Skip to contentSkip to conversation

Change data reservations to use the ticketing infra

PATCH Superseded Josef Bacik 1 participant 24 messages Jul 8, 2020 Btrfs file system

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover Change data reservations to use the ticketing infra

v2->v3:
- Rebased onto a recent misc-next

v1->v2:
- Adjusted a comment in may_commit_transaction.
- Fixed one of the intermediate patches to properly update ->reclaim_size.

We've had two different things in place to reserve data and metadata space, because generally speaking data is much simpler. However the data reservations still suffered from the same issues that plagued metadata reservations, you could get multiple tasks racing in to get reservations. This causes problems with cases like write/delete loops where we should be able to fill up the fs, delete everything, and go again. You would sometimes race with a flusher that's trying to unpin the free space, take it's reservations, and then it would fail.

Fix this by moving the data reservations under the metadata ticketing infrastructure. This gets rid of that problem, and simplifies our enospc code more by consolidating it into a single path. Thanks,

Josef

#1 lore raw

PATCH v3 1/23 btrfs: change nr to u64 in btrfs_start_delalloc_roots

We have btrfs_wait_ordered_roots() which takes a u64 for nr, but btrfs_start_delalloc_roots() that takes an int for nr, which makes using them in conjunction, especially for something like (u64)-1, annoying and inconsistent. Fix btrfs_start_delalloc_roots() to take a u64 for nr and adjust start_delalloc_inodes() and it's callers appropriately.

This means we've adjusted start_delalloc_inodes() to take a pointer of nr since we want to preserve the ability for start-delalloc_inodes() to return an error, so simply make it do the nr adjusting as necessary.

Part of adjusting the callers to this means changing btrfs_writeback_inodes_sb_nr() to take a u64 for items. This may be confusing because it seems unrelated, but the caller of btrfs_writeback_inodes_sb_nr() already passes in a u64, it's just the function variable that needs to be changed.

---
 fs/btrfs/ctree.h       |  2 +-
 fs/btrfs/dev-replace.c |  2 +-
 fs/btrfs/inode.c       | 27 +++++++++++----------------
 fs/btrfs/ioctl.c       |  2 +-
 fs/btrfs/space-info.c  |  2 +-
 5 files changed, 15 insertions(+), 20 deletions(-)
fs/btrfs/ctree.h +1 -1
index c96c27f66999..6f1ba19c6705 100644
@@ -2929,7 +2929,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 			       u32 min_type);
 
 int btrfs_start_delalloc_snapshot(struct btrfs_root *root);
-int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr);
+int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr);
 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
 			      unsigned int extra_bits,
 			      struct extent_state **cached_state);
fs/btrfs/dev-replace.c +1 -1
index db93909b25e0..769ac3098880 100644
@@ -630,7 +630,7 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	 * flush all outstanding I/O and inode extent mappings before the
 	 * copy operation is declared as being finished
 	 */
-	ret = btrfs_start_delalloc_roots(fs_info, -1);
+	ret = btrfs_start_delalloc_roots(fs_info, U64_MAX);
 	if (ret) {
 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 		return ret;
fs/btrfs/inode.c +11 -16
index d894d9e41aad..57a9a6f91572 100644
@@ -9384,7 +9384,8 @@ static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode
  * some fairly slow code that needs optimization. This walks the list
  * of all the inodes with pending delalloc and forces them to disk.
  */
-static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
+static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr,
+				 bool snapshot)
 {
 	struct btrfs_inode *binode;
 	struct inode *inode;
@@ -9424,9 +9425,11 @@ static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
 		list_add_tail(&work->list, &works);
 		btrfs_queue_work(root->fs_info->flush_workers,
 				 &work->work);
-		ret++;
-		if (nr != -1 && ret >= nr)
-			goto out;
+		if (*nr != U64_MAX) {
+			(*nr)--;
+			if (*nr == 0)
+				goto out;
+		}
 		cond_resched();
 		spin_lock(&root->delalloc_lock);
 	}
@@ -9451,18 +9454,15 @@ static int start_delalloc_inodes(struct btrfs_root *root, int nr, bool snapshot)
 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
-	int ret;
+	u64 nr = U64_MAX;
 
 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
 		return -EROFS;
 
-	ret = start_delalloc_inodes(root, -1, true);
-	if (ret > 0)
-		ret = 0;
-	return ret;
+	return start_delalloc_inodes(root, &nr, true);
 }
 
-int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
+int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr)
 {
 	struct btrfs_root *root;
 	struct list_head splice;
@@ -9485,15 +9485,10 @@ int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
 			       &fs_info->delalloc_roots);
 		spin_unlock(&fs_info->delalloc_root_lock);
 
-		ret = start_delalloc_inodes(root, nr, false);
+		ret = start_delalloc_inodes(root, &nr, false);
 		btrfs_put_root(root);
 		if (ret < 0)
 			goto out;
-
-		if (nr != -1) {
-			nr -= ret;
-			WARN_ON(nr < 0);
-		}
 		spin_lock(&fs_info->delalloc_root_lock);
 	}
 	spin_unlock(&fs_info->delalloc_root_lock);
fs/btrfs/ioctl.c +1 -1
index ab34179d7cbc..d4df1a7f53a8 100644
@@ -4848,7 +4848,7 @@ long btrfs_ioctl(struct file *file, unsigned int
 	case BTRFS_IOC_SYNC: {
 		int ret;
 
-		ret = btrfs_start_delalloc_roots(fs_info, -1);
+		ret = btrfs_start_delalloc_roots(fs_info, U64_MAX);
 		if (ret)
 			return ret;
 		ret = btrfs_sync_fs(inode->i_sb, 1);
fs/btrfs/space-info.c +1 -1
index c7bd3fdd7792..ef6e264746fc 100644
@@ -477,7 +477,7 @@ void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
 }
 
 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
-					 unsigned long nr_pages, int nr_items)
+					 unsigned long nr_pages, u64 nr_items)
 {
 	struct super_block *sb = fs_info->sb;
 
Signature · 1 line
2.24.1
#2 lore raw

PATCH v3 2/23 btrfs: remove orig from shrink_delalloc

In reply to Josef Bacik

We don't use this anywhere inside of shrink_delalloc, remove it.

---
 fs/btrfs/space-info.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/btrfs/space-info.c +2 -2
index ef6e264746fc..10cae5b55235 100644
@@ -517,7 +517,7 @@ static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
  * shrink metadata reservation for delalloc
  */
 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
-			    u64 orig, bool wait_ordered)
+			    bool wait_ordered)
 {
 	struct btrfs_space_info *space_info;
 	struct btrfs_trans_handle *trans;
@@ -742,7 +742,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		break;
 	case FLUSH_DELALLOC:
 	case FLUSH_DELALLOC_WAIT:
-		shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
+		shrink_delalloc(fs_info, num_bytes * 2,
 				state == FLUSH_DELALLOC_WAIT);
 		break;
 	case FLUSH_DELAYED_REFS_NR:
Signature · 1 line
2.24.1
#3 lore raw

PATCH v3 3/23 btrfs: handle U64_MAX for shrink_delalloc

In reply to Josef Bacik

Data allocations are going to want to pass in U64_MAX for flushing space, adjust shrink_delalloc to handle this properly.

