Skip to contentSkip to conversation

btrfs: Turn fs_info member buffer_radix into XArray

PATCH In review Gabriel Niebler 3 participants 5 messages Apr 21, 2022 – May 2, 2022 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 btrfs: Turn fs_info member buffer_radix into XArray

… named 'extent_buffers'. Also adjust all usages of this object to use the XArray API, which greatly simplifies the code as it takes care of locking and is generally easier to use and understand, providing notionally simpler array semantics.

Also perform some light refactoring.

---

Changes from v1:
- Fixed first line of commit message

---
 fs/btrfs/ctree.h             |   4 +-
 fs/btrfs/disk-io.c           |   4 +-
 fs/btrfs/extent_io.c         | 118 ++++++++++++++---------------------
 fs/btrfs/tests/btrfs-tests.c |  22 +------
 4 files changed, 55 insertions(+), 93 deletions(-)
fs/btrfs/ctree.h +2 -2
index b7631b88426e..833a95be041b 100644
@@ -994,10 +994,10 @@ struct btrfs_fs_info {
 
 	struct btrfs_delayed_root *delayed_root;
 
-	/* Extent buffer radix tree */
+	/* Extent buffer xarray */
 	spinlock_t buffer_lock;
 	/* Entries are eb->start / sectorsize */
-	struct radix_tree_root buffer_radix;
+	struct xarray extent_buffers;
 
 	/* next backup root to be overwritten */
 	int backup_root_index;
fs/btrfs/disk-io.c +2 -2
index 126f244cdf88..4ab3eba6578d 100644
@@ -486,7 +486,7 @@ static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
 		uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
 						       fs_info->nodesize);
 
-		/* A dirty eb shouldn't disappear from buffer_radix */
+		/* A dirty eb shouldn't disappear from extent_buffers */
 		if (WARN_ON(!eb))
 			return -EUCLEAN;
 
@@ -3133,7 +3133,7 @@ static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 {
 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
-	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
+	xa_init_flags(&fs_info->extent_buffers, GFP_ATOMIC);
 	INIT_LIST_HEAD(&fs_info->trans_list);
 	INIT_LIST_HEAD(&fs_info->dead_roots);
 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
fs/btrfs/extent_io.c +48 -70

174 lines not on this page · raw message

fs/btrfs/tests/btrfs-tests.c +3 -19
index d8e56edd6991..c8c4efc9a3fb 100644
@@ -150,8 +150,8 @@ struct btrfs_fs_info *btrfs_alloc_dummy_fs_info(u32 nodesize, u32 sectorsize)
 
 void btrfs_free_dummy_fs_info(struct btrfs_fs_info *fs_info)
 {
-	struct radix_tree_iter iter;
-	void **slot;
+	unsigned long index;
+	struct extent_buffer *eb;
 	struct btrfs_device *dev, *tmp;
 
 	if (!fs_info)
@@ -163,25 +163,9 @@ void btrfs_free_dummy_fs_info(struct btrfs_fs_info *fs_info)
 
 	test_mnt->mnt_sb->s_fs_info = NULL;
 
-	spin_lock(&fs_info->buffer_lock);
-	radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter, 0) {
-		struct extent_buffer *eb;
-
-		eb = radix_tree_deref_slot_protected(slot, &fs_info->buffer_lock);
-		if (!eb)
-			continue;
-		/* Shouldn't happen but that kind of thinking creates CVE's */
-		if (radix_tree_exception(eb)) {
-			if (radix_tree_deref_retry(eb))
-				slot = radix_tree_iter_retry(&iter);
-			continue;
-		}
-		slot = radix_tree_iter_resume(slot, &iter);
-		spin_unlock(&fs_info->buffer_lock);
+	xa_for_each(&fs_info->extent_buffers, index, eb) {
 		free_extent_buffer_stale(eb);
-		spin_lock(&fs_info->buffer_lock);
 	}
-	spin_unlock(&fs_info->buffer_lock);
 
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 	list_for_each_entry_safe(dev, tmp, &fs_info->fs_devices->devices,
Signature · 1 line
2.35.3
#1 lore raw
Quoted 8 lines · Gabriel Niebler

… named 'extent_buffers'. Also adjust all usages of this object to use the XArray API, which greatly simplifies the code as it takes care of locking and is generally easier to use and understand, providing notionally simpler array semantics.

