_set_iter() + * │ + * v + * ┌─y─ has an attr fork? + * │ | + * │ n + * │ | + * │ V + * │ add a fork + * │ │ + * └──────────┤ + * │ + * V + * ┌─y─ is shortform? + * │ │ + * │ V + * │ xfs_attr_set_fmt + * │ | + * │ V + * │ xfs_attr_try_sf_addname + * │ │ + * │ V + * │ had enough ──y──> done + * │ space? + * n │ + * │ n + * │ │ + * │ V + * │ transform to leaf + * │ │ + * │ V + * │ hold the leaf buffer + * │ │ + * │ V + * │ return -EAGAIN + * │ Re-enter in + * │ leaf form + * │ + * └─> release leaf buffer + * if needed + * │ + * V + * ┌───n── fork has + * │ only 1 blk? + * │ │ + * │ y + * │ │ + * │ v + * │ xfs_attr_leaf_try_add() + * │ │ + * │ v + * │ had enough ──────────────y───────────────??? + * │ space? │ + * │ │ │ + * │ n │ + * │ │ │ + * │ v │ + * │ return -EAGAIN │ + * │ re-enter in │ + * │ node form │ + * │ │ │ + * └──────────┤ │ + * │ │ + * V │ + * xfs_attr_node_addname_find_attr │ + * determines if this │ + * is create or rename │ + * find space to store attr │ + * │ │ + * v │ + * xfs_attr_node_addname │ + * │ │ + * v │ + * fits in a node leaf? ────n─────??? │ + * │ ^ v │ + * │ │ single leaf node? │ + * │ │ │ │ │ + * y │ y n │ + * │ │ │ │ │ + * v │ v v │ + * update │ grow the leaf split if │ + * hashvals └─── return -EAGAIN needed │ + * │ retry leaf add │ │ + * │ on reentry │ │ + * ├─────────────────────────────┘ │ + * │ │ + * v │ + * need to alloc │ + * ┌─y── or flip flag? │ + * │ │ │ + * │ n │ + * │ │ │ + * │ v │ + * │ done │ + * │ │ + * │ │ + * │ XFS_DAS_FOUND_LBLK <──────────────────┘ + * │ │ + * │ V + * │ xfs_attr_leaf_addname() + * │ │ + * │ v + * │ ┌──first time through? + * │ │ │ + * │ │ y + * │ │ │ + * │ n v + * │ │ if we have rmt blks + * │ │ find space for them + * │ │ │ + * │ └──────────┤ + * │ │ + * │ v + * │ still have + * │ ┌─n─ blks to alloc? <──??? + * │ │ │ │ + * │ │ y │ + * │ │ │ │ + * │ │ v │ + * │ │ alloc one blk │ + * │ │ return -EAGAIN ──┘ + * │ │ re-enter with one + * │ │ less blk to alloc + * │ │ + * │ │ + * │ └───> set the rmt + * │ value + * │ │ + * │ v + * │ was this + * │ a rename? ──n─??? + * │ │ │ + * │ y │ + * │ │ │ + * │ v │ + * │ flip incomplete │ + * │ flag │ + * │ │ │ + * │ v │ + * │ XFS_DAS_FLIP_LFLAG │ + * │ │ │ + * │ v │ + * │ remove │ + * │ ┌───> old name │ + * │ │ │ │ + * │ XFS_DAS_RM_LBLK │ │ + * │ ^ │ │ + * │ │ v │ + * │ └──y── more to │ + * │ remove │ + * │ │ │ + * │ n │ + * │ │ │ + * │ v │ + * │ done <──────┘ + * │ + * └──────> XFS_DAS_FOUND_NBLK + * │ + * v + * ┌─────n── need to + * │ alloc blks? + * │ │ + * │ y + * │ │ + * │ v + * │ find space + * │ │ + * │ v + * │ ┌─>XFS_DAS_ALLOC_NODE + * │ │ │ + * │ │ v + * │ │ alloc blk + * │ │ │ + * │ │ v + * │ └──y── need to alloc + * │ more blocks? + * │ │ + * │ n + * │ │ + * │ v + * │ set the rmt value + * │ │ + * │ v + * │ was this + * └────────> a rename? ──n─??? + * │ │ + * y │ + * │ │ + * v │ + * flip incomplete │ + * flag │ + * │ │ + * v │ + * XFS_DAS_FLIP_NFLAG │ + * │ │ + * v │ + * remove │ + * ┌────────> old name │ + * │ │ │ + * XFS_DAS_RM_NBLK │ │ + * ^ │ │ + * │ v │ + * └──────y── more to │ + * remove │ + * │ │ + * n │ + * │ │ + * v │ + * done <──────┘ + * */ /* @@ -174,12 +401,20 @@ enum xfs_delattr_state { XFS_DAS_UNINIT = 0, /* No state has been set yet */ XFS_DAS_RMTBLK, /* Removing remote blks */ 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 */ }; /* * Defines for xfs_delattr_context.flags */ #define XFS_DAC_DEFER_FINISH 0x01 /* finish the transaction */ +#define XFS_DAC_LEAF_ADDNAME_INIT 0x02 /* xfs_attr_leaf_addname init*/ /* * Context used for keeping track of delayed attribute operations @@ -187,6 +422,11 @@ 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; @@ -213,7 +453,6 @@ 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 908521e7..fc71f10 100644 --- a/fs/xfs/libxfs/xfs_attr_remote.c +++ b/fs/xfs/libxfs/xfs_attr_remote.c @@ -439,9 +439,9 @@ xfs_attr_rmtval_get( /* * Find a "hole" in the attribute address space large enough for us to drop the - * new attribute's value into + * new attributes 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. */ @@ -669,37 +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..8ad68d5 100644 --- a/fs/xfs/libxfs/xfs_attr_remote.h +++ b/fs/xfs/libxfs/xfs_attr_remote.h @@ -10,9 +10,12 @@ int xfs_attr3_rmt_blocks(struct xfs_mount *mp, int attrlen); int xfs_attr_rmtval_get(struct xfs_da_args *args); int xfs_attr_rmtval_set(struct xfs_da_args *args); -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_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 e74bbb6..0c16d46 100644 --- a/fs/xfs/xfs_trace.h +++ b/fs/xfs/xfs_trace.h @@ -1944,7 +1944,6 @@ 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