---
 fs/btrfs/space-info.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
fs/btrfs/space-info.c +14 -3
index 10cae5b55235..78ad8f01785c 100644
@@ -530,8 +530,19 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
 	int loops;
 
 	/* Calc the number of the pages we need flush for space reservation */
-	items = calc_reclaim_items_nr(fs_info, to_reclaim);
-	to_reclaim = items * EXTENT_SIZE_PER_ITEM;
+	if (to_reclaim == U64_MAX) {
+		items = U64_MAX;
+	} else {
+		/*
+		 * to_reclaim is set to however much metadata we need to
+		 * reclaim, but reclaiming that much data doesn't really track
+		 * exactly, so increase the amount to reclaim by 2x in order to
+		 * make sure we're flushing enough delalloc to hopefully reclaim
+		 * some metadata reservations.
+		 */
+		items = calc_reclaim_items_nr(fs_info, to_reclaim) * 2;
+		to_reclaim = items * EXTENT_SIZE_PER_ITEM;
+	}
 
 	trans = (struct btrfs_trans_handle *)current->journal_info;
 	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
@@ -742,7 +753,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		break;
 	case FLUSH_DELALLOC:
 	case FLUSH_DELALLOC_WAIT:
-		shrink_delalloc(fs_info, num_bytes * 2,
+		shrink_delalloc(fs_info, num_bytes,
 				state == FLUSH_DELALLOC_WAIT);
 		break;
 	case FLUSH_DELAYED_REFS_NR:
Signature · 1 line
2.24.1
#4 lore raw

PATCH v3 4/23 btrfs: make shrink_delalloc take space_info as an arg

In reply to Josef Bacik

Currently shrink_delalloc just looks up the metadata space info, but this won't work if we're trying to reclaim space for data chunks. We get the right space_info we want passed into flush_space, so simply pass that along to shrink_delalloc.

---
 fs/btrfs/space-info.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
fs/btrfs/space-info.c +4 -5
index 78ad8f01785c..1bf40328b0ee 100644
@@ -516,10 +516,10 @@ static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
 /*
  * shrink metadata reservation for delalloc
  */
-static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
-			    bool wait_ordered)
+static void shrink_delalloc(struct btrfs_fs_info *fs_info,
+			    struct btrfs_space_info *space_info,
+			    u64 to_reclaim, bool wait_ordered)
 {
-	struct btrfs_space_info *space_info;
 	struct btrfs_trans_handle *trans;
 	u64 delalloc_bytes;
 	u64 dio_bytes;
@@ -545,7 +545,6 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
 	}
 
 	trans = (struct btrfs_trans_handle *)current->journal_info;
-	space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
 
 	delalloc_bytes = percpu_counter_sum_positive(
 						&fs_info->delalloc_bytes);
@@ -753,7 +752,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		break;
 	case FLUSH_DELALLOC:
 	case FLUSH_DELALLOC_WAIT:
-		shrink_delalloc(fs_info, num_bytes,
+		shrink_delalloc(fs_info, space_info, num_bytes,
 				state == FLUSH_DELALLOC_WAIT);
 		break;
 	case FLUSH_DELAYED_REFS_NR:
Signature · 1 line
2.24.1
#5 lore raw

PATCH v3 5/23 btrfs: make ALLOC_CHUNK use the space info flags

In reply to Josef Bacik

We have traditionally used flush_space() to flush metadata space, so we've been unconditionally using btrfs_metadata_alloc_profile() for our profile to allocate a chunk. However if we're going to use this for data we need to use btrfs_get_alloc_profile() on the space_info we pass in.

---
 fs/btrfs/space-info.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
fs/btrfs/space-info.c +2 -1
index 1bf40328b0ee..8450864c5b77 100644
@@ -777,7 +777,8 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 			break;
 		}
 		ret = btrfs_chunk_alloc(trans,
-				btrfs_metadata_alloc_profile(fs_info),
+				btrfs_get_alloc_profile(fs_info,
+							space_info->flags),
 				(state == ALLOC_CHUNK) ? CHUNK_ALLOC_NO_FORCE :
 					CHUNK_ALLOC_FORCE);
 		btrfs_end_transaction(trans);
Signature · 1 line
2.24.1
#6 lore raw

PATCH v3 6/23 btrfs: call btrfs_try_granting_tickets when freeing reserved bytes

In reply to Josef Bacik

We were missing a call to btrfs_try_granting_tickets in btrfs_free_reserved_bytes, so add it to handle the case where we're able to satisfy an allocation because we've freed a pending reservation.

---
 fs/btrfs/block-group.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/block-group.c +2 -0
index 3aa78952a2b7..daf88891a40a 100644
@@ -3091,6 +3091,8 @@ void btrfs_free_reserved_bytes(struct btrfs_block_group *cache,
 	if (delalloc)
 		cache->delalloc_bytes -= num_bytes;
 	spin_unlock(&cache->lock);
+
+	btrfs_try_granting_tickets(cache->fs_info, space_info);
 	spin_unlock(&space_info->lock);
 }
 
Signature · 1 line
2.24.1
#7 lore raw

PATCH v3 7/23 btrfs: call btrfs_try_granting_tickets when unpinning anything

In reply to Josef Bacik

When unpinning we were only calling btrfs_try_granting_tickets() if global_rsv->space_info == space_info, which is problematic because we use ticketing for SYSTEM chunks, and want to use it for DATA as well. Fix this by moving this call outside of that if statement.

---
 fs/btrfs/extent-tree.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
fs/btrfs/extent-tree.c +3 -4
index c0bc35f932bf..4b8c59318f6e 100644
@@ -2844,11 +2844,10 @@ static int unpin_extent_range(struct btrfs_fs_info *fs_info,
 				len -= to_add;
 			}
 			spin_unlock(&global_rsv->lock);
-			/* Add to any tickets we may have */
-			if (len)
-				btrfs_try_granting_tickets(fs_info,
-							   space_info);
 		}
+		/* Add to any tickets we may have */
+		if (!readonly && return_free_space && len)
+			btrfs_try_granting_tickets(fs_info, space_info);
 		spin_unlock(&space_info->lock);
 	}
 
Signature · 1 line
2.24.1
#8 lore raw

PATCH v3 8/23 btrfs: call btrfs_try_granting_tickets when reserving space

In reply to Josef Bacik

If we have compression on we could free up more space than we reserved, and thus be able to make a space reservation. Add the call for this scenario.