Also perform some light refactoring.

I've added the patch to for-next, seems that it survived one night test round, so it seems to be fine. I'll move it to misc-next next week, Nikolay was interested in reviewing the xarray patches.

Reviewed-by on v2 1/1
#2 lore raw
In reply to Gabriel Niebler
Quoted 5 lines · Gabriel Niebler

… named 'extent_buffers'. Also adjust all usages of this object to use the XArray API, which greatly simplifies the code as it takes care of locking and is generally easier to use and understand, providing notionally simpler array semantics.

Quoted 8 lines

Also perform some light refactoring.

---

Changes from v1:
- Fixed first line of commit message

One minor suggestion below though.

<snip>

Quoted 50 lines
(quoted diff) +9 -21
@@ -6313,10 +6306,8 @@ static int release_extent_buffer(struct extent_buffer *eb)
 
  			spin_unlock(&eb->refs_lock);
 
-			spin_lock(&fs_info->buffer_lock);
-			radix_tree_delete(&fs_info->buffer_radix,
-					  eb->start >> fs_info->sectorsize_bits);
-			spin_unlock(&fs_info->buffer_lock);
+			xa_erase(&fs_info->extent_buffers,
+				 eb->start >> fs_info->sectorsize_bits);
  		} else {
  			spin_unlock(&eb->refs_lock);
  		}
@@ -7249,41 +7240,28 @@ void memmove_extent_buffer(const struct extent_buffer *dst,
  	}
  }
 
-#define GANG_LOOKUP_SIZE	16
  static struct extent_buffer *get_next_extent_buffer(
  		struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
  {
-	struct extent_buffer *gang[GANG_LOOKUP_SIZE];
+	struct extent_buffer *eb;
  	struct extent_buffer *found = NULL;
+	unsigned long index;
  	u64 page_start = page_offset(page);
-	u64 cur = page_start;
 
  	ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
  	lockdep_assert_held(&fs_info->buffer_lock);
 
-	while (cur < page_start + PAGE_SIZE) {
-		int ret;
-		int i;
-
-		ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
-				(void **)gang, cur >> fs_info->sectorsize_bits,
-				min_t(unsigned int, GANG_LOOKUP_SIZE,
-				      PAGE_SIZE / fs_info->nodesize));
-		if (ret == 0)
-			goto out;
-		for (i = 0; i < ret; i++) {
-			/* Already beyond page end */
-			if (gang[i]->start >= page_start + PAGE_SIZE)
-				goto out;
+	xa_for_each_start(&fs_info->extent_buffers, index, eb,
+			  page_start >> fs_info->sectorsize_bits) {
+		if (eb->start >= page_start + PAGE_SIZE)
+		        /* Already beyond page end */
+			break;

Review void memmove_extent_buffer(const struct extent_buffer *dst, · line 7263

Quoted 10 lines
(quoted diff) +3 -5
+		if (eb->start >= bytenr) {
  			/* Found one */
-			if (gang[i]->start >= bytenr) {
-				found = gang[i];
-				goto out;
-			}
+			found = eb;
+			break; >   		}
-		cur = gang[ret - 1]->start + gang[ret - 1]->len;
  	}

nit: The body of the loop can be turned into:

fs/btrfs/extent_io.c +5 -8
index da3d9dc186cd..7c1d5fec59dd 100644
@@ -7318,16 +7318,13 @@ static struct extent_buffer *get_next_extent_buffer(
 
        xa_for_each_start(&fs_info->extent_buffers, index, eb,
                          page_start >> fs_info->sectorsize_bits) {
-               if (eb->start >= page_start + PAGE_SIZE)
+               if (in_range(eb->start, page_start, PAGE_SIZE))
+                       return eb;
+               else if (eb->start >= page_start + PAGE_SIZE)
                        /* Already beyond page end */
-                       break;
-               if (eb->start >= bytenr) {
-                       /* Found one */
-                       found = eb;
-                       break;
-               }
+                       return NULL;
        }
-       return found;
+       return NULL;
 }
 
