dax: Fix missed wakeup with PMD faults

RocksDB can hang indefinitely when using a DAX file.  This is due to
a bug in the XArray conversion when handling a PMD fault and finding a
PTE entry.  We use the wrong index in the hash and end up waiting on
the wrong waitqueue.

There's actually no need to wait; if we find a PTE entry while looking
for a PMD entry, we can return immediately as we know we should fall
back to a PTE fault (which may not conflict with the lock held).

We reuse the XA_RETRY_ENTRY to signal a conflicting entry was found.
This value can never be found in an XArray while holding its lock, so
it does not create an ambiguity.

Cc: <[email protected]>
Link: http://lkml.kernel.org/r/CAPcyv4hwHpX-MkUEqxwdTj7wCCZCN4RV-L4jsnuwLGyL_UEG4A@mail.gmail.com
Fixes: b15cd80 ("dax: Convert page fault handlers to XArray")
Signed-off-by: Matthew Wilcox (Oracle) <[email protected]>
Tested-by: Dan Williams <[email protected]>
Reported-by: Robert Barror <[email protected]>
Reported-by: Seema Pandit <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
Signed-off-by: Dan Williams <[email protected]>

Author: Matthew Wilcox (Oracle)

fs/dax.c

@@ -123,6 +123,15 @@ static int dax_is_empty_entry(void *entry)
 	return xa_to_value(entry) & DAX_EMPTY;
 }
 
+/*
+ * true if the entry that was found is of a smaller order than the entry
+ * we were looking for
+ */
+static bool dax_is_conflict(void *entry)
+{
+	return entry == XA_RETRY_ENTRY;
+}
+
 /*
  * DAX page cache entry locking
  */
@@ -195,11 +204,13 @@ static void dax_wake_entry(struct xa_state *xas, void *entry, bool wake_all)
  * Look up entry in page cache, wait for it to become unlocked if it
  * is a DAX entry and return it.  The caller must subsequently call
  * put_unlocked_entry() if it did not lock the entry or dax_unlock_entry()
- * if it did.
+ * if it did.  The entry returned may have a larger order than @order.
+ * If @order is larger than the order of the entry found in i_pages, this
+ * function returns a dax_is_conflict entry.
  *
  * Must be called with the i_pages lock held.
  */
-static void *get_unlocked_entry(struct xa_state *xas)
+static void *get_unlocked_entry(struct xa_state *xas, unsigned int order)
 {
 	void *entry;
 	struct wait_exceptional_entry_queue ewait;
@@ -210,6 +221,8 @@ static void *get_unlocked_entry(struct xa_state *xas)
 
 	for (;;) {
 		entry = xas_find_conflict(xas);
+		if (dax_entry_order(entry) < order)
+			return XA_RETRY_ENTRY;
 		if (!entry || WARN_ON_ONCE(!xa_is_value(entry)) ||
 				!dax_is_locked(entry))
 			return entry;
@@ -254,7 +267,7 @@ static void wait_entry_unlocked(struct xa_state *xas, void *entry)
 static void put_unlocked_entry(struct xa_state *xas, void *entry)
 {
 	/* If we were the only waiter woken, wake the next one */
-	if (entry)
+	if (entry && dax_is_conflict(entry))
 		dax_wake_entry(xas, entry, false);
 }
 
@@ -461,28 +474,25 @@ void dax_unlock_page(struct page *page, dax_entry_t cookie)
  * overlap with xarray value entries.
  */
 static void *grab_mapping_entry(struct xa_state *xas,
-		struct address_space *mapping, unsigned long size_flag)
+		struct address_space *mapping, unsigned int order)
 {
 	unsigned long index = xas->xa_index;
 	bool pmd_downgrade = false; /* splitting PMD entry into PTE entries? */
 	void *entry;
 
 retry:
 	xas_lock_irq(xas);
-	entry = get_unlocked_entry(xas);
+	entry = get_unlocked_entry(xas, order);
 
 	if (entry) {
+		if (dax_is_conflict(entry))
+			goto fallback;
 		if (!xa_is_value(entry)) {
 			xas_set_err(xas, EIO);
 			goto out_unlock;
 		}
 