---
 fs/btrfs/block-group.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/btrfs/block-group.c +7 -0
index daf88891a40a..c7097cb4fc7d 100644
@@ -3058,6 +3058,13 @@ int btrfs_add_reserved_bytes(struct btrfs_block_group *cache,
 						      space_info, -ram_bytes);
 		if (delalloc)
 			cache->delalloc_bytes += num_bytes;
+
+		/*
+		 * Compression can use less space than we reserved, so wake
+		 * tickets if that happens.
+		 */
+		if (num_bytes < ram_bytes)
+			btrfs_try_granting_tickets(cache->fs_info, space_info);
 	}
 	spin_unlock(&cache->lock);
 	spin_unlock(&space_info->lock);
Signature · 1 line
2.24.1
#9 lore raw

PATCH v3 9/23 btrfs: use the btrfs_space_info_free_bytes_may_use helper for delalloc

In reply to Josef Bacik

We are going to use the ticket infrastructure for data, so use the btrfs_space_info_free_bytes_may_use() helper in btrfs_free_reserved_data_space_noquota() so we get the try_granting_tickets call when we free our reservation.

---
 fs/btrfs/delalloc-space.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
fs/btrfs/delalloc-space.c +1 -3
index 0e354e9e57d0..0a41bdcc14d1 100644
@@ -277,9 +277,7 @@ void btrfs_free_reserved_data_space_noquota(struct btrfs_fs_info *fs_info,
 	ASSERT(IS_ALIGNED(len, fs_info->sectorsize));
 
 	data_sinfo = fs_info->data_sinfo;
-	spin_lock(&data_sinfo->lock);
-	btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, -len);
-	spin_unlock(&data_sinfo->lock);
+	btrfs_space_info_free_bytes_may_use(fs_info, data_sinfo, len);
 }
 
 /*
Signature · 1 line
2.24.1
#10 lore raw

PATCH v3 10/23 btrfs: use btrfs_start_delalloc_roots in shrink_delalloc

In reply to Josef Bacik

The original iteration of flushing had us flushing delalloc and then checking to see if we could make our reservation, thus we were very careful about how many pages we would flush at once.

But now that everything is async and we satisfy tickets as the space becomes available we don't have to keep track of any of this, simply try and flush the number of dirty inodes we may have in order to reclaim space to make our reservation. This cleans up our delalloc flushing significantly.

The async_pages stuff is dropped because btrfs_start_delalloc_roots() handles the case that we generate async extents for us, so we no longer require this extra logic.

---
 fs/btrfs/space-info.c | 55 +------------------------------------------
 1 file changed, 1 insertion(+), 54 deletions(-)
fs/btrfs/space-info.c +1 -54
index 8450864c5b77..de8a49de5fdb 100644
@@ -476,28 +476,6 @@ void btrfs_dump_space_info(struct btrfs_fs_info *fs_info,
 	up_read(&info->groups_sem);
 }
 
-static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
-					 unsigned long nr_pages, u64 nr_items)
-{
-	struct super_block *sb = fs_info->sb;
-
-	if (down_read_trylock(&sb->s_umount)) {
-		writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
-		up_read(&sb->s_umount);
-	} else {
-		/*
-		 * We needn't worry the filesystem going from r/w to r/o though
-		 * we don't acquire ->s_umount mutex, because the filesystem
-		 * should guarantee the delalloc inodes list be empty after
-		 * the filesystem is readonly(all dirty pages are written to
-		 * the disk).
-		 */
-		btrfs_start_delalloc_roots(fs_info, nr_items);
-		if (!current->journal_info)
-			btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
-	}
-}
-
 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
 					u64 to_reclaim)
 {
@@ -523,10 +501,8 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
 	struct btrfs_trans_handle *trans;
 	u64 delalloc_bytes;
 	u64 dio_bytes;
-	u64 async_pages;
 	u64 items;
 	long time_left;
-	unsigned long nr_pages;
 	int loops;
 
 	/* Calc the number of the pages we need flush for space reservation */
@@ -567,37 +543,8 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
 
 	loops = 0;
 	while ((delalloc_bytes || dio_bytes) && loops < 3) {
-		nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
-
-		/*
-		 * Triggers inode writeback for up to nr_pages. This will invoke
-		 * ->writepages callback and trigger delalloc filling
-		 *  (btrfs_run_delalloc_range()).
-		 */
-		btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
-
-		/*
-		 * We need to wait for the compressed pages to start before
-		 * we continue.
-		 */
-		async_pages = atomic_read(&fs_info->async_delalloc_pages);
-		if (!async_pages)
-			goto skip_async;
-
-		/*
-		 * Calculate how many compressed pages we want to be written
-		 * before we continue. I.e if there are more async pages than we
-		 * require wait_event will wait until nr_pages are written.
-		 */
-		if (async_pages <= nr_pages)
-			async_pages = 0;
-		else
-			async_pages -= nr_pages;
+		btrfs_start_delalloc_roots(fs_info, items);
 
-		wait_event(fs_info->async_submit_wait,
-			   atomic_read(&fs_info->async_delalloc_pages) <=
-			   (int)async_pages);
-skip_async:
 		spin_lock(&space_info->lock);
 		if (list_empty(&space_info->tickets) &&
 		    list_empty(&space_info->priority_tickets)) {
Signature · 1 line
2.24.1
#11 lore raw

PATCH v3 11/23 btrfs: check tickets after waiting on ordered extents

In reply to Josef Bacik

Right now if the space is free'd up after the ordered extents complete (which is likely since the reservations are held until they complete), we would do extra delalloc flushing before we'd notice that we didn't have any more tickets. Fix this by moving the tickets check after our wait_ordered_extents check.

---
 fs/btrfs/space-info.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)
fs/btrfs/space-info.c +9 -8
index de8a49de5fdb..3b5064a2a972 100644
@@ -545,14 +545,6 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
 	while ((delalloc_bytes || dio_bytes) && loops < 3) {
 		btrfs_start_delalloc_roots(fs_info, items);
 
-		spin_lock(&space_info->lock);
-		if (list_empty(&space_info->tickets) &&
-		    list_empty(&space_info->priority_tickets)) {
-			spin_unlock(&space_info->lock);
-			break;
-		}
-		spin_unlock(&space_info->lock);
-
 		loops++;
 		if (wait_ordered && !trans) {
 			btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
@@ -561,6 +553,15 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
 			if (time_left)
 				break;
 		}
+
+		spin_lock(&space_info->lock);
+		if (list_empty(&space_info->tickets) &&
+		    list_empty(&space_info->priority_tickets)) {
+			spin_unlock(&space_info->lock);
+			break;
+		}
+		spin_unlock(&space_info->lock);
+
 		delalloc_bytes = percpu_counter_sum_positive(
 						&fs_info->delalloc_bytes);
 		dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes);
Signature · 1 line
2.24.1
#12 lore raw

PATCH v3 12/23 btrfs: add flushing states for handling data reservations

In reply to Josef Bacik

Currently the way we do data reservations is by seeing if we have enough space in our space_info. If we do not and we're a normal inode we'll

1) Attempt to force a chunk allocation until we can't anymore.
2) If that fails we'll flush delalloc, then commit the transaction, then run the delayed iputs.

If we are a free space inode we're only allowed to force a chunk allocation. In order to use the normal flushing mechanism we need to encode this into a flush state array for normal inodes. Since both will start with allocating chunks until the space info is full there is no need to add this as a flush state, this will be handled specially.

