[PATCH] btrfs: add a way to monitor for ENOSPC events on a file system

[Josef Bacik <josef@toxicpanda.com>]

At Facebook's scale of btrfs deployment it's difficult to determine if
there's a systemic problem in our ENOSPC handling, or if it's simply
misbehaving tasks.  Part of monitoring btrfs at any scale is having data
about what is happening on a file system.  To that end, export the
number of ENOSPC events we've had per space_info via sysfs.  This
provides production users of btrfs to better monitor if they're having
problems, as sometimes userspace fails in different and interesting ways
that may be difficult to tie back to an errant (or even legitimate)
ENOSPC.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/space-info.c |  8 ++++++++
 fs/btrfs/space-info.h |  6 ++++++
 fs/btrfs/sysfs.c      | 12 ++++++++++++
 3 files changed, 26 insertions(+)

diff --git a/fs/btrfs/space-info.c b/fs/btrfs/space-info.c
index fc329aff478f..af467e888545 100644
--- a/fs/btrfs/space-info.c
+++ b/fs/btrfs/space-info.c
@@ -213,6 +213,7 @@ static int create_space_info(struct btrfs_fs_info *info, u64 flags)
 	INIT_LIST_HEAD(&space_info->ro_bgs);
 	INIT_LIST_HEAD(&space_info->tickets);
 	INIT_LIST_HEAD(&space_info->priority_tickets);
+	atomic_set(&space_info->enospc_events, 0);
 	space_info->clamp = 1;
 
 	ret = btrfs_sysfs_add_space_info_type(info, space_info);
@@ -1674,6 +1675,11 @@ int btrfs_reserve_metadata_bytes(struct btrfs_root *root,
 			ret = 0;
 	}
 	if (ret == -ENOSPC) {
+		if (flush == BTRFS_RESERVE_FLUSH_ALL ||
+		    flush == BTRFS_RESERVE_FLUSH_ALL_STEAL ||
+		    flush == BTRFS_RESERVE_FLUSH_EVICT)
+			atomic_inc(&block_rsv->space_info->enospc_events);
+
 		trace_btrfs_space_reservation(fs_info, "space_info:enospc",
 					      block_rsv->space_info->flags,
 					      orig_bytes, 1);
@@ -1707,6 +1713,8 @@ int btrfs_reserve_data_bytes(struct btrfs_fs_info *fs_info, u64 bytes,
 
 	ret = __reserve_bytes(fs_info, data_sinfo, bytes, flush);
 	if (ret == -ENOSPC) {
+		if (flush == BTRFS_RESERVE_FLUSH_DATA)
+			atomic_inc(&data_sinfo->enospc_events);
 		trace_btrfs_space_reservation(fs_info, "space_info:enospc",
 					      data_sinfo->flags, bytes, 1);
 		if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
diff --git a/fs/btrfs/space-info.h b/fs/btrfs/space-info.h
index b1a8ffb03b3e..11eff2139aae 100644
--- a/fs/btrfs/space-info.h
+++ b/fs/btrfs/space-info.h
@@ -73,6 +73,12 @@ struct btrfs_space_info {
 	 */
 	u64 tickets_id;
 
+	/*
+	 * Counter for the number of times user facing flush actions have
+	 * failed.
+	 */
+	atomic_t enospc_events;
+
 	struct rw_semaphore groups_sem;
 	/* for block groups in our same type */
 	struct list_head block_groups[BTRFS_NR_RAID_TYPES];
diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
index 4b508938e728..52c5311873d3 100644
--- a/fs/btrfs/sysfs.c
+++ b/fs/btrfs/sysfs.c
@@ -674,6 +674,15 @@ static ssize_t btrfs_space_info_show_total_bytes_pinned(struct kobject *kobj,
 	return scnprintf(buf, PAGE_SIZE, "%lld\n", val);
 }
 
+static ssize_t btrfs_space_info_show_enospc_events(struct kobject *kobj,
+						   struct kobj_attribute *a,
+						   char *buf)
+{
+	struct btrfs_space_info *sinfo = to_space_info(kobj);
+	int events = atomic_read(&sinfo->enospc_events);
+	return scnprintf(buf, PAGE_SIZE, "%d\n", events);
+}
+
 SPACE_INFO_ATTR(flags);
 SPACE_INFO_ATTR(total_bytes);
 SPACE_INFO_ATTR(bytes_used);
@@ -686,6 +695,8 @@ SPACE_INFO_ATTR(disk_used);
 SPACE_INFO_ATTR(disk_total);
 BTRFS_ATTR(space_info, total_bytes_pinned,
 	   btrfs_space_info_show_total_bytes_pinned);
+BTRFS_ATTR(space_info, enospc_events,
+	   btrfs_space_info_show_enospc_events);
 
 static struct attribute *space_info_attrs[] = {
 	BTRFS_ATTR_PTR(space_info, flags),
@@ -699,6 +710,7 @@ static struct attribute *space_info_attrs[] = {
 	BTRFS_ATTR_PTR(space_info, disk_used),
 	BTRFS_ATTR_PTR(space_info, disk_total),
 	BTRFS_ATTR_PTR(space_info, total_bytes_pinned),
+	BTRFS_ATTR_PTR(space_info, enospc_events),
 	NULL,
 };
 ATTRIBUTE_GROUPS(space_info);
-- 
2.26.3