-		if (size_flag & DAX_PMD) {
-			if (dax_is_pte_entry(entry)) {
-				put_unlocked_entry(xas, entry);
-				goto fallback;
-			}
-		} else { /* trying to grab a PTE entry */
+		if (order == 0) {
 			if (dax_is_pmd_entry(entry) &&
 			    (dax_is_zero_entry(entry) ||
 			     dax_is_empty_entry(entry))) {
@@ -523,7 +533,11 @@ static void *grab_mapping_entry(struct xa_state *xas,
 	if (entry) {
 		dax_lock_entry(xas, entry);
 	} else {
-		entry = dax_make_entry(pfn_to_pfn_t(0), size_flag | DAX_EMPTY);
+		unsigned long flags = DAX_EMPTY;
+
+		if (order > 0)
+			flags |= DAX_PMD;
+		entry = dax_make_entry(pfn_to_pfn_t(0), flags);
 		dax_lock_entry(xas, entry);
 		if (xas_error(xas))
 			goto out_unlock;
@@ -594,7 +608,7 @@ struct page *dax_layout_busy_page(struct address_space *mapping)
 		if (WARN_ON_ONCE(!xa_is_value(entry)))
 			continue;
 		if (unlikely(dax_is_locked(entry)))
-			entry = get_unlocked_entry(&xas);
+			entry = get_unlocked_entry(&xas, 0);
 		if (entry)
 			page = dax_busy_page(entry);
 		put_unlocked_entry(&xas, entry);
@@ -621,7 +635,7 @@ static int __dax_invalidate_entry(struct address_space *mapping,
 	void *entry;
 
 	xas_lock_irq(&xas);
-	entry = get_unlocked_entry(&xas);
+	entry = get_unlocked_entry(&xas, 0);
 	if (!entry || WARN_ON_ONCE(!xa_is_value(entry)))
 		goto out;
 	if (!trunc &&
@@ -849,7 +863,7 @@ static int dax_writeback_one(struct xa_state *xas, struct dax_device *dax_dev,
 	if (unlikely(dax_is_locked(entry))) {
 		void *old_entry = entry;
 
-		entry = get_unlocked_entry(xas);
+		entry = get_unlocked_entry(xas, 0);
 
 		/* Entry got punched out / reallocated? */
 		if (!entry || WARN_ON_ONCE(!xa_is_value(entry)))
@@ -1510,7 +1524,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	 * entry is already in the array, for instance), it will return
 	 * VM_FAULT_FALLBACK.
 	 */
-	entry = grab_mapping_entry(&xas, mapping, DAX_PMD);
+	entry = grab_mapping_entry(&xas, mapping, PMD_ORDER);
 	if (xa_is_internal(entry)) {
 		result = xa_to_internal(entry);
 		goto fallback;
@@ -1659,11 +1673,10 @@ dax_insert_pfn_mkwrite(struct vm_fault *vmf, pfn_t pfn, unsigned int order)
 	vm_fault_t ret;
 
 	xas_lock_irq(&xas);
-	entry = get_unlocked_entry(&xas);
+	entry = get_unlocked_entry(&xas, order);
 	/* Did we race with someone splitting entry or so? */
-	if (!entry ||
-	    (order == 0 && !dax_is_pte_entry(entry)) ||
-	    (order == PMD_ORDER && !dax_is_pmd_entry(entry))) {
+	if (!entry || dax_is_conflict(entry) ||
+	    (order == 0 && !dax_is_pte_entry(entry))) {
 		put_unlocked_entry(&xas, entry);
 		xas_unlock_irq(&xas);
 		trace_dax_insert_pfn_mkwrite_no_entry(mapping->host, vmf,