---
 fs/btrfs/ctree.h      | 2 ++
 fs/btrfs/space-info.c | 6 ++++++
 2 files changed, 8 insertions(+)
fs/btrfs/ctree.h +2 -0
index 6f1ba19c6705..4eabad429440 100644
@@ -2572,6 +2572,8 @@ enum btrfs_reserve_flush_enum {
 	 */
 	BTRFS_RESERVE_FLUSH_LIMIT,
 	BTRFS_RESERVE_FLUSH_EVICT,
+	BTRFS_RESERVE_FLUSH_DATA,
+	BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE,
 	BTRFS_RESERVE_FLUSH_ALL,
 	BTRFS_RESERVE_FLUSH_ALL_STEAL,
 };
fs/btrfs/space-info.c +6 -0
index 3b5064a2a972..94da7b43e152 100644
@@ -1018,6 +1018,12 @@ static const enum btrfs_flush_state evict_flush_states[] = {
 	COMMIT_TRANS,
 };
 
+static const enum btrfs_flush_state data_flush_states[] = {
+	FLUSH_DELALLOC_WAIT,
+	COMMIT_TRANS,
+	RUN_DELAYED_IPUTS,
+};
+
 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 				struct btrfs_space_info *space_info,
 				struct reserve_ticket *ticket,
Signature · 1 line
2.24.1
#13 lore raw

PATCH v3 13/23 btrfs: add the data transaction commit logic into may_commit_transaction

In reply to Josef Bacik

Data space flushing currently unconditionally commits the transaction twice in a row, and the last time it checks if there's enough pinned extents to satisfy it's reservation before deciding to commit the transaction for the 3rd and final time.

Encode this logic into may_commit_transaction(). In the next patch we will pass in U64_MAX for bytes_needed the first two times, and the final time we will pass in the actual bytes we need so the normal logic will apply.

This patch exists soley to make the logical changes I will make to the flushing state machine separate to make it easier to bisect any performance related regressions.

---
 fs/btrfs/space-info.c | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)
fs/btrfs/space-info.c +22 -6
index 94da7b43e152..be0f117d4ccf 100644
@@ -579,21 +579,33 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
  * will return -ENOSPC.
  */
 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
-				  struct btrfs_space_info *space_info)
+				  struct btrfs_space_info *space_info,
+				  u64 bytes_needed)
 {
 	struct reserve_ticket *ticket = NULL;
 	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
 	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
 	struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv;
 	struct btrfs_trans_handle *trans;
-	u64 bytes_needed;
 	u64 reclaim_bytes = 0;
 	u64 cur_free_bytes = 0;
+	bool do_commit = false;
 
 	trans = (struct btrfs_trans_handle *)current->journal_info;
 	if (trans)
 		return -EAGAIN;
 
+	/*
+	 * If we are data and have passed in U64_MAX we just want to
+	 * unconditionally commit the transaction to match the previous data
+	 * flushing behavior.
+	 */
+	if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
+	   bytes_needed == U64_MAX) {
+		do_commit = true;
+		goto check_pinned;
+	}
+
 	spin_lock(&space_info->lock);
 	cur_free_bytes = btrfs_space_info_used(space_info, true);
 	if (cur_free_bytes < space_info->total_bytes)
@@ -607,7 +619,7 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	else if (!list_empty(&space_info->tickets))
 		ticket = list_first_entry(&space_info->tickets,
 					  struct reserve_ticket, list);
-	bytes_needed = (ticket) ? ticket->bytes : 0;
+	bytes_needed = (ticket) ? ticket->bytes : bytes_needed;
 
 	if (bytes_needed > cur_free_bytes)
 		bytes_needed -= cur_free_bytes;
@@ -618,6 +630,7 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	if (!bytes_needed)
 		return 0;
 
+check_pinned:
 	trans = btrfs_join_transaction(fs_info->extent_root);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
@@ -627,7 +640,8 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	 * we have block groups that are going to be freed, allowing us to
 	 * possibly do a chunk allocation the next loop through.
 	 */
-	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
+	if (do_commit ||
+	    test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
 	    __percpu_counter_compare(&space_info->total_bytes_pinned,
 				     bytes_needed,
 				     BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
@@ -635,7 +649,9 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 
 	/*
 	 * See if there is some space in the delayed insertion reservation for
-	 * this reservation.
+	 * this reservation.  If the space_info's don't match (like for DATA or
+	 * SYSTEM) then just enospc, reclaiming this space won't recover any
+	 * space to satisfy those reservations.
 	 */
 	if (space_info != delayed_rsv->space_info)
 		goto enospc;
@@ -743,7 +759,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		btrfs_wait_on_delayed_iputs(fs_info);
 		break;
 	case COMMIT_TRANS:
-		ret = may_commit_transaction(fs_info, space_info);
+		ret = may_commit_transaction(fs_info, space_info, num_bytes);
 		break;
 	default:
 		ret = -ENOSPC;
Signature · 1 line
2.24.1
#14 lore raw

PATCH v3 14/23 btrfs: add btrfs_reserve_data_bytes and use it

In reply to Josef Bacik

Create a new function btrfs_reserve_data_bytes() in order to handle data reservations. This uses the new flush types and flush states to handle making data reservations.

This patch specifically does not change any functionality, and is purposefully not cleaned up in order to make bisection easier for the future patches. The new helper is identical to the old helper in how it handles data reservations. We first try to force a chunk allocation, and then we run through the flush states all at once and in the same order that they were done with the old helper.

Subsequent patches will clean this up and change the behavior of the flushing, and it is important to keep those changes separate so we can easily bisect down to the patch that caused the regression, rather than the patch that made us start using the new infrastructure.

---
 fs/btrfs/delalloc-space.c | 119 ++------------------------------------
 fs/btrfs/space-info.c     |  92 +++++++++++++++++++++++++++++
 fs/btrfs/space-info.h     |   2 +
 3 files changed, 98 insertions(+), 115 deletions(-)
fs/btrfs/delalloc-space.c +4 -115
index 0a41bdcc14d1..bacee09b7bfd 100644
@@ -115,126 +115,15 @@ int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
 {
 	struct btrfs_root *root = inode->root;
 	struct btrfs_fs_info *fs_info = root->fs_info;
-	struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
-	u64 used;
-	int ret = 0;
-	int need_commit = 2;
-	int have_pinned_space;
+	enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_DATA;
 
 	/* Make sure bytes are sectorsize aligned */
 	bytes = ALIGN(bytes, fs_info->sectorsize);
 