 

That is use the in_range macro to detect when we have an eb between page_start and page_start + PAGE_SIZE in which case we can directly return it, and the in_range is self-documenting. And directly return NULL in case of eb->start going beyond the current page and in case we didn't find anything. David, what do you think?

Quoted 4 lines
(quoted diff) +0 -1
-out:
  	return found;
  }
 

<snip>

#3 lore raw
Quoted 31 lines · Nikolay Borisov

Review

Quoted 2 lines
(quoted diff) +0 -1
-		cur = gang[ret - 1]->start + gang[ret - 1]->len;
  	}

nit: The body of the loop can be turned into:

fs/btrfs/extent_io.c +5 -8
index da3d9dc186cd..7c1d5fec59dd 100644
@@ -7318,16 +7318,13 @@ static struct extent_buffer *get_next_extent_buffer(
 
         xa_for_each_start(&fs_info->extent_buffers, index, eb,
                           page_start >> fs_info->sectorsize_bits) {
-               if (eb->start >= page_start + PAGE_SIZE)
+               if (in_range(eb->start, page_start, PAGE_SIZE))
+                       return eb;
+               else if (eb->start >= page_start + PAGE_SIZE)
                         /* Already beyond page end */
-                       break;
-               if (eb->start >= bytenr) {
-                       /* Found one */
-                       found = eb;
-                       break;
-               }
+                       return NULL;
         }
-       return found;
+       return NULL;
  }
 
 
Quoted 5 lines

That is use the in_range macro to detect when we have an eb between page_start and page_start + PAGE_SIZE in which case we can directly return it, and the in_range is self-documenting. And directly return NULL in case of eb->start going beyond the current page and in case we didn't find anything. David, what do you think?

I like it and folded to the patch, thanks. The variable 'found' becomes unused so removed as well.

#4 lore raw

Am 29.04.22 um 20:39 schrieb David Sterba:

Quoted 32 lines · Nikolay Borisov
Quoted 31 lines

Review

Quoted 2 lines
(quoted diff) +0 -1
-		cur = gang[ret - 1]->start + gang[ret - 1]->len;
   	}

nit: The body of the loop can be turned into:

fs/btrfs/extent_io.c +5 -8
index da3d9dc186cd..7c1d5fec59dd 100644
@@ -7318,16 +7318,13 @@ static struct extent_buffer *get_next_extent_buffer(
 
          xa_for_each_start(&fs_info->extent_buffers, index, eb,
                            page_start >> fs_info->sectorsize_bits) {
-               if (eb->start >= page_start + PAGE_SIZE)
+               if (in_range(eb->start, page_start, PAGE_SIZE))
+                       return eb;
+               else if (eb->start >= page_start + PAGE_SIZE)
                          /* Already beyond page end */
-                       break;
-               if (eb->start >= bytenr) {
-                       /* Found one */
-                       found = eb;
-                       break;
-               }
+                       return NULL;
          }
-       return found;
+       return NULL;
   }
 
 
Quoted 8 lines
Quoted 5 lines

That is use the in_range macro to detect when we have an eb between page_start and page_start + PAGE_SIZE in which case we can directly return it, and the in_range is self-documenting. And directly return NULL in case of eb->start going beyond the current page and in case we didn't find anything. David, what do you think?

I like it and folded to the patch, thanks. The variable 'found' becomes unused so removed as well.

Nice! I didn't know there was such a macro - learned something again.

Thank you both!