-	if (btrfs_is_free_space_inode(inode)) {
-		need_commit = 0;
-		ASSERT(current->journal_info);
-	}
-
-again:
-	/* Make sure we have enough space to handle the data first */
-	spin_lock(&data_sinfo->lock);
-	used = btrfs_space_info_used(data_sinfo, true);
-
-	if (used + bytes > data_sinfo->total_bytes) {
-		struct btrfs_trans_handle *trans;
-
-		/*
-		 * If we don't have enough free bytes in this space then we need
-		 * to alloc a new chunk.
-		 */
-		if (!data_sinfo->full) {
-			u64 alloc_target;
-
-			data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
-			spin_unlock(&data_sinfo->lock);
-
-			alloc_target = btrfs_data_alloc_profile(fs_info);
-			/*
-			 * It is ugly that we don't call nolock join
-			 * transaction for the free space inode case here.
-			 * But it is safe because we only do the data space
-			 * reservation for the free space cache in the
-			 * transaction context, the common join transaction
-			 * just increase the counter of the current transaction
-			 * handler, doesn't try to acquire the trans_lock of
-			 * the fs.
-			 */
-			trans = btrfs_join_transaction(root);
-			if (IS_ERR(trans))
-				return PTR_ERR(trans);
-
-			ret = btrfs_chunk_alloc(trans, alloc_target,
-						CHUNK_ALLOC_NO_FORCE);
-			btrfs_end_transaction(trans);
-			if (ret < 0) {
-				if (ret != -ENOSPC)
-					return ret;
-				else {
-					have_pinned_space = 1;
-					goto commit_trans;
-				}
-			}
-
-			goto again;
-		}
-
-		/*
-		 * If we don't have enough pinned space to deal with this
-		 * allocation, and no removed chunk in current transaction,
-		 * don't bother committing the transaction.
-		 */
-		have_pinned_space = __percpu_counter_compare(
-			&data_sinfo->total_bytes_pinned,
-			used + bytes - data_sinfo->total_bytes,
-			BTRFS_TOTAL_BYTES_PINNED_BATCH);
-		spin_unlock(&data_sinfo->lock);
-
-		/* Commit the current transaction and try again */
-commit_trans:
-		if (need_commit) {
-			need_commit--;
-
-			if (need_commit > 0) {
-				btrfs_start_delalloc_roots(fs_info, -1);
-				btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
-							 (u64)-1);
-			}
-
-			trans = btrfs_join_transaction(root);
-			if (IS_ERR(trans))
-				return PTR_ERR(trans);
-			if (have_pinned_space >= 0 ||
-			    test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
-				     &trans->transaction->flags) ||
-			    need_commit > 0) {
-				ret = btrfs_commit_transaction(trans);
-				if (ret)
-					return ret;
-				/*
-				 * The cleaner kthread might still be doing iput
-				 * operations. Wait for it to finish so that
-				 * more space is released.  We don't need to
-				 * explicitly run the delayed iputs here because
-				 * the commit_transaction would have woken up
-				 * the cleaner.
-				 */
-				ret = btrfs_wait_on_delayed_iputs(fs_info);
-				if (ret)
-					return ret;
-				goto again;
-			} else {
-				btrfs_end_transaction(trans);
-			}
-		}
-
-		trace_btrfs_space_reservation(fs_info,
-					      "space_info:enospc",
-					      data_sinfo->flags, bytes, 1);
-		return -ENOSPC;
-	}
-	btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes);
-	spin_unlock(&data_sinfo->lock);
+	if (btrfs_is_free_space_inode(inode))
+		flush = BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE;
 
-	return 0;
+	return btrfs_reserve_data_bytes(fs_info, bytes, flush);
 }
 
 int btrfs_check_data_free_space(struct btrfs_inode *inode,
fs/btrfs/space-info.c +92 -0

95 lines not on this page · raw message

fs/btrfs/space-info.h +2 -0
index c3c64019950a..5646393b928c 100644
@@ -149,5 +149,7 @@ static inline void btrfs_space_info_free_bytes_may_use(
 	btrfs_try_granting_tickets(fs_info, space_info);
 	spin_unlock(&space_info->lock);
 }
+int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
+			     enum btrfs_reserve_flush_enum flush);
 
 #endif /* BTRFS_SPACE_INFO_H */
Signature · 1 line
2.24.1
#15 lore raw

PATCH v3 15/23 btrfs: use ticketing for data space reservations

In reply to Josef Bacik

Now that we have all the infrastructure in place, use the ticketing infrastructure to make data allocations. This still maintains the exact same flushing behavior, but now we're using tickets to get our reservations satisfied.

---
 fs/btrfs/space-info.c | 126 +++++++++++++++++++++++-------------------
 1 file changed, 68 insertions(+), 58 deletions(-)
fs/btrfs/space-info.c +68 -58
index 3a61cdfe4b21..a107370d144c 100644
@@ -1070,6 +1070,54 @@ static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
 	} while (flush_state < states_nr);
 }
 
+static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
+					struct btrfs_space_info *space_info,
+					struct reserve_ticket *ticket,
+					const enum btrfs_flush_state *states,
+					int states_nr)
+{
+	int flush_state = 0;
+	int commit_cycles = 2;
+
+	while (!space_info->full) {
+		flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE);
+		spin_lock(&space_info->lock);
+		if (ticket->bytes == 0) {
+			spin_unlock(&space_info->lock);
+			return;
+		}
+		spin_unlock(&space_info->lock);
+	}
+again:
+	while (flush_state < states_nr) {
+		u64 flush_bytes = U64_MAX;
+
+		if (!commit_cycles) {
+			if (states[flush_state] == FLUSH_DELALLOC_WAIT) {
+				flush_state++;
+				continue;
+			}
+			if (states[flush_state] == COMMIT_TRANS)
+				flush_bytes = ticket->bytes;
+		}
+
+		flush_space(fs_info, space_info, flush_bytes,
+			    states[flush_state]);
+		spin_lock(&space_info->lock);
+		if (ticket->bytes == 0) {
+			spin_unlock(&space_info->lock);
+			return;
+		}
+		spin_unlock(&space_info->lock);
+		flush_state++;
+	}
+	if (commit_cycles) {
+		commit_cycles--;
+		flush_state = 0;
+		goto again;
+	}
+}
+
 static void wait_reserve_ticket(struct btrfs_fs_info *fs_info,
 				struct btrfs_space_info *space_info,
 				struct reserve_ticket *ticket)
@@ -1136,6 +1184,15 @@ static int handle_reserve_ticket(struct btrfs_fs_info *fs_info,
 						evict_flush_states,
 						ARRAY_SIZE(evict_flush_states));
 		break;
+	case BTRFS_RESERVE_FLUSH_DATA:
+		priority_reclaim_data_space(fs_info, space_info, ticket,
+					data_flush_states,
+					ARRAY_SIZE(data_flush_states));
+		break;
+	case BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE:
+		priority_reclaim_data_space(fs_info, space_info, ticket,
+					    NULL, 0);
+		break;
 	default:
 		ASSERT(0);
 		break;
@@ -1343,78 +1400,31 @@ int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
 			     enum btrfs_reserve_flush_enum flush)
 {
 	struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
-	const enum btrfs_flush_state *states = NULL;
 	u64 used;
-	int states_nr = 0;
-	int commit_cycles = 2;
 	int ret = -ENOSPC;
 
 	ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA);
 
-	if (flush == BTRFS_RESERVE_FLUSH_DATA) {
-		states = data_flush_states;
-		states_nr = ARRAY_SIZE(data_flush_states);
-	}
-
 	spin_lock(&data_sinfo->lock);
-again:
 	used = btrfs_space_info_used(data_sinfo, true);
 
 	if (used + bytes > data_sinfo->total_bytes) {
-		u64 prev_total_bytes = data_sinfo->total_bytes;
-		int flush_state = 0;
+		struct reserve_ticket ticket;
 
+		init_waitqueue_head(&ticket.wait);
+		ticket.bytes = bytes;
+		ticket.error = 0;
+		list_add_tail(&ticket.list, &data_sinfo->priority_tickets);
+		data_sinfo->reclaim_size += bytes;
 		spin_unlock(&data_sinfo->lock);
 
-		/*
-		 * Everybody can force chunk allocation, so try this first to
-		 * see if we can just bail here and make our reservation.
-		 */
-		flush_space(fs_info, data_sinfo, bytes, ALLOC_CHUNK_FORCE);
-		spin_lock(&data_sinfo->lock);
-		if (prev_total_bytes < data_sinfo->total_bytes)
-			goto again;
+		ret = handle_reserve_ticket(fs_info, data_sinfo, &ticket,
+					    flush);
+	} else {
+		btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes);
+		ret = 0;
 		spin_unlock(&data_sinfo->lock);
-
-		/*
-		 * Cycle through the rest of the flushing options for our flush
-		 * type, then try again.
-		 */
-		while (flush_state < states_nr) {
-			u64 flush_bytes = U64_MAX;
-
-			/*
-			 * Previously we unconditionally committed the
-			 * transaction twice before finally checking against
-			 * pinned space before committing the final time.  We
-			 * also skipped flushing delalloc the final pass
-			 * through.
-			 */
-			if (!commit_cycles) {
-				if (states[flush_state] == FLUSH_DELALLOC_WAIT) {
-					flush_state++;
-					continue;
-				}
-				if (states[flush_state] == COMMIT_TRANS)
-					flush_bytes = bytes;
-			}
-
-			flush_space(fs_info, data_sinfo, flush_bytes,
-				    states[flush_state]);
-			flush_state++;
-		}
-
-		if (!commit_cycles)
-			goto out;
-
-		commit_cycles--;
-		spin_lock(&data_sinfo->lock);
-		goto again;
 	}
-	btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes);
-	ret = 0;
-	spin_unlock(&data_sinfo->lock);
-out:
 	if (ret)
 		trace_btrfs_space_reservation(fs_info,
 					      "space_info:enospc",
Signature · 1 line
2.24.1
#16 lore raw

PATCH v3 16/23 btrfs: serialize data reservations if we are flushing

In reply to Josef Bacik

Nikolay reported a problem where generic/371 would fail sometimes with a slow drive. The gist of the test is that we fallocate a file in parallel with a pwrite of a different file. These two files combined are smaller than the file system, but sometimes the pwrite would ENOSPC.

A fair bit of investigation uncovered the fact that the fallocate workload was racing in and grabbing the free space that the pwrite workload was trying to free up so it could make its own reservation. After a few loops of this eventually the pwrite workload would error out with an ENOSPC.

We've had the same problem with metadata as well, and we serialized all metadata allocations to satisfy this problem. This wasn't usually a problem with data because data reservations are more straightforward, but obviously could still happen.

Fix this by not allowing reservations to occur if there are any pending tickets waiting to be satisfied on the space info.

---
 fs/btrfs/space-info.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/btrfs/space-info.c +5 -1
index a107370d144c..afdc774cebe9 100644
@@ -1402,13 +1402,17 @@ int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
 	struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
 	u64 used;
 	int ret = -ENOSPC;
+	bool pending_tickets;
 
 	ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA);
 
 	spin_lock(&data_sinfo->lock);
 	used = btrfs_space_info_used(data_sinfo, true);
+	pending_tickets = !list_empty(&data_sinfo->tickets) ||
+		!list_empty(&data_sinfo->priority_tickets);
 
-	if (used + bytes > data_sinfo->total_bytes) {
+	if (pending_tickets ||
+	    used + bytes > data_sinfo->total_bytes) {
 		struct reserve_ticket ticket;
 
 		init_waitqueue_head(&ticket.wait);
Signature · 1 line
2.24.1
#17 lore raw

PATCH v3 17/23 btrfs: use the same helper for data and metadata reservations

In reply to Josef Bacik

Now that data reservations follow the same pattern as metadata reservations we can simply rename __reserve_metadata_bytes to __reserve_bytes and use that helper for data reservations.

Things to keep in mind, btrfs_can_overcommit() returns 0 for data, because we can never overcommit. We also will never pass in FLUSH_ALL for data, so we'll simply be added to the priority list and go straight into handle_reserve_ticket.

---
 fs/btrfs/space-info.c | 49 +++++++++++++------------------------------
 1 file changed, 14 insertions(+), 35 deletions(-)
fs/btrfs/space-info.c +14 -35
index afdc774cebe9..605be9f4fcac 100644
@@ -1252,10 +1252,9 @@ static inline bool is_normal_flushing(enum btrfs_reserve_flush_enum flush)
  * regain reservations will be made and this will fail if there is not enough
  * space already.
  */
-static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
-				    struct btrfs_space_info *space_info,
-				    u64 orig_bytes,
-				    enum btrfs_reserve_flush_enum flush)
+static int __reserve_bytes(struct btrfs_fs_info *fs_info,
+			   struct btrfs_space_info *space_info, u64 orig_bytes,
+			   enum btrfs_reserve_flush_enum flush)
 {
 	struct reserve_ticket ticket;
 	u64 used;
@@ -1367,8 +1366,8 @@ int btrfs_reserve_metadata_bytes(struct btrfs_root *root,
 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
 	int ret;
 
-	ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
-				       orig_bytes, flush);
+	ret = __reserve_bytes(fs_info, block_rsv->space_info, orig_bytes,
+			      flush);
 	if (ret == -ENOSPC &&
 	    unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
 		if (block_rsv != global_rsv &&
@@ -1400,38 +1399,18 @@ int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
 			     enum btrfs_reserve_flush_enum flush)
 {
 	struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
-	u64 used;
-	int ret = -ENOSPC;
-	bool pending_tickets;
+	int ret;
 
+	ASSERT(flush == BTRFS_RESERVE_FLUSH_DATA ||
+	       flush == BTRFS_RESERVE_FLUSH_FREE_SPACE_INODE);
 	ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_DATA);
 
-	spin_lock(&data_sinfo->lock);
-	used = btrfs_space_info_used(data_sinfo, true);
-	pending_tickets = !list_empty(&data_sinfo->tickets) ||
-		!list_empty(&data_sinfo->priority_tickets);
-
-	if (pending_tickets ||
-	    used + bytes > data_sinfo->total_bytes) {
-		struct reserve_ticket ticket;
-
-		init_waitqueue_head(&ticket.wait);
-		ticket.bytes = bytes;
-		ticket.error = 0;
-		list_add_tail(&ticket.list, &data_sinfo->priority_tickets);
-		data_sinfo->reclaim_size += bytes;
-		spin_unlock(&data_sinfo->lock);
-
-		ret = handle_reserve_ticket(fs_info, data_sinfo, &ticket,
-					    flush);
-	} else {
-		btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes);
-		ret = 0;
-		spin_unlock(&data_sinfo->lock);
-	}
-	if (ret)
-		trace_btrfs_space_reservation(fs_info,
-					      "space_info:enospc",
+	ret = __reserve_bytes(fs_info, data_sinfo, bytes, flush);
+	if (ret == -ENOSPC) {
+		trace_btrfs_space_reservation(fs_info, "space_info:enospc",
 					      data_sinfo->flags, bytes, 1);
+		if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
+			btrfs_dump_space_info(fs_info, data_sinfo, bytes, 0);
+	}
 	return ret;
 }
Signature · 1 line
2.24.1
#18 lore raw

PATCH v3 18/23 btrfs: drop the commit_cycles stuff for data reservations

In reply to Josef Bacik

This was an old wart left over from how we previously did data reservations. Before we could have people race in and take a reservation while we were flushing space, so we needed to make sure we looped a few times before giving up. Now that we're using the ticketing infrastructure we don't have to worry about this and can drop the logic altogether.

---
 fs/btrfs/space-info.c | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
fs/btrfs/space-info.c +2 -20
index 605be9f4fcac..5efb3eb9253d 100644
@@ -1077,7 +1077,6 @@ static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
 					int states_nr)
 {
 	int flush_state = 0;
-	int commit_cycles = 2;
 
 	while (!space_info->full) {
 		flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE);
@@ -1088,21 +1087,9 @@ static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
 		}
 		spin_unlock(&space_info->lock);
 	}
-again:
-	while (flush_state < states_nr) {
-		u64 flush_bytes = U64_MAX;
-
-		if (!commit_cycles) {
-			if (states[flush_state] == FLUSH_DELALLOC_WAIT) {
-				flush_state++;
-				continue;
-			}
-			if (states[flush_state] == COMMIT_TRANS)
-				flush_bytes = ticket->bytes;
-		}
 
-		flush_space(fs_info, space_info, flush_bytes,
-			    states[flush_state]);
+	while (flush_state < states_nr) {
+		flush_space(fs_info, space_info, U64_MAX, states[flush_state]);
 		spin_lock(&space_info->lock);
 		if (ticket->bytes == 0) {
 			spin_unlock(&space_info->lock);
@@ -1111,11 +1098,6 @@ static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info,
 		spin_unlock(&space_info->lock);
 		flush_state++;
 	}
-	if (commit_cycles) {
-		commit_cycles--;
-		flush_state = 0;
-		goto again;
-	}
 }
 
 static void wait_reserve_ticket(struct btrfs_fs_info *fs_info,
Signature · 1 line
2.24.1
#19 lore raw

PATCH v3 19/23 btrfs: don't force commit if we are data

In reply to Josef Bacik

We used to unconditionally commit the transaction at least 2 times and then on the 3rd try check against pinned space to make sure committing the transaction was worth the effort. This is overkill, we know nobody is going to steal our reservation, and if we can't make our reservation with the pinned amount simply bail out.

This also cleans up the passing of bytes_needed to may_commit_transaction, as that was the thing we added into place in order to accomplish this behavior. We no longer need it so remove that mess.

---
 fs/btrfs/space-info.c | 24 +++++-------------------
 1 file changed, 5 insertions(+), 19 deletions(-)
fs/btrfs/space-info.c +5 -19
index 5efb3eb9253d..5b46835766e3 100644
@@ -579,8 +579,7 @@ static void shrink_delalloc(struct btrfs_fs_info *fs_info,
  * will return -ENOSPC.
  */
 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
-				  struct btrfs_space_info *space_info,
-				  u64 bytes_needed)
+				  struct btrfs_space_info *space_info)
 {
 	struct reserve_ticket *ticket = NULL;
 	struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
@@ -588,24 +587,13 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	struct btrfs_block_rsv *trans_rsv = &fs_info->trans_block_rsv;
 	struct btrfs_trans_handle *trans;
 	u64 reclaim_bytes = 0;
+	u64 bytes_needed;
 	u64 cur_free_bytes = 0;
-	bool do_commit = false;
 
 	trans = (struct btrfs_trans_handle *)current->journal_info;
 	if (trans)
 		return -EAGAIN;
 
-	/*
-	 * If we are data and have passed in U64_MAX we just want to
-	 * unconditionally commit the transaction to match the previous data
-	 * flushing behavior.
-	 */
-	if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
-	   bytes_needed == U64_MAX) {
-		do_commit = true;
-		goto check_pinned;
-	}
-
 	spin_lock(&space_info->lock);
 	cur_free_bytes = btrfs_space_info_used(space_info, true);
 	if (cur_free_bytes < space_info->total_bytes)
@@ -619,7 +607,7 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	else if (!list_empty(&space_info->tickets))
 		ticket = list_first_entry(&space_info->tickets,
 					  struct reserve_ticket, list);
-	bytes_needed = (ticket) ? ticket->bytes : bytes_needed;
+	bytes_needed = (ticket) ? ticket->bytes : 0;
 
 	if (bytes_needed > cur_free_bytes)
 		bytes_needed -= cur_free_bytes;
@@ -630,7 +618,6 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	if (!bytes_needed)
 		return 0;
 
-check_pinned:
 	trans = btrfs_join_transaction(fs_info->extent_root);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
@@ -640,8 +627,7 @@ static int may_commit_transaction(struct btrfs_fs_info *fs_info,
 	 * we have block groups that are going to be freed, allowing us to
 	 * possibly do a chunk allocation the next loop through.
 	 */
-	if (do_commit ||
-	    test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
+	if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
 	    __percpu_counter_compare(&space_info->total_bytes_pinned,
 				     bytes_needed,
 				     BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
@@ -759,7 +745,7 @@ static void flush_space(struct btrfs_fs_info *fs_info,
 		btrfs_wait_on_delayed_iputs(fs_info);
 		break;
 	case COMMIT_TRANS:
-		ret = may_commit_transaction(fs_info, space_info, num_bytes);
+		ret = may_commit_transaction(fs_info, space_info);
 		break;
 	default:
 		ret = -ENOSPC;
Signature · 1 line
2.24.1
#20 lore raw

PATCH v3 20/23 btrfs: run delayed iputs before committing the transaction for data

In reply to Josef Bacik

Before we were waiting on iputs after we committed the transaction, but this doesn't really make much sense. We want to reclaim any space we may have in order to be more likely to commit the transaction, due to pinned space being added by running the delayed iputs. Fix this by making delayed iputs run before committing the transaction.

---
 fs/btrfs/space-info.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/btrfs/space-info.c +1 -1
index 5b46835766e3..092f3f62a5ef 100644
@@ -1022,8 +1022,8 @@ static const enum btrfs_flush_state evict_flush_states[] = {
 
 static const enum btrfs_flush_state data_flush_states[] = {
 	FLUSH_DELALLOC_WAIT,
-	COMMIT_TRANS,
 	RUN_DELAYED_IPUTS,
+	COMMIT_TRANS,
 };
 
 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
Signature · 1 line
2.24.1
#21 lore raw

PATCH v3 21/23 btrfs: flush delayed refs when trying to reserve data space

In reply to Josef Bacik

We can end up with free'd extents in the delayed refs, and thus may_commit_transaction() may not think we have enough pinned space to commit the transaction and we'll ENOSPC early. Handle this by running the delayed refs in order to make sure pinned is uptodate before we try to commit the transaction.

---
 fs/btrfs/space-info.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/space-info.c +1 -0
index 092f3f62a5ef..fce61f800150 100644
@@ -1023,6 +1023,7 @@ static const enum btrfs_flush_state evict_flush_states[] = {
 static const enum btrfs_flush_state data_flush_states[] = {
 	FLUSH_DELALLOC_WAIT,
 	RUN_DELAYED_IPUTS,
+	FLUSH_DELAYED_REFS,
 	COMMIT_TRANS,
 };
 
Signature · 1 line
2.24.1
#22 lore raw

PATCH v3 22/23 btrfs: do async reclaim for data reservations

In reply to Josef Bacik

Now that we have the data ticketing stuff in place, move normal data reservations to use an async reclaim helper to satisfy tickets. Before we could have multiple tasks race in and both allocate chunks, resulting in more data chunks than we would necessarily need. Serializing these allocations and making a single thread responsible for flushing will only allocate chunks as needed, as well as cut down on transaction commits and other flush related activities.

Priority reservations will still work as they have before, simply trying to allocate a chunk until they can make their reservation.

---
 fs/btrfs/ctree.h      |   3 +-
 fs/btrfs/disk-io.c    |   3 +-
 fs/btrfs/space-info.c | 123 ++++++++++++++++++++++++++++++------------
 fs/btrfs/super.c      |   1 +
 4 files changed, 93 insertions(+), 37 deletions(-)
fs/btrfs/ctree.h +2 -1
index 4eabad429440..aa6532976404 100644
@@ -494,7 +494,7 @@ enum btrfs_orphan_cleanup_state {
 	ORPHAN_CLEANUP_DONE	= 2,
 };
 
-void btrfs_init_async_reclaim_work(struct work_struct *work);
+void btrfs_init_async_reclaim_work(struct btrfs_fs_info *fs_info);
 
 /* fs_info */
 struct reloc_control;
@@ -917,6 +917,7 @@ struct btrfs_fs_info {
 
 	/* Used to reclaim the metadata space in the background. */
 	struct work_struct async_reclaim_work;
+	struct work_struct async_data_reclaim_work;
 
 	spinlock_t unused_bgs_lock;
 	struct list_head unused_bgs;
fs/btrfs/disk-io.c +2 -1
index aa97a39e4b15..08140caceb33 100644
@@ -2779,7 +2779,7 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	fs_info->check_integrity_print_mask = 0;
 #endif
 	btrfs_init_balance(fs_info);
-	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
+	btrfs_init_async_reclaim_work(fs_info);
 
 	spin_lock_init(&fs_info->block_group_cache_lock);
 	fs_info->block_group_cache_tree = RB_ROOT;
@@ -4099,6 +4099,7 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	btrfs_cleanup_defrag_inodes(fs_info);
 
 	cancel_work_sync(&fs_info->async_reclaim_work);
+	cancel_work_sync(&fs_info->async_data_reclaim_work);
 
 	/* Cancel or finish ongoing discard work */
 	btrfs_discard_cleanup(fs_info);
fs/btrfs/space-info.c +88 -35

186 lines not on this page · raw message

fs/btrfs/super.c +1 -0
index f90ee35465ab..b44e859c3d5c 100644
@@ -1868,6 +1868,7 @@ static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 		 * the filesystem is busy.
 		 */
 		cancel_work_sync(&fs_info->async_reclaim_work);
+		cancel_work_sync(&fs_info->async_data_reclaim_work);
 
 		btrfs_discard_cleanup(fs_info);
 
Signature · 1 line
2.24.1
#23 lore raw

PATCH v3 23/23 btrfs: add a comment explaining the data flush steps

In reply to Josef Bacik

The data flushing steps are not obvious to people other than myself and Chris. Write a giant comment explaining the reasoning behind each flush step for data as well as why it is in that particular order.

---
 fs/btrfs/space-info.c | 47 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 47 insertions(+)
fs/btrfs/space-info.c +47 -0
index 680218a7b0e5..ba2bb47cbb18 100644
@@ -998,6 +998,53 @@ static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
 	} while (flush_state <= COMMIT_TRANS);
 }
 
+/*
+ * FLUSH_DELALLOC_WAIT:
+ *   Space is free'd from flushing delalloc in one of two ways.
+ *
+ *   1) compression is on and we allocate less space than we reserved.
+ *   2) We are overwriting existing space.
+ *
+ *   For #1 that extra space is reclaimed as soon as the delalloc pages are
+ *   cow'ed, by way of btrfs_add_reserved_bytes() which adds the actual extent
+ *   length to ->bytes_reserved, and subtracts the reserved space from
+ *   ->bytes_may_use.
+ *
+ *   For #2 this is trickier.  Once the ordered extent runs we will drop the
+ *   extent in the range we are overwriting, which creates a delayed ref for
+ *   that freed extent.  This however is not reclaimed until the transaction
+ *   commits, thus the next stages.
+ *
+ * RUN_DELAYED_IPUTS
+ *   If we are freeing inodes, we want to make sure all delayed iputs have
+ *   completed, because they could have been on an inode with i_nlink == 0, and
+ *   thus have been trunated and free'd up space.  But again this space is not
+ *   immediately re-usable, it comes in the form of a delayed ref, which must be
+ *   run and then the transaction must be committed.
+ *
+ * FLUSH_DELAYED_REFS
+ *   The above two cases generate delayed refs that will affect
+ *   ->total_bytes_pinned.  However this counter can be inconsistent with
+ *   reality if there are outstanding delayed refs.  This is because we adjust
+ *   the counter based soley on the current set of delayed refs and disregard
+ *   any on-disk state which might include more refs.  So for example, if we
+ *   have an extent with 2 references, but we only drop 1, we'll see that there
+ *   is a negative delayed ref count for the extent and assume that the space
+ *   will be free'd, and thus increase ->total_bytes_pinned.
+ *
+ *   Running the delayed refs gives us the actual real view of what will be
+ *   freed at the transaction commit time.  This stage will not actually free
+ *   space for us, it just makes sure that may_commit_transaction() has all of
+ *   the information it needs to make the right decision.
+ *
+ * COMMIT_TRANS
+ *   This is where we reclaim all of the pinned space generated by the previous
+ *   two stages.  We will not commit the transaction if we don't think we're
+ *   likely to satisfy our request, which means if our current free space +
+ *   total_bytes_pinned < reservation we will not commit.  This is why the
+ *   previous states are actually important, to make sure we know for sure
+ *   whether committing the transaction will allow us to make progress.
+ */
 static const enum btrfs_flush_state data_flush_states[] = {
 	FLUSH_DELALLOC_WAIT,
 	RUN_DELAYED_IPUTS,
Signature · 1 line
2.